How do machines learn? Evaluating the AIcon2abs method

How do machines learn? Evaluating the AIcon2abs method | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article How do machines learn? Evaluating the AIcon2abs method Rubens Lacerda Queiroz, Cabral Lima, Fábio Ferrentini Sampaio, and 1 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-5728177/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract This study is an expansion of a previous work aiming to evaluate the AIcon2abs method ( AI from Concrete to Abstract: Demystifying Artificial Intelligence to the general public ), an innovative method aimed at increasing the public (including children) understanding of machine learning (ML). The approach employs the WiSARD algorithm, a weightless neural network known for its simplicity, and user accessibility. WiSARD does not require Internet, making it ideal for non-technical users and resource-limited environments. This method enables participants to intuitively visualize and interact with ML processes through engaging, hands-on activities, as if they were the algorithms themselves. The method allows users to intuitively visualize and understand the internal processes of training and classification through practical activities. Once WiSARD's functionality does not require an Internet connection, it can learn effectively from a minimal dataset, even from a single example. This feature enables users to observe how the machine improves its accuracy incrementally as it receives more data. Moreover, WiSARD generates mental images representing what it has learned, highlighting essential features of the classified data. AIcon2abs was tested through a six-hour remote course with 34 Brazilian participants, including 5 children (8–11 yo), 5 adolescents (12–17 yo), and 24 adults (21–72 yo). Data analysis was conducted from two perspectives: a mixed-method pre-experiment (including hypothesis testing), and a qualitative phenomenological analysis. Nearly 100% of participants rated AIcon2abs positively, with the results demonstrating a high degree of satisfaction in achieving the intended outcomes. This research was approved by the CEP/HUCFF/UFRJ Research Ethics Committee. Demystifying AI with AIcon2abs Machine -learning Education WiSARD algorithm Accessible AI tools Hands-on learning Easy machine learning visualization Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Figure 7 Figure 8 Figure 9 Figure 10 Figure 11 Figure 12 Figure 13 1. Introduction As noted by Queiroz et al. (2021), Artificial Intelligence (AI) has been increasingly integrated into a wide variety of fields, influencing numerous facets of daily life at an ever-expanding rate. In response to this rapid growth, several scholars have highlighted the critical need to develop a strategic framework for the advancement of AI, stressing that such a strategy must emerge from a reflective process that involves all sectors of society (Almeida, 2018; Caputo, 2019; Medina, 2004; Yogeshwar, 2018). Nonetheless, a pivotal question arises: How can individuals effectively participate in discussions and decision-making regarding the future of AI without possessing a foundational understanding of the field? This dilemma underscores an urgent necessity to develop accessible methods that enable the general public to become informed and engaged participants in debates and decisions concerning the adoption and implementation of AI technologies. A major challenge in achieving this goal lies in the inherent complexity of the techniques and concepts underpinning most AI systems (Sakulkueakulsuk et al., 2018). In response, researchers have begun to explore solutions aimed at overcoming these obstacles and fostering a more accessible understanding of AI. Russell and Norvig (2020) classify the AI domain into three principal areas: Search, Knowledge Representation, and Machine Learning. Among these, Machine Learning has emerged as one of the most significant fields, largely responsible for the recent "AI boom." Most AI systems that individuals encounter in their daily lives are built on Machine Learning principles, as in the case of Generative AI 1 as chat GPT 2 and other LLMs 3 , which combines machine learning mechanisms with other AI approaches to produce generative AI 4 . Consequently, most initiatives focused on demystifying AI for the public have concentrated on the Machine Learning component, as evidenced by the Multivocal Literature Review (MLR) 5 conducted by Queiroz et al. (2021). This review encompassed scientific databases and general internet searches, leading to the identification of various artifacts, approaches, characteristics, and unresolved issues. From these findings, the AIcon2abs method was developed as a potential solution (Queiroz et al., 2021). The AIcon2abs method was designed to facilitate an introductory understanding of AI for the general public, using Machine Learning as an accessible entry point. In its most basic form, Machine Learning involves providing an algorithm with examples of a specific class, which enables the system to later classify new observations as belonging or not to that class . For the purposes of this study, the foundational understanding of Machine Learning is framed by two core concepts: (i) Machine Learning is essentially a process by which computers learn from examples , and (ii) this learning occurs through the execution of a computer program. The AIcon2abs method comprises several key components, including a block-based programming environment, a series of interactive educational activities based on the WiSARD weightless artificial neural network model, a set of educational robotics materials, and a standard instructional unit (IU). The block-based programming environment, termed BlockWiSARD (Queiroz et al., 2021), was specifically designed to support individuals with no prior programming experience in (i) creating programs that enable their computers to learn how to recognize images and (ii) understanding the underlying operations of these programs. Notably, BlockWiSARD can be operated on low-cost computers without the need for a GPU or internet connection. In BlockWiSARD, higher-level tasks necessary for machine learning, as well as the instantiation of the learning algorithm, are incorporated as commands within the program being developed. This feature is intended to facilitate a clearer distinction between traditional computer programs and those capable of learning from examples. Specifically, BlockWiSARD includes four programming blocks directly related to the training and classification process. By combining these blocks with conventional programming commands, users can easily construct systems that (i) instantiate a learning mechanism, (ii) input data for learning, (iii) assign corresponding labels to that data, and (iv) classify new input data. The concept of incorporating machine learning primitives into programming languages was originally proposed by Tom Mitchell (2006). In addition to the machine learning blocks, BlockWiSARD features a block responsible for generating a mental image of a given class. These mental images enable users to visualize a prototype of a class learned from training patterns, thereby enhancing their understanding of how machine learning models conceptualize and categorize data. An essential principle of BlockWiSARD is that each command block within the program must correspond to a tangible action in the concrete world, such as displaying an image on a screen, hearing a sound played a specific number of times, or seeing a light blink. This premise must also be valid for the machine learning related programming blocks. But how can we visualize or observe the internal processes carried out by the machine when the learning algorithm is instantiated? What happens when the Learn and Recognize commands are executed in the program? Additionally, how can we enable individuals to replicate these actions as if they were the algorithm themselves? To this end, the AIcon2abs method incorporates games, animations, and interactive activities centered around the WiSARD model, designed to engage users in a hands-on, experiential learning process (Queiroz et al., 2021; Queiroz, 2023). As Papert (1994, p.31) said, "The reason you are not a mathematician might well be that you think that math has nothing to do with the body; you have kept your body out of it because it is supposed to be abstract, or perhaps a teacher scolded you for using your fingers to add numbers! This idea […] has inspired me to use the computer as a medium to allow children to put their bodies back into their mathematics." WiSARD is a perfect tool to put people's bodies into learning AI through ludic activities. These activities also enable learners to observe the internal operations of the machine when the learning algorithm is instantiated within BlockWiSARD. They also allow the visualization of what the machine does when the Learn and Recognize blocks are used in the BlockWiSARD programs. Moreover, the ludic activities make it possible for people to perform all those actions as if they were the WiSARD algorithm itself. BlockWiSARD operates in three modes: (i) Computer Only; (ii) Computer + Arduino 6 , or (iii) Raspberry Pi 7 . The latter two modes enable the integration of BlockWiSARD with low-cost educational robotics platforms. In addition to the core materials, a set of robotics tools compatible with BlockWiSARD was developed as part of the AIcon2abs method (Queiroz et al., 2021). While not essential to the method, these materials complement the learning process and demonstrate how robotics can enhance the educational experience when combined with BlockWiSARD. For the effective implementation of the AIcon2abs method, the inclusion of a standard instructional unit (IU) is essential. This IU includes an asynchronous e-learning implementation with a workload of approximately 6 hours which was used to evaluate the method, as discussed in Section 3 of this paper. This evaluation is the focus of this paper. The theoretical foundations supporting the design of the AIcon2abs method are rooted in several core concepts and technologies, including: (i) Piaget's (1950) Constructivist Theory, (ii) Papert's (1993) Constructionist Theory, (iii) Russell and Norvig's (2020) concept of Intelligent Agents, (iv) the Machine Learning process proposed by Mitchell (1997), (v) Turing's (1950) Imitation Game, (vi) Feynman's (1989) notion of learning through construction, (vii) Block-based programming (Resnick et al., 2009), (viii) the WiSARD weightless artificial network (Aleksander et al., 2009), (ix) Active Learning methodologies (Bonwell & Eison, 1991), and (x) educational robotics (Eguchi, 2010; Papert, 1994). These concepts and their interconnections with the AIcon2abs method are explored in further detail by Queiroz et al. (2021) and Queiroz (2023). 1.1 Research question Queiroz et al. (2021) provide a comprehensive theoretical framework for the conceptualization of the AIcon2abs method and translate these theoretical ideas into functional artifacts with significant potential to address the problem the method aims to solve. However, in that first study, the authors did not conduct an empirical evaluation of the method . In light of this, the present study seeks to empirically evaluate the AIcon2abs method in order to answer the following research question: Is AIcon2abs a suitable method to support a diverse audience of different ages and backgrounds in understanding the concept of machine learning in its most basic form in an uncomplicated and satisfactory way? 1.2. Objectives 1.2.1. General objective The primary aim of this study is to evaluate the AIcon2abs method in order to assess its potential strengths and weaknesses as a tool for helping an audience of eight years old and over to understand satisfactorily, in an uncomplicated way, the concept of machine learning in its most basic form. The concept of satisfactory solution , as defined by Simon (1996), refers to a solution that adequately addresses a given problem. In this context, a satisfactory solution may be achieved in two ways: (i) through consensus among the stakeholders involved in the problem or (ii) by advancing the current solution relative to previous solutions provided by earlier artifacts (Dresch et al., 2015). 1.2.2. Specific objectives The specific objectives of this research are as follows: (i) to evaluate the artifacts developed as part of the AIcon2abs method, ensuring they meet the specified requirements; (ii) to assess the validity of the theoretical conjectures formulated in Subsection 1.4; and (iii) to formalize the construction and contingency heuristics of the AIcon2abs method (see Subsections 4.3.4 and 4.3.5), which constitute the primary research contributions outlined by the Design Science Research (DSR) methodology proposed by Dresch et al. (2015) (Section 2). 1.3. AIcon2abs method requirements The AIcon2abs method was developed with the intention of fulfilling the following requirements (Queiroz et al., 2021): It must be suitable for individuals both with and without fully developed abstraction capabilities. It must incorporate a machine learning algorithm that facilitates the easy visualization and manipulation of its underlying logic through interactive visual elements. It must provide a block-based programming environment that enables non-expert users (including children) to create basic programs capable of learning. The program environment must include the chosen learning algorithm's instantiation, training, and classification commands among its primitives , and it must be compatible with low-cost robotic platforms. The method should adopt Active Learning methodologies, including narratives, challenges, interactive activities, and games. It must be designed to operate effectively on low-cost computers which do not require a Graphics Processing Unit (GPU) or an internet connection. 1.4. AIcon2abs theoretical conjectures Theoretical conjectures represent assumptions regarding human or social behavior, grounded in theories, models, and constructs from various disciplines. Within the framework of Design Science Research (DSR), these conjectures guide the design of the artifact, and the artifact is then employed to test the validity of these conjectures (Pimentel et al., 2019). The core theoretical conjecture that informed the design and development of the AIcon2abs method is as follows: Learning abstract concepts from computer science can be facilitated through practices that highlight observable aspects of these concepts in the concrete world. Additional theoretical conjectures that contributed to the design of the AIcon2abs method include: Algorithms with explicit visual representations enhance understanding for non-expert audiences, including children. A machine learning algorithm capable of online learning (interleaving training and classification) and generalizing from only one or a few trained examples facilitates the understanding of the machine learning process described by Mitchell (1997). General users, including children, can grasp the distinction between conventional computer programs and those capable of learning by: (i) integrating the instantiation of the machine learning algorithm, along with training and classification primitives, into basic programming constructs; (ii) adopting an agent-based approach; and (iii) utilizing the machine learning process framework articulated by Mitchell (1997). The essence of the Imitation Game (Turing, 1950) provides a valuable tool for comprehending AI, particularly by comparing the performance of a human and a machine in tasks requiring human-like intelligence. 1.5. Expected research subjects’ performance By adhering to the theoretical conjectures and fulfilling the outlined requirements, the AIcon2abs method is expected to enable its target audience to achieve the following outcomes: Develop the ability to create programs utilizing basic programming structures such as loops and conditionals. Replicate, through ludic activities, the specific steps performed by the machine learning algorithm used in the programs they develop . Understand the fundamental process of teaching a machine to learn from examples using machine learning . 2. Research methodology and text structure To address our research question, we employed the Design Science Research (DSR) method outlined by Dresch et al. (2015), incorporating adaptations from the DSR map elements developed by Pimentel et al. (2019). Notably, the DSR map by Pimentel et al. (2019) does not constitute a research method per se; rather, it serves as a framework to support DSR-oriented investigations, irrespective of the specific DSR method adopted. By integrating these approaches, we established a process comprising 14 steps organized into six clusters (see Figure 1). This article is structured in alignment with the DSR steps depicted in the diagram presented in Fig. 1, with a particular focus on the evaluation of the AIcon2abs method and the communication of findings, covering steps 10 through 14. Section 1 provides the context for this study, presenting an overview of the AIcon2abs method’s conception, structure, and the objectives of this research. Section 2 delineates the research methodology, followed by Section 3, which details the design, development, and implementation of the AIcon2abs standardized user interface (UI). Sections 4 and 5 then concentrate on the evaluation of the AIcon2abs method. A comprehensive discussion of steps 1 through 9 can be found in Queiroz et al. (2021) and Queiroz (2023), the latter being Queiroz's Doctoral Thesis. Dresch et al. (2015) suggest that DSR-based research may integrate data collection and analysis techniques from other methodological frameworks, such as case studies and experimental designs. In this study, we adopted a pre-experimental design alongside a phenomenological approach to data collection and analysis. A pre-experiment involves limited control over the variables involved in the process, employing a single group (the experimental group) where the intervention is implemented without a control group (Cohen et al., 2002). For the phenomenological component, we applied the approach suggested by Creswell (2016), which frames phenomenology as a qualitative method to explore the shared meanings individuals ascribe to their lived experiences regarding a specific phenomenon (Van Manen, 1990). 3. AIcon2abs instructional unit Figure 2 illustrates a comprehensive diagram of the AIcon2abs method's components. The standardized AIcon2abs interface (IU) was implemented as the primary tool in the empirical study to rigorously assess and validate the method. For additional information on the BlockWiSARD block-based programming environment, Ludic Activities related to the WiSARD Weightless Artificial Neural Network (WANN), and AIcon2abs Educational Robotics Materials, refer to Queiroz et al. (2021). 3.1. AIcon2abs IU components The development of the AIcon2abs IU adhered to the ADDIE instructional design model (Analyze, Design, Develop, Implement, and Evaluate) as advocated by Branch (2009) and Schlegel (2012). Among various instructional design frameworks, ADDIE is recognized for its efficacy in addressing complex educational scenarios, making it well-suited for the creation of comprehensive learning resources (Branch, 2009). To facilitate this empirical evaluation, forty-one video lessons and fourteen interactive ludic activities were developed in Portuguese for the AIcon2abs IU. These learning resources were organized into seven instructional units as follows: (1) Introduction; (2) Introducing the Story of Zrow; (3) Transforming Your Computer into a Zrow Prototype; (4) Exploring BlockWiSARD – Zrow's Programming Environment; (5) Teaching Zrow about Star and Flower Shapes; (6) WiSARD – The Learning Engine of Zrow; and (7) Zrow's Departure. All AIcon2abs reference materials (in Portuguese) are available in Queiroz (2023). The Portuguese video lessons and ludic activities can also be accessed via the following YouTube 8 playlists: Invitation to the AIcon2abs Course (https://www.youtube.com/playlist?list=PLGOt2oAva9YDbXosELwFw_tv3CHL4kkOg) AI for the General Public: Introduction (https://www.youtube.com/playlist?list=PLGOt2oAva9YCHZaaYQeVq2D8TiasBnwc-) AI for the General Public: Introducing the Story of Zrow (https://www.youtube.com/playlist?list=PLGOt2oAva9YCjFb3Of5NrZdN2PJV1SWn6) AIcon2abs Course Videos (https://www.youtube.com/playlist?list=PLGOt2oAva9YB-nlAZgqpeuHp53uQLTE7x) Examples of interactive activities in English are also available at: The Digitized Game (https://scratch.mit.edu/projects/453213702) You Are a WiSARD! (https://scratch.mit.edu/projects/440606183) You Are a WiSARD with Bleaching! (https://scratch.mit.edu/projects/440606183) The instructional design included the formation of three Google Classroom groups segmented by age: children (8–12 years), adolescents (12–17 years), and adults (18+ years), in compliance with the Brazilian National Research Ethics Committee (Comissão Nacional de Ética em Pesquisa – CONEP 9 ) guidelines. All groups accessed the same learning content and environment. Prior to deployment, navigability testing was conducted: all video lessons were viewed, and all activities were completed to identify and rectify any potential issues. To promote the course, materials were created in three formats—text, image, and video—and disseminated via YouTube, Facebook 10 , WhatsApp 11 , and Instagram 12 . Video-based student guides were also developed, providing instructions on course progression and usage of the virtual learning platform. Pre-test and post-test questionnaires were prepared and diostributed through Google Forms. 3.2. AIcon2abs IU implementation Following the development and testing of the AIcon2abs standardized IU, the instructional materials created for this method were delivered through an online course titled "AIcon2abs" ( AI from Concrete to Abstract, Demystifying Artificial Intelligence to the General Public ). Details of this process are provided in Subsections 3.2.1 and 3.2.2. To access the AIcon2abs course for the method evaluation, users can follow the steps below using a Google account: Use the invitation link: https://classroom.google.com/c/NjE3NDgyNjEwMzE3?cjc=6bndekk or Access Google Classroom . Click on the (+) icon in the upper corner of the page. Choose the option Join Class . Enter the class code 6bndekk and click Join . Finally, click on the classwork tab. 3.1.1. Initiation, conduction, and completion of instruction Participants had the flexibility to engage with the course activities at their convenience, allowing for asynchronous participation and independent task completion. This design permitted participants to join the course at different times and progress at their own pace. Any questions or feedback from participants were managed through messages within the platform. The AIcon2abs course was completed by thirty-four participants aged 8 to 72 . This cohort included twenty-four adults (aged 21–72), five adolescents (aged 12–17), and five children (aged 8–11). Of the participants, 53% had previous programming experience, while 23% had prior knowledge of AI or machine learning. Participants’ educational backgrounds were diverse, with 32% in Social, Biological, and Human Sciences, 38% in the exact sciences and engineering, and 29% in basic education. Geographically, 47% resided in Southeast Brazil, and 53% in the Southern region of the country. 3.1.2. Data collection instruments The primary instrument for data collection was a set of pre-test and post-test questionnaires comprising both open-ended and closed questions. Closed questions included nominal (yes/no and multiple-choice) and ordinal (Likert scale) formats. The pre-test questionnaire contained 11 items, while the post-test questionnaire consisted of 15 items. These questionnaires are available in Queiroz (2023). Secondary data, including public posts, private messages within the learning platform, and email communications, also contributed to the data collection process. 4. Results As outlined in Section 2, data from the pre-test and post-test questionnaires, along with secondary data sources, were analyzed from two perspectives: (i) as part of a mixed-methods pre-experiment ( including exploratory data analysis and hypothesis testing ) and (ii) from a phenomenological perspective (a qualitative examination capturing the essence of participants' experiences with the AIcon2abs course). This section presents findings related to participants' comprehension of the machine learning process and their evaluation of the usability and clarity of the AIcon2abs method components (see Table 1). The questions used in this analysis, available in Queiroz (2023), focused on participants' basic understanding of machine learning ( in its most basic form ) before and after the course, as well as their insights regarding the course materials and their overall experience in the AIcon2abs course. Table 1 List of questions used for the data analysis presented in this article Kind ID Question Data type Pre-post Q03 If you believe these devices can learn new things, tell me how you think they are taught Categorized Pre-post Q04 If you believe that the computer can recognize an image it has never "seen" but is similar to other photos it has "seen," tell me: how do you think it can do that? Categorized Pre-post Q07 What does it take to "transform" a conventional computer program into a program capable of learning? Categorized Pre-post Q10 What is machine learning? Categorized Post Q14 Has your understanding of machine learning changed after you took the course? Categorical Post Q15 What has changed? Open Post Q18 Did you understand the programs made with BlockWiSARD? Ordinal Post Q20 Did you understand how WiSARD works by doing the Scratch 13 [interactive ludic] activities? Ordinal Post Q23 Did you like to take the AIcon2abs course? Were you satisfied with the things you learned during the process? Ordinal Post Q25 What did you learn by taking this course? Open Post Q26 Did the course bring you any reflections? Which? Open For this analysis, we employed two types of categorical data: (i) inherently categorical data and (ii) derived categorical data . Inherently categorical data were obtained from responses to questions with predefined options, such as Yes/No and multiple-choice answers. Derived categorical data, on the other hand, were generated by categorizing responses to open-ended questions. To avoid influencing participants' responses, some questions were intentionally left open-ended without predefined options. Categorization was applied when responses could be grouped by thematic similarity, enabling both visualization and quantification of the data. Consequently, wherever feasible, answers to open-ended questions were systematically categorized. Table 2 presents examples of categorized responses to the question: " If you believe these devices can learn new things, explain how you think they are taught ." A comprehensive categorization of all open-ended responses is available in AIcon2abs (2023). The categorization and scoring process for these responses adhered to the methodologies proposed by Smith (2000) and Woike (2007). Categorical and categorized data visualization and analysis are presented for the full participant dataset, with certain visualizations organized by age group where relevant. Since the analyzed data are exclusively categorical or ordinal, calculations and visualization tools appropriate solely for interval and ratio data in exploratory analysis were not employed in this study. Table 2 Examples of categorization of open-ended answers Original answer Categorization They are taught through the presentation of some examples of a particular class so that, later, reading, and interpreting patterns. With examples I'm not sure, but I think these devices learn from training. For example, a robot can be "trained" to recognize facial expressions . Trained by humans They are taught with programs and images . Computer programs With data provided by different users over time . Collecting data Machine Learning . Machine learning Do not learn . Do not learn In the case of pre- and post-test questions, a modified application and interpretation of the confusion matrix was utilized as a data visualization tool, which we termed the Pre-Posttest Matrix . Originating with Pearson (1904), the confusion matrix, or contingency table, is defined as follows: "A confusion matrix for an N-way classification task is an N by N matrix where the cell (x, y) contains the number of times an item with correct classification x was classified by the model as y" (Jurafsky & Martin, 2008, p.35). Correspondingly, a Pre-posttest Matrix for an N -possible-answers pretest-posttest question Q is an N by N matrix where the cell ( x , y ) contains the number of times a question Q answer changed its value from x in the pre-test to y in the post-test . The Pre-Posttest Matrix is designed to illustrate more explicitly the shifts in participants’ responses to a given question between the pre-test and post-test . It is also noteworthy that, prior to the implementation of the AIcon2abs course, a beta test 14 was conducted to evaluate the ludic activities centered around the WiSARD model. This test was administered to undergraduate students enrolled in an introductory computer programming course. After engaging in these activities, the students successfully implemented fully functional WiSARDs based solely on the knowledge they acquired from the exercises. All participants rated the activities involving the WiSARD model as accessible and engaging. Informed consent was obtained from all students prior to their participation in the beta test. 4.1. Data analysis from a pre-experiment perspective This section provides an exploratory analysis and hypothesis testing using a mixed-methods approach, based on data collected and analyzed within a pre-experimental design framework. For the purpose of this analysis, the adult participant group was subdivided into two distinct age categories: (i) Adults, comprising individuals aged 21 to 59, and (ii) Seniors, comprising individuals aged 60 to 72. This age-based division aligns with the United Nations guidelines 15 . . 4.1.1. Pre-posttest questions exploratory analysis Q03-Pre-posttest: If you believe these devices can learn new things, tell me how you think they are taught. (Categorized) The bar chart in Figure 3a illustrates that, during the pre-test phase (labeled as Q1), none of the participants expressed the concept of using examples to teach a computer. Additionally, only three participants identified training as a method for teaching a computer. However, in the post-test phase (labeled as Q2), the number of participants who verbalized these concepts increased to 16, indicating an 80% rise. Notably, of the six participants who either did not know or did not respond to the question during the pre-test, only one retained this non-response in the post-test. The stacked bar charts in Figure 3b and Figure 3c further highlight a perceptible improvement in the quality of responses to this question across all age groups from pre-test to post-test. The Pre-Post Test Matrix presented in Figure 4 enables the observation of specific response changes among research participants. For instance, four participants who initially responded with "Collecting data" in the pre-test modified their answer to "With examples" in the post-test. Additionally, the matrix reveals the distribution of responses provided in the post-test by the 11 participants who had initially answered "Other" in the pre-test, as well as by the 6 participants who had either not responded or indicated a lack of knowledge regarding this question prior to participating in the course. Q04-Pre-posttest: If you believe that the computer can recognize an image it has never "seen" but is similar to other photos it has "seen," tell me: how do you think it can do that? (Categorized) The bar chart in Figure 5a indicates that approximately 44% of participants (35.3% + 8.8%) who did not verbalize concepts related to training, classification, and generalization for classifying new observations in the pre-test articulated these concepts in the post-test. Furthermore, the stacked bar charts in Figure 5b and Figure 5c demonstrate that this shift in responses occurred consistently across all age groups. Q07-Pre-posttest : What is required to "transform" a conventional computer program into a program capable of learning? (Categorized) A comparison of the pre-test and post-test responses to Q07 (as shown in Figure 6) reveals that numerous participants completed the course by providing answers aligned with the material presented in the AIcon2abs course. Ideas such as "Include a learning mechanism" and "Train with examples," which were absent in the pre-test responses, were verbalized by many participants in the post-test. Q10-Pre-posttest: What is machine learning? (Categorized) The analysis of the bar chart related to responses to question Q10 (Figure 7) reveals that approximately 32% of participants (26.5% + 5.9%) who did not initially articulate the concepts of "Teaching the computer with examples" or "Humans training computers" as definitions of machine learning in the pre-test expressed these ideas in the post-test. Additionally, all participants who were unsure or did not respond to the pre-test question (35.3%) provided answers in the post-test. Notably, eleven participants (approximately 32% of the total) began the course by offering responses already consistent with the course material. The Pre-Posttest Matrix in Figure 8 further indicates that among those participants who did not answer or stated they did not know the answer in the pre-test, three changed their responses to "Teaching the computer with examples," one to "Pattern recognition," two to "Humans training computers," two to "A computer program," while two did not incorporate any of the machine learning concepts covered in the course into their post-test responses. Additionally, of the twelve individuals who provided alternative responses in the pre-test, one adjusted their answer to "The machine generalizing what was taught to it," five to "Teaching the computer with examples," and two to "A computer program." Four participants concluded the course without demonstrating clear concepts of machine learning in their responses. Q14-Pos-test: Has your understanding of machine learning changed after you took the course? (Categorical) / Q15-Pos-test: What has changed? (Open) The bar chart displaying responses to Question 14 (Figure 9) suggests that the course contributed significantly to altering the participants' understanding of machine learning. Illustrative responses to the open-ended question “What has changed in your understanding of machine learning?” (Q15) include: (i) “I now understand how this learning process functions internally,” (ii) “I have learned that machines acquire knowledge through examples,” (iii) “I realized that machines can be continuously trained and improve their accuracy over time,” and (iv) “I previously believed there was only one approach to learning”. Q18-Pos-test: Did you understand the programs made with BlockWiSARD Been 1 = "I did not understand" and 5 = "I totally understand" how would you rate your understanding of the programs? (Ordinal) / Q20-Posttest: Did you understand how WiSARD works by doing the Scratch [interactive ludic] activities? Choose rate from 1 to 5, where 1 = "I cannot understand" and 5 = "I fully understand". (Ordinal) Bar chart in Figure 10 shows that most of the research subjects could understand the programming challenges performed with BlockWiSARD. Regarding the Scratch ludic activities about the WiSARD model, the results indicate a good understanding of most participants. As shown in Figure 11, scores 4 and 5 represent m 16 ost of the scores used to evaluate the method. Once the empirical evaluation was conducted remotely (Due to the COVID-19 pandemic), the IU utilized in the method evaluation did not include the AIcon2abs robotics materials. However, Alpha15 tests were performed with these materials, as shown in the links below A Raspberry Pi wheeled robot that can sense, learn, and act (https://youtu.be/KfEOnvSr_kU) A wheeled robot (made from simple robotics) sensing, learning, and acting through a program developed with BlockWiSARD (https://youtu.be/GK6GHzuix10) BlockWiSARD working alongside an Arduino board (https://youtu.be/xOLXF5LvziQ) 4.1.2. Hypothesis testing (Inferential analysis) Beyond exploratory data analysis, non-parametric hypothesis tests (Siegel, 1957) were conducted on the responses to pre- and post-test questions Q03, Q04, Q07, and Q10. Two non-parametric tests were utilized: the Wilcoxon signed-rank test and the Sign test (Cohen et al., 2002; Kirk, 1978; Pett, 2015; Siegel, 1957). The hypothesis testing described in this section was designed to evaluate the following hypotheses: H0 ( Null Hypothesis ): The AIcon2abs course had no effect on participants' understanding of the content covered regarding machine learning concepts and processes. H1 ( Alternative hypothesis ): The AIcon2abs course had a positive effect on participants' understanding of the content covered regarding machine learning concepts and processes. To conduct these hypothesis tests, scores were assigned to each response category defined for the pre- and post-test questions Q03, Q04, Q07, and Q10 (see introductory paragraphs of Section 4). Table 3 presents the scoring scale applied to attribute scalar values to the four categorized questions related to machine learning concepts and processes addressed in the AIcon2abs course. All hypothesis testing procedures were carried out using IBM® SPSS® Statistics 17 . Table 3 Criteria adopted to score the categorized answers Score Answer's coherence level regarding the content covered in the course 5 Answer consistent with the course content 4 Answer relatively consistent with the course content 3 Answer slightly consistent with the course content 2 Answer with little consistency in relation to the course content 1 Answer not consistent with the course content 0 Answer equal to “Do not know” or “Blank answer” Hypothesis test part 1 For the Wilcoxon signed-rank test to yield robust and reliable results, a primary requirement is that the distribution of the differences between the two dependent samples should be approximately symmetrical. This condition was met between the two dependent samples when considering the total score across all four questions for all 34 research participants. Table 4 and Table 5 display the results of the Wilcoxon signed-rank test and the Sign test for questions Q03, Q04, Q07, and Q10. As indicated, both tests produced identical results. Table 4 Wilcoxon signed-rank test results for Q03+Q04+Q07+Q10 Scores Significance Decision N (+) dif (-) dif Tais Q03 + Q04 + Q07 + Q10 <,001. Reject H0 34 28 2 4 Table 5 Sign test results for Q03+Q04+Q07+Q10 Scores Significance Decision N (+) dif (-) dif Tais Q03 + Q04 + Q07 + Q10 <,001. Reject H0 34 28 2 4 Hypothesis test part 2 The Wilcoxon signed-rank test assumption of symmetry was not met when analyzing the scores for each of the four questions individually (even without dividing the data by age groups). Consequently, only the Sign test was applied for these hypothesis tests. Table 6 presents the Sing tests results. Table 6 Sign test results for Q03, Q04, Q07, and Q10 (separately) regarding the pre-posttest scores of the 34 research subjects Scores Significance Decision N (+) dif (-) dif Tais Q03 ,002 Reject H0 34 20 4 10 Q04 <,001 Reject H0 34 20 3 11 Q07 <,001 Reject H0 34 19 2 13 Q10 ,002 Reject H0 34 22 5 7 The p-value for statistical significance was set at 0.05 when defining test parameters in IBM® SPSS® Statistics. A p-value of 0.05 is a standard threshold in hypothesis testing, indicating a 5% probability that the observed results could occur purely by chance. However, in most cases, the statistical significance calculated by the tests yielded a p-value below 0.001. This level of significance implies that the likelihood of the observed results occurring by chance is approximately 1 in 1,000, or 0.1%. The results obtained from the hypothesis tests (leading to a rejection of H0) support the positive effect of the AIcon2abs course on the research participants. In addition to statistical significance, the marked prevalence of positive post-test outcomes compared to pre-test results is also noteworthy. 4.2. Data Analysis from a phenomenological perspective This study also employed a phenomenological analysis approach. As outlined in Section 2, we adopted the phenomenological methodology proposed by Creswell (2016), which involves describing the phenomenon experienced by research subjects and the context in which this experience unfolded. Additionally, in a few comprehensive paragraphs, the researcher presents a portrayal of the essence of participants' perceptions of their lived experience. This narrative is constructed by reducing and organizing participants' testimonies into significant quotations classified by thematic categories. 4.2.1. The lived experience and its context Section 3 detailed the context in which participants experienced the observed phenomenon: an online course conducted as part of research on a method intended to provide insight into the inner workings of machine learning and AI. Participation was voluntary, and motivations varied, including (i) curiosity and interest in the topics covered (computer programming, AI, and machine learning), (ii) the desire to expand knowledge, and (iii) the opportunity to contribute to scientific research. In Subsection 4.2.2, titled The AIcon2abs Course , we present a narrative capturing the essence of participants' experiences. This narrative synthesizes perceptions derived from responses to post-test questions 23, 25, and 26 (Table 1), alongside additional comments on the learning platform. To develop the narrative, we organized participants' responses into four categories: (i) Feelings, (ii) Didactics and Content, (iii) Learning, and (iv) Reflections. The thematic categorization and original statements in Portuguese are available in AIcon2abs (2023). 4.2.2. The AIcon2abs course The following narrative serves as a composite testimonial reflecting the common feelings, perceptions, and reflections of the 34 participants regarding the AIcon2abs course: “I chose to enroll in the AIcon2abs course out of curiosity, hoping to learn more about artificial intelligence and machine learning. The course was offered online, featuring pre-recorded lessons, and began with the story of Zrow, a robot from another planet on a quest to learn about Earth. The narrative of Zrow’s adventures provided a stimulating framework for the course, evoking the dreams of my childhood. This method of presenting the course content was highly engaging. The course materials were of exceptional quality, both technically and pedagogically. The content was intuitive and well-explained, requiring no prior knowledge. This shows that complex topics, like AI and machine learning, can be made accessible to a broad audience. The themes were explored through interactive, simple, and creative activities, including Scratch-based games and animations, which were very engaging. The course was designed to accommodate various age groups, with thoughtful consideration for older adults and children alike. Throughout the course, I gained insights into programming, machine learning, and AI. I learned that machines acquire knowledge through examples and that they operate based on programmed instructions. I also learned about WiSARD, which was surprisingly easy to understand. I gained a sense of the processes that allow machines to ‘learn’ and discovered different methods for training AI. This course opened a new world to me, a realm in which I had been an outsider. Some activities, however, could benefit from additional time and examples. I also believe that face-to-face interaction could enhance learning, particularly for younger participants who may find even brief video content tiring. The course prompted me to reflect on the need for greater awareness of AI. The general public should understand AI to participate in discussions about its future. This is knowledge that should permeate our schools and academic institutions. The course raised several questions: What applications of AI influence our daily lives without our awareness? Will most people remain uninformed, leaving knowledge holders to shape decisions for us? I reflected on the implications of AI for free will and its potential for transforming our understanding of autonomy. This experience broadened my understanding of the world and affirmed the remarkable capacity of the human mind. Ultimately, I found the course to be a truly rewarding experience.” 4.3. Discussion This section discusses the results through the lens of the DSR method as outlined by Dresch et al. (2015) and Pimentel et al. (2019), which guided this research (see Figure 1). We will first evaluate the three primary assessments expected in DSR-based research per Pimentel et al. (2019). Next, we will formalize the construction and contingency heuristics , the core contributions of the DSR method as proposed by Dresch et al. (2015). Finally, we will address some limitations of this study. 4.3.1. Is the artifact valid? Results in Subsections 4.1 and 4.2 indicate that (i) the research objectives (Subsection 1.2) were satisfactorily achieved, (ii) the requirements (Subsection 1.3) were met, and (iii) the anticipated performance (Subsection 1.5) was realized. These outcomes support the conclusion that the AIcon2abs method is valid , meeting the requirements of Evaluation 1 in the DSR map presented in Section 2. 4.3.2. Was the problem solved? The data indicate that most participants found the AIcon2abs course activities easy and that many developed a foundational understanding of machine learning in its most basic form. Furthermore, participants’ satisfaction with the course suggests that learning through AIcon2abs was a rewarding experience. According to Simon (1996), consensus among participants serves as a criterion for determining whether a solution is satisfactory. The data presented here show that participants were generally satisfied, indicating that AIcon2abs is a suitable method for addressing its intended problem . 4.3.3. Do the theoretical conjectures seem to be valid? In Subsection 1.4, we outlined six theoretical conjectures guiding the development of the AIcon2abs method. The findings suggest the validity of these conjectures, particularly the central one, which posits that learning abstract concepts from computer science can be facilitated through practices that highlight observable aspects of these concepts in the concrete world . It is valid to mention that none of the research or approaches found in the Queiroz et al. (2021) literature review had the same target audience as the AIcon2abs method: the general public. Those studies were exclusively dedicated to a specific age group: (i) children, (ii) adolescents, or (iii) undergraduate course students. Therefore, it is worth stating that this study revealed that this theoretical conjecture seems valid for a diverse audience of different ages and backgrounds (including children) . Regarding conjecture 5 (Subsection 1.4), after implementing the AIcon2abs course, we realized that the agent-based approach could be enhanced by integrating it with the Input-Process-Output (IPO) model 18 . This combination could further facilitate understanding the fundamental logic of machine learning in its most basic form. 4.3.4. AIcon2abs construction heuristics Construction heuristics establish rules for the internal environment of the artifact and for its construction. When formalized, these heuristics can aid other researchers in developing artifacts to address similar issues. Construction heuristics represent a core contribution to knowledge within the DSR framework (Dresch et al., 2015; Gregor & Jones, 2007; Simon, 1996; Venable, 2006). Fig.12 presents the formalized construction heuristic derived from the AIcon2abs method’s design, development, implementation, and evaluation. 4.3.5. AIcon2abs contingency heuristics Contingency heuristics specify, when applicable, the adjustments required to adapt an artifact for effective functioning in different external environments (Dresch et al., 2015). For the AIcon2abs method, the external environment is defined as its target audience. During data analysis, some participants highlighted certain aspects of the AIcon2abs method that made it difficult to understand some specific content. Notably, these challenges were exclusively reported by adult participants. The most frequently cited issues included: Some content was perceived as overly "complex," requiring more than a single engagement with the related activity for adequate comprehension. Activities associated with more complex content had insufficient examples to support understanding. The recommended course duration was inadequate for consolidating certain content. These findings represent contingencies that should be addressed to meet the learning objectives of the AIcon2abs method, particularly when applied to an audience with the specified learning needs. The identified contingencies focus on the difficulty of adequately understanding some themes worked on in the AIcon2abs IU. In alignment with the Differentiated Learning educational approach (Tomlinson, 1999), we propose the contingency heuristic illustrated in Figure 13. This heuristic is founded on established theoretical principles and pedagogical practices. However, a new Design Science Research (DSR) cycle, encompassing steps from Artifact Design to Evaluation (see Figure 1), is recommended to validate its efficacy. 4.3.6. Limitations of the study While efforts were made to address threats to the validity of this research, some limitations remain. Regarding external validity , despite the promising results observed for the AIcon2abs method within the context of this study, its effectiveness in other settings is not assured. The method was evaluated based on a relatively small sample of 34 participants. While the sample exhibited diversity in certain respects, it was specific in others: (i) all adults had either completed or were attending higher education, and (ii) all children and adolescents had access to quality education. Concerning internal validity , data were not collected from participants who withdrew from the AIcon2abs course, precluding comparisons between those who completed the course and those who did not. Furthermore, all participants were presumed to be prepared to engage with the course content, potentially influencing the outcomes. These considerations highlight the need for further evaluations of the AIcon2abs method across diverse contexts. Suggestions for future research and potential directions are presented in Subsection 5.2. 5. Conclusion 5.1. Concluding summary This study aimed to address the question: Is AIcon2abs a suitable method to support a diverse audience of different ages and backgrounds in understanding the concept of machine learning in its most basic form in an uncomplicated and satisfactory way ? The results presented here affirmatively answer this question. Moreover, they demonstrate that the AIcon2abs method effectively achieved its objectives by incorporating the following four strategies, which constitute the primary contribution of this research: Employing Active Learning strategies. Presenting the fundamental concept of machine learning (in its most basic form) through concrete examples that are familiar to the target audience and aligned with the adopted machine learning model constructs. Integrating the learning algorithm instantiation, along with training and classification primitives, as basic commands within a conventional programming structure. Utilizing a machine learning algorithm whose training and classification processes are readily observable and replicable, allowing learners to observe the internal actions of the machine during the learning process. Additionally, the formalization of construction and contingency heuristics for the AIcon2abs method (outlined in Subsections 4.3.4 and 4.3.5) constitutes a valuable contribution of this study. Other notable contributions include: (i) the Pre-Posttest Matrix data visualization tool , an adapted use and interpretation of the Confusion Matrix (Section 4), and (ii) the blending of the Design Science Research (DSR) method by Dresch et al. (2015) with elements from the DSR map proposed by Pimentel et al. (2019) (Section 2). The AIcon2abs components were highly rated by nearly all participants—children, adolescents, and adults—who took part in the empirical evaluation. Having been rigorously evaluated and validated , the AIcon2abs method is now available to those interested in demystifying artificial intelligence, particularly machine learning, to the general public. 5.2. Future work To further enhance the AIcon2abs method and broaden its applicability to include new audiences, the following future work would be of significant value: Develop English versions of the instructional videos and playful activities using the WiSARD model. Incorporate the Input-Process-Output (IPO) concept alongside the Intelligent Agents concept in the design of the AIcon2abs instructional unit, to reinforce understanding of the fundamental differences between traditional computer systems and those utilizing AI mechanisms. Produce the additional learning materials proposed in the formalized contingency heuristics. Offer a face-to-face version of the AIcon2abs course to: (i) apply the contingency heuristic developed in this study and (ii) incorporate educational robotics materials in the BlockWiSARD activities. Adapt the AIcon2abs method for illiterate learners. Investigate the feasibility of adapting the AIcon2abs method for individuals with disabilities, such as the visually impaired and the deaf. Explore the potential of using AIcon2abs construction heuristics to develop artifacts introducing other AI paradigms to the general public. Apply the AIcon2abs construction heuristics to design materials addressing other complex computing concepts, such as Object-Oriented Programming Paradigm. Declarations Ethics statement This research has been approved by the CEP/HUCFF/FM/UFRJ 19 human research ethics committee: CAAE: 44444621.7.0000.5257, approval date: June 22, 2021, approval order number: 4.861.208, 2021. All adults, parents, or legal guardians gave their informed consent, and all minors gave their informed assent before participating in this study. Conflict of interest statement The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper. Data availability The datasets used for the data analysis presented in this paper are available at AIcon2abs (2023). Acknowledgements This study was financed in part by the Coordenação de Aperfeiçoamento de Pessoal de Nível Superior-Brasil (CAPES)-Finance Code 001. Funding Coordenação de Aperfeiçoamento de Pessoal de Nível Superior-Brasil (CAPES)- Finance Code 001. ( Graduate Fellowship ). Author Contribution Authors’ ContributionsQUEIROZ, R.L.: Conceptualization, Data curation, Formal analysis, Investigation, Methodology, Project administration, Software, Validation, Visualization, Writing - original draft, review & editing. LIMA, P.M.V.: Conceptualization, Formal analysis, Funding acquisition, Investigation, Methodology, Project administration, Resources, Supervision, Validation, Visualization, and Writing, review & editing. LIMA, C.: Formal analysis, Project administration, Supervision, Validation, Writing - review & editing. SAMPAIO, F.F.: Formal analysis, Project administration, Supervision, Validation, Writing - review & editing. References AIcon2abs (2023). AIcon2abs Empirical Evaluation Data. Mendeley Data , V1. https://doi.org/10.17632/ch6ys3gz82.1 Aleksander, I., De Gregorio, M., França, F.M.G., Lima, P.M.V., & Morton, H, (2009), A brief introduction to weightless neural systems. Proceedings of ESANN 2009 , 299–305. Almeida, V. (2018, August 8). Brasil precisa de estratégia para inteligência artificial (Transl.: Brazil needs a strategy for artificial intelligence). Academia Brasileira de Ciências. Retrieved January 28, 2024, from http://www.abc.org.br/2018/08/08/brasil-precisa-de-estrategia-para-inteligencia-artificial/ Banzi, M. (2012, June 27). How Arduino is open-sourcing imagination . TEDGlobal Edinburgh. Retrieved January 28, 2024, from https://www.ted.com/talks/ massimo_banzi_how_arduino_is_open_sourcing_imagination . Bonwell, C.C., & Eison, J.A. (1991). Active learning: Creating excitement in the classroom . ASHE-ERIC higher education reports. George Washington University. Retrieved January 28, 2024, from https://files.eric.ed.gov/fulltext/ ED336049.pdf. Branch, R.M. (2009). Instructional design: The ADDIE approach . Springer. Caltech (1988). Richard Feynman’s Blackboard at the Time of His Death. Caltech Archives. Retrieved January 28, 2024, from http://archives-dc.library.caltech.edu/ islandora/object/ct1%3A483 Caputo, M. (2019, December 21) MCTIC lança consulta pública para a Estratégia Brasileira de Inteligência Artificial (Transl.: MCTIC launches public consultation for the Brazilian Artificial Intelligence Strategy). Academia Brasileira de Ciências. Retrieved January 28, 2024, from http://www.abc.org.br/2019/12/21/mctic-lanca-consulta-publica-para-a-estrategia-brasileira-de-inteligencia-artificial/ Cohen, L., Manion, L., & Morrison, K. (2002). Research methods in education . Routledge. Creswell, J.W., & Poth, C.N. (2016). Qualitative inquiry and research design: Choosing among five approaches . Sage publications. Dick, W., & Carey, L.M. (1977). Formative Evaluation. In: L.J. Briggs (Ed.), Instructional design: Principles and applications . (pp. 311-333). Educational Technology Publications. Dash, A. (2023).The LLM Advantage: How to Unlock the Power of Language Models for Business Success: A Practical Guide to Leveraging AI for Innovation, Growth, and Competitive Edge . United Kindon. Grazing Minds Publishing, Dresch, A., Lacerda, D.P., & Antunes Jr, J.A.V. (2015). Design science research . Springer. Retrieved January 28, 2024, from http://www.springer.com/gb/book/ 9783319073736 . Eguchi, A. (2010). What is educational robotics? Theories behind it and practical implementation. Proceedings of the Society for Information Technology & Teacher Education International Conference , 4006–4014. Feynman, R.P. (1989). Feynman’s office: the last blackboards. Physics Today , 42 (2), 88. https://doi.org/10.1063/1.2810904 Fleming, N.D. (1995). I'm different; not dumb. Modes of presentation (VARK) in the tertiary classroom. Proceedings of the 1995 Annual Conference of the Higher Education Research and Development Society of Australasia (HERDSA) , 308–313. Gregor, S., & Jones, D. (2007). The anatomy of a design theory. Journal of the Association for Information Systems, 8(5), 312-335. https://doi.org/10.17705/1jais.00129 Kalota, F. 2024. A Primer on Generative Artificial Intelligence. Education Sciences . 14 (2), 172. https://doi.org/10.3390/educsci1402017 Kaner, C., Bach, J., & Pettichord, B. (2001). Lessons learned in software testing . John Wiley & Sons. Kirk, R.E. (1978). Introductory statistics . Brooks/Cole Publishing Company. Maharaj, S., Koon, M.K., Gookool, C., Taylor, M., Birbal, R., & Mohammed, N. (2023). Cambridge Primary Computing Learner's Book - Stage 5 . Hodder Education. Medina, E.M. (2004). Beyond the ballot box: Computer science education and social responsibility. ACM SIGCSE Bulletin , 36(4), 7–10. https://doi.org/10.1145/1041624.1041626 Mitchell, T.M. (1997). Machine intelligence . McGraw-Hill. Mitchell, T.M. (2006). The discipline of machine learning . Carnegie Mellon University, School of Computer Science, Machine Learning Department. Ogawa, R.T., & Malen, B. (1991). Towards rigor in reviews of multivocal literatures: Applying the exploratory case study method. Review of Educational Research , 61(3), 265–286. https://doi.org/10.3102/ 00346543061003265 Papert, S. (1994). The Children’s Machine: Rethinking School in the Age of the Computer . Basic Books. Papert, S.A. (1993). Mindstorms: Children, Computers, and Powerful Ideas . Basic Books. Pearson, K. (1904). Mathematical contributions to the theory of evolution, XIII: On the theory of contingency and its relation to association and normal correlation . Draper’s Co. Research memoirs in biometry series, 1. Reprinted in Pearson, E.A. (Ed). (1948). Karl Pearson’s Early Statistical Papers , Cambridge University Press. Pett, M. A. (2015). Nonparametric statistics for health care research: Statistics for small samples and unusual distributions. Sage Publications. Piaget, J. (1950). The psychology of intelligence . London Routledge & Kegan Paul. Pimentel, M., Filippo, D., & Santoro, F.M. (2019). Design Science Research: fazendo pesquisas científicas rigorosas atreladas ao desenvolvimento de artefatos computacionais projetados para a educação (Transl.: Design Science Research: carrying out rigorous scientific research linked to developing computational artifacts designed for education). Metodologia de Pesquisa Em Informática Na Educação: Concepção Da Pesquisa. SBC. Retrieved January 28, 2024, from https://metodologia.ceie-br.org/wp-content/uploads/2018/10/cap1_5.pdf Queiroz, R.L. (2023). AI from concrete to abstract: demystifying artificial intelligence to the general public. Federal University of Rio de Janeiro (UFRJ). Retrieved January 28, 2024, from https://bit.ly/4cYUuXK Queiroz, R.L., Sampaio, F.F., Lima, C., & Lima, P.M.V. (2021). AI from concrete to abstract: demystifying artificial intelligence to the general public. AI & SOCIETY, 36(3), 877-893. https://doi.org/10.1007/s00146-021-01151-x Resnick, M., Maloney, J., Monroy-Hernández, A., Rusk, N., Eastmond, E., Brennan, K., Millner, A., Rosenbaum, E., Silver, J., & Silverman, B. (2009). Scratch: programming for all. Communications of the ACM , 52(11), 60–67. https://doi.org/10.1145/1592761.1592779 Russell, S.J., & Norvig, P. (2020). Artificial intelligence: a modern approach. Prentice Hall. Sakulkueakulsuk, B., Witoon, S., Ngarmkajornwiwat, P., Pataranutaporn. P., Surareungchai, W., Pataranutaporn, P., & Subsoontorn, P. (2018). Kids making AI: Integrating machine learning, gamification, and social context in STEM education. Proceedings of the 2018 IEEE International Conference on Teaching, Assessment, and Learning for Engineering (TALE), 1005–1010. https://doi.org/10.1109/ TALE.2018.8615249 Schlegel, M.J. (1995). A Handbook of Instructional and Training Program Design. ERIC - Institute of Education Sciences. Retrieved January 28, 2024, from https://files.eric.ed.gov/fulltext/ED383281.pdf Severance, C. (2013). Eben Upton: Raspberry Pi. Computer, 46(10), 14-16. https://doi.org/10.1109/MC.2013.349 Siegel, S. (1957). Nonparametric statistics. The American Statistician , 11(3), 13–19. https://doi.org/10.1080/00031305 .1957.10501091 Simon, H.A. (1996). The sciences of the artificial . MIT press. Tessmer, M. (2013). Planning and conducting formative evaluations . Routledge. Tomlinson, C.A. (1999). Mapping a route toward differentiated instruction. Educational leadership , 57(1), 12-17. https://education.illinoisstate.edu/downloads/linc/linccurriculummodule/Tomlinson.pdf. Accessed 28 January 2024 Van Manen, M. (1990). Beyond assumptions: Shifting the limits of action research. Theory into Practice , 29 (3), 152–157. https://doi.org/10.1080/00405849009543448 Venable, J. (2006). The role of theory and theorizing in design science research. Proceedings of the 1st International Conference on Design Science in Information Systems and Technology (DESRIST 2006), 1–18. Yogeshwar, R. (2018, October 18). Os dados que você entrega acabarão por estrangulá-lo no sistema (Transl.: The data you deliver will eventually strangle you in the system). El País. Retrieved January 28, 2024, from Footnotes Simply put, generative AI generates content, which could be text, images, or multimedia. Chat GPT is an AI Large Language Model (LLM) developed by Open AI to follow instructions in a prompt and provide detailed responses. https://openai.com/blog/chatgpt “LLMs (Large Language Models) are a type of artificial intelligence (AI) that can generate, understand, and interact with natural language, such as text and speech. Natural language is the way we humans communicate with each other, using words, sentences and grammar” (Dash, 2023, p.1) “Generative AI is a form of artificial intelligence that utilizes techniques from machine learning and deep learning to generate original data” (Kalota, 2024) A Multivocal Literature Review (MLR) comprises all accessible writings on a usual, often contemporary topic (Ogawa & Malen, 1991). Arduino is a prototyping platform simple enough to be used by any student, including children. The board's design allows people to "build things that work" without attending a complete electronics course (Banzi, 2012; Severance, 2013). Its low-cost, quality, flexibility, and ease of use have made Arduino electronic prototyping boards an excellent option for developing projects on Educational Robotics http://www.arduino.cc/ Raspberry Pi is a low-cost single-board computer provided with a set of General-Purpose Input/Output pins where one can connect robotics devices. Roughly speaking, we can say that the Raspberry Pi is a single-board computer with a microcontroller (like an Arduino) built into it. The flexibility of the Raspberry Pi allows one to use it for the most diverse purposes, such as experimentation, residential automation, or simply fun. (Severance, 2013). https://www.raspberrypi.org/ https://www.youtube.com/ http://conselho.saude.gov.br/comissoes-cns/conep/ https://www.facebook.com/ https://www.whatsapp.com/ https://www.instagram.com/ https://scratch.mit.edu/ A type of user testing that uses testers who are NOT part of your organization and who are members of your product’s target market. The product under test is typically very close to completion.” (kaner et al., 2008, p.34) https://www.un.org/development/desa/pd/sites/www.un.org.development.desa.pd/files/unpd_egm _201902_s1_sergeischerbov.pdf “In-house testing performed by the test team (and possibly other interested, friendly insiders).” (Kaner et al., 2008, p.34) https://www.ibm.com/spss A computer works based on a cycle defined as Input-Process-Output (IPO). This is an elementary concept widely used in introductory computing and can be summarized as: Input (Anything that is put into a system) - > Process (The steps or actions taken by the system) -> Output (The result of the process) (Maharaj et al., 2023, p.147). https://www.hucff.ufrj.br/comite-de-etica-em-pesquisa/ Additional Declarations No competing interests reported. Cite Share Download PDF Status: Posted Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-5728177","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":397331034,"identity":"8676095a-c823-4619-9e61-c9767681d214","order_by":0,"name":"Rubens Lacerda Queiroz","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA8klEQVRIiWNgGAWjYFACHijNDCIqgJidgYGxgXgtZ4B8ZqK1gABjGxFazGfkHnxc8Mc6sZ+d9+HDn/NsEvczMx98OIPBTk4Xhz6ZG3nJxjPb0hNnNrMbG0huS0vsYWZLNtzAkGxsdgC7FgmJHDNp3obDiRsOs7FJGG47DNTCYyb5gOFA4jbcWsx/8/w5nLgfpCVxDnFazJh52IC2MAO1HGyAatmATwvPu2Rp3rZ04xmH2ZgNG46lGfccBvplhgEev7DnHvzM88datr//GOPDHzU2su3tzQcf9lTYyeHSAgXM6AIGeJVj1TIKRsEoGAWjAAEAxT9UInlTwUIAAAAASUVORK5CYII=","orcid":"","institution":"PPGI, Federal University of Rio de Janeiro","correspondingAuthor":true,"prefix":"","firstName":"Rubens","middleName":"Lacerda","lastName":"Queiroz","suffix":""},{"id":397331035,"identity":"ed8f887a-aaf6-4b5a-a224-25e0c7d9a823","order_by":1,"name":"Cabral Lima","email":"","orcid":"","institution":"PPGI, Federal University of Rio de Janeiro","correspondingAuthor":false,"prefix":"","firstName":"Cabral","middleName":"","lastName":"Lima","suffix":""},{"id":397331036,"identity":"7c284c66-264a-4baf-8531-0ed3b8fac59b","order_by":2,"name":"Fábio Ferrentini Sampaio","email":"","orcid":"","institution":"Polytechnic University of Setúbal","correspondingAuthor":false,"prefix":"","firstName":"Fábio","middleName":"Ferrentini","lastName":"Sampaio","suffix":""},{"id":397331037,"identity":"5f0928cc-7780-4739-affa-e6e8877d8a3b","order_by":3,"name":"Priscila Machado Vieira Lima","email":"","orcid":"","institution":"Federal University of Rio de Janeiro","correspondingAuthor":false,"prefix":"","firstName":"Priscila","middleName":"Machado Vieira","lastName":"Lima","suffix":""}],"badges":[],"createdAt":"2024-12-29 03:08:06","currentVersionCode":1,"declarations":{"humanSubjects":false,"vertebrateSubjects":false,"conflictsOfInterestStatement":false,"humanSubjectEthicalGuidelines":false,"humanSubjectConsent":false,"humanSubjectClinicalTrial":false,"humanSubjectCaseReport":false,"vertebrateSubjectEthicalGuidelines":false},"doi":"10.21203/rs.3.rs-5728177/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-5728177/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":73087417,"identity":"bedaed47-df4b-492b-a57e-a4d2a3bd7391","added_by":"auto","created_at":"2025-01-06 15:05:15","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":369229,"visible":true,"origin":"","legend":"\u003cp\u003eDSR steps and text structure. Adapted from Dresch et al. (2015) and Pimentel et al. (2019)\u003c/p\u003e","description":"","filename":"1.png","url":"https://assets-eu.researchsquare.com/files/rs-5728177/v1/f7b7ac33bfbb3805dc348ee3.png"},{"id":73085251,"identity":"2063bcc0-588a-46a2-bede-1df4dce0c5db","added_by":"auto","created_at":"2025-01-06 14:49:16","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":44497,"visible":true,"origin":"","legend":"\u003cp\u003eAIcon2abs components' method summary\u003c/p\u003e","description":"","filename":"2.png","url":"https://assets-eu.researchsquare.com/files/rs-5728177/v1/1eb81382b3e97ad74a269c92.png"},{"id":73085256,"identity":"5128d403-db84-4867-99b9-5b969a9cde67","added_by":"auto","created_at":"2025-01-06 14:49:16","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":547795,"visible":true,"origin":"","legend":"\u003cp\u003eBar charts for Question 3 answers. a Pre-posttest all subjects’ chart; b Pre-posttest age group chart (Absolute Frequency); c Pre-posttest age group chart (Relative Frequency)\u003c/p\u003e","description":"","filename":"3.png","url":"https://assets-eu.researchsquare.com/files/rs-5728177/v1/e3a2eeb4c9f3e6b486543b95.png"},{"id":73085247,"identity":"8921d873-a23f-4fb3-a292-f3bfb7e9fc74","added_by":"auto","created_at":"2025-01-06 14:49:15","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":198837,"visible":true,"origin":"","legend":"\u003cp\u003eQ03-Pre-posttest Matrix\u003c/p\u003e","description":"","filename":"4.png","url":"https://assets-eu.researchsquare.com/files/rs-5728177/v1/bfcd2ca6510c683f5d49230b.png"},{"id":73085269,"identity":"cf68a1a2-3b9d-47bb-8ecd-9d42fa6afdf8","added_by":"auto","created_at":"2025-01-06 14:49:16","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":577373,"visible":true,"origin":"","legend":"\u003cp\u003eBar charts for Question 4 answers. a Pre-posttest all subjects’ chart; b Pre-posttest age group chart (Absolute Frequency); c Pre-posttest age group chart (Relative Frequency)\u003c/p\u003e","description":"","filename":"5.png","url":"https://assets-eu.researchsquare.com/files/rs-5728177/v1/804fdc8f4cc063b443954ab5.png"},{"id":73085262,"identity":"3b920f77-7e4a-45ee-b455-6e885986bbe6","added_by":"auto","created_at":"2025-01-06 14:49:16","extension":"png","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":308032,"visible":true,"origin":"","legend":"\u003cp\u003eQ07-Pre-posttest all subjects’ chart\u003c/p\u003e","description":"","filename":"6.png","url":"https://assets-eu.researchsquare.com/files/rs-5728177/v1/560d8bf19e75702daa860617.png"},{"id":73086938,"identity":"de1afed6-954b-4741-bb2b-b8cc549d33db","added_by":"auto","created_at":"2025-01-06 14:57:19","extension":"png","order_by":7,"title":"Figure 7","display":"","copyAsset":false,"role":"figure","size":261711,"visible":true,"origin":"","legend":"\u003cp\u003eQ10-Pre-posttest all subjects’ charts\u003c/p\u003e","description":"","filename":"7.png","url":"https://assets-eu.researchsquare.com/files/rs-5728177/v1/74347f3a01e6448aa999f1e6.png"},{"id":73085255,"identity":"d7b2c80d-49b4-452b-8e79-131e2ed9c1a5","added_by":"auto","created_at":"2025-01-06 14:49:16","extension":"png","order_by":8,"title":"Figure 8","display":"","copyAsset":false,"role":"figure","size":188529,"visible":true,"origin":"","legend":"\u003cp\u003eQ10-Pre-posttest Matrix\u003c/p\u003e","description":"","filename":"8.png","url":"https://assets-eu.researchsquare.com/files/rs-5728177/v1/0bfad78a857473b6096b2321.png"},{"id":73085246,"identity":"691de3a8-9403-45ef-b123-c3a0236af346","added_by":"auto","created_at":"2025-01-06 14:49:15","extension":"png","order_by":9,"title":"Figure 9","display":"","copyAsset":false,"role":"figure","size":42217,"visible":true,"origin":"","legend":"\u003cp\u003eQ14-Pos-test all subjects’ charts\u003c/p\u003e","description":"","filename":"9.png","url":"https://assets-eu.researchsquare.com/files/rs-5728177/v1/e797f31defed3b0851e50f32.png"},{"id":73085258,"identity":"0aa05f18-4997-433e-9c0c-d44901f80283","added_by":"auto","created_at":"2025-01-06 14:49:16","extension":"png","order_by":10,"title":"Figure 10","display":"","copyAsset":false,"role":"figure","size":104100,"visible":true,"origin":"","legend":"\u003cp\u003eQ18-Pos-test all subjects’ chart\u003c/p\u003e","description":"","filename":"10.png","url":"https://assets-eu.researchsquare.com/files/rs-5728177/v1/f959f0818f16e2777a00c53f.png"},{"id":73085272,"identity":"48f4c218-52d5-45c4-84b6-90741e708601","added_by":"auto","created_at":"2025-01-06 14:49:17","extension":"png","order_by":11,"title":"Figure 11","display":"","copyAsset":false,"role":"figure","size":103628,"visible":true,"origin":"","legend":"\u003cp\u003eQ20 Pos-test all subjects’ chart\u003c/p\u003e","description":"","filename":"11.png","url":"https://assets-eu.researchsquare.com/files/rs-5728177/v1/70966849a4b6caebe8a8b88a.png"},{"id":73085332,"identity":"df0b290a-ece5-4726-93ae-01ae94d6a34f","added_by":"auto","created_at":"2025-01-06 14:49:19","extension":"png","order_by":12,"title":"Figure 12","display":"","copyAsset":false,"role":"figure","size":434107,"visible":true,"origin":"","legend":"\u003cp\u003eAIcon2abs construction heuristic\u003c/p\u003e","description":"","filename":"12.png","url":"https://assets-eu.researchsquare.com/files/rs-5728177/v1/1e6cd11a1fd773cb91dda18c.png"},{"id":73086937,"identity":"42c697fc-92f1-449b-8d09-fe5027819232","added_by":"auto","created_at":"2025-01-06 14:57:19","extension":"png","order_by":13,"title":"Figure 13","display":"","copyAsset":false,"role":"figure","size":211844,"visible":true,"origin":"","legend":"\u003cp\u003eAIcon2abs contingency heuristic\u003c/p\u003e","description":"","filename":"13.png","url":"https://assets-eu.researchsquare.com/files/rs-5728177/v1/d80aed9bbadc3f5d1baab0d7.png"},{"id":80675761,"identity":"22cd90f6-f726-4c21-9345-fcb208acc68c","added_by":"auto","created_at":"2025-04-15 22:46:28","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":5371916,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-5728177/v1/ada81a14-40be-4962-9721-ca7146eff126.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"How do machines learn? Evaluating the AIcon2abs method","fulltext":[{"header":"1. Introduction","content":"\u003cp\u003eAs noted by Queiroz et al. (2021), Artificial Intelligence (AI) has been increasingly integrated into a wide variety of fields, influencing numerous facets of daily life at an ever-expanding rate. In response to this rapid growth, several scholars have highlighted the critical need to develop a strategic framework for the advancement of AI, stressing that such a strategy must emerge from a reflective process that involves all sectors of society (Almeida, 2018; Caputo, 2019; Medina, 2004; Yogeshwar, 2018).\u003c/p\u003e\n\u003cp\u003eNonetheless, a pivotal question arises: How can individuals effectively participate in discussions and decision-making regarding the future of AI without possessing a foundational understanding of the field? This dilemma underscores an urgent necessity to develop accessible methods that enable the general public to become informed and engaged participants in debates and decisions concerning the adoption and implementation of AI technologies. A major challenge in achieving this goal lies in the inherent complexity of the techniques and concepts underpinning most AI systems (Sakulkueakulsuk et al., 2018). In response, researchers have begun to explore solutions aimed at overcoming these obstacles and fostering a more accessible understanding of AI.\u003c/p\u003e\n\u003cp\u003eRussell and Norvig (2020) classify the AI domain into three principal areas: Search, Knowledge Representation, and Machine Learning. Among these, Machine Learning has emerged as one of the most significant fields, largely responsible for the recent \u0026quot;AI boom.\u0026quot; Most AI systems that individuals encounter in their daily lives are built on Machine Learning principles, as in the case of Generative AI\u003csup\u003e1\u003c/sup\u003e as chat GPT\u003csup\u003e2\u003c/sup\u003e and other LLMs\u003csup\u003e3\u003c/sup\u003e, which combines machine learning mechanisms with other AI approaches to produce generative AI\u003csup\u003e4\u003c/sup\u003e. Consequently, most initiatives focused on demystifying AI for the public have concentrated on the Machine Learning component, as evidenced by the Multivocal Literature Review (MLR)\u003csup\u003e5\u003c/sup\u003e conducted by Queiroz et al. (2021). This review encompassed scientific databases and general internet searches, leading to the identification of various artifacts, approaches, characteristics, and unresolved issues. From these findings, the AIcon2abs method was developed as a potential solution (Queiroz et al., 2021).\u003c/p\u003e\n\u003cp\u003eThe AIcon2abs method was designed to facilitate an introductory understanding of AI for the general public, using Machine Learning as an accessible entry point. \u003cem\u003eIn its most basic form, Machine Learning involves providing an algorithm with examples of a specific class, which enables the system to later classify new observations as belonging or not to that class\u003c/em\u003e. For the purposes of this study, the foundational understanding of Machine Learning is framed by two core concepts: \u003cem\u003e(i) \u003c/em\u003e\u003cstrong\u003e\u003cem\u003eMachine Learning is essentially a process by which computers learn from examples\u003c/em\u003e\u003c/strong\u003e\u003cem\u003e, \u003c/em\u003e\u003cstrong\u003e\u003cem\u003eand \u003c/em\u003e\u003c/strong\u003e\u003cem\u003e(ii)\u003c/em\u003e\u003cstrong\u003e\u003cem\u003e this learning occurs through the execution of a computer program.\u003c/em\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe AIcon2abs method comprises several key components, including a block-based programming environment, a series of interactive educational activities based on the WiSARD weightless artificial neural network model, a set of educational robotics materials, and a standard instructional unit (IU). The block-based programming environment, termed \u003cstrong\u003eBlockWiSARD\u003c/strong\u003e (Queiroz et al., 2021), was specifically designed to support individuals with no prior programming experience in (i) creating programs that enable their computers to learn how to recognize images and (ii) understanding the underlying operations of these programs. Notably, BlockWiSARD can be operated on low-cost computers without the need for a GPU or internet connection.\u003c/p\u003e\n\u003cp\u003eIn BlockWiSARD, higher-level tasks necessary for machine learning, as well as the instantiation of the learning algorithm, are incorporated as commands within the program being developed. This feature is intended to facilitate a clearer distinction between traditional computer programs and those capable of learning from examples. Specifically, BlockWiSARD includes four programming blocks directly related to the training and classification process. By combining these blocks with conventional programming commands, users can easily construct systems that (i) instantiate a learning mechanism, (ii) input data for learning, (iii) assign corresponding labels to that data, and (iv) classify new input data.\u003c/p\u003e\n\u003cp\u003eThe concept of incorporating machine learning primitives into programming languages was originally proposed by Tom Mitchell (2006). In addition to the machine learning blocks, BlockWiSARD features a block responsible for generating a \u003cem\u003emental image\u003c/em\u003e of a given class. These mental images enable users to visualize a prototype of a class learned from training patterns, thereby enhancing their understanding of how machine learning models conceptualize and categorize data.\u003c/p\u003e\n\u003cp\u003eAn essential principle of BlockWiSARD is that each command block within the program must correspond to a tangible action in the concrete world, such as displaying an image on a screen, hearing a sound played a specific number of times, or seeing a light blink. This premise must also be valid for the machine learning related programming blocks. But how can we visualize or observe the internal processes carried out by the machine when the learning algorithm is instantiated? What happens when the Learn and Recognize commands are executed in the program? Additionally, how can we enable individuals to replicate these actions as if they were the algorithm themselves? To this end, the AIcon2abs method incorporates games, animations, and interactive activities centered around the WiSARD model, designed to engage users in a hands-on, experiential learning process (Queiroz et al., 2021; Queiroz, 2023).\u003c/p\u003e\n\u003cp\u003eAs Papert (1994, p.31) said, \u0026quot;The reason you are not a mathematician might well be that you think that math has nothing to do with the body; you have kept your body out of it because it is supposed to be abstract, or perhaps a teacher scolded you for using your fingers to add numbers! This idea [\u0026hellip;] has inspired me to use the computer as a medium to allow children to put their bodies back into their mathematics.\u0026quot; WiSARD is a perfect tool to put people\u0026apos;s bodies into learning AI through ludic activities. These activities also enable learners to observe the internal operations of the machine when the learning algorithm is instantiated within BlockWiSARD. They also allow the visualization of what the machine does when the Learn and Recognize blocks are used in the BlockWiSARD programs. Moreover, the ludic activities make it possible for people to perform all those actions as if they were the WiSARD algorithm itself.\u003c/p\u003e\n\u003cp\u003eBlockWiSARD operates in three modes: (i) Computer Only; (ii) Computer + Arduino\u003csup\u003e6\u003c/sup\u003e, or (iii) Raspberry Pi\u003csup\u003e7\u003c/sup\u003e. The latter two modes enable the integration of BlockWiSARD with low-cost educational robotics platforms. In addition to the core materials, a set of robotics tools compatible with BlockWiSARD was developed as part of the AIcon2abs method (Queiroz et al., 2021). While not essential to the method, these materials complement the learning process and demonstrate how robotics can enhance the educational experience when combined with BlockWiSARD. \u003c/p\u003e\n\u003cp\u003eFor the effective implementation of the AIcon2abs method, the inclusion of a standard instructional unit (IU) is essential. This IU includes an asynchronous e-learning implementation with a workload of approximately 6 hours which was used to evaluate the method, as discussed in Section 3 of this paper.\u003cem\u003e \u003c/em\u003e\u003cstrong\u003e\u003cem\u003eThis evaluation is the focus of this paper.\u003c/em\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe theoretical foundations supporting the design of the AIcon2abs method are rooted in several core concepts and technologies, including: (i) Piaget\u0026apos;s (1950) Constructivist Theory, (ii) Papert\u0026apos;s (1993) Constructionist Theory, (iii) Russell and Norvig\u0026apos;s (2020) concept of Intelligent Agents, (iv) the Machine Learning process proposed by Mitchell (1997), (v) Turing\u0026apos;s (1950) Imitation Game, (vi) Feynman\u0026apos;s (1989) notion of learning through construction, (vii) Block-based programming (Resnick et al., 2009), (viii) the WiSARD weightless artificial network (Aleksander et al., 2009), (ix) Active Learning methodologies (Bonwell \u0026amp; Eison, 1991), and (x) educational robotics (Eguchi, 2010; Papert, 1994). These concepts and their interconnections with the AIcon2abs method are explored in further detail by Queiroz et al. (2021) and Queiroz (2023).\u003c/p\u003e\n\u003cp\u003e1.1 Research question\u003c/p\u003e\n\u003cp\u003eQueiroz et al. (2021) provide a comprehensive theoretical framework for the conceptualization of the AIcon2abs method and translate these theoretical ideas into functional artifacts with significant potential to address the problem the method aims to solve. \u003cem\u003eHowever, in that first study, the authors did not conduct an empirical evaluation of the method\u003c/em\u003e. In light of this, the present study seeks to empirically evaluate the AIcon2abs method in order to answer the following research question: \u003cstrong\u003e\u003cem\u003eIs\u003c/em\u003e\u003cem\u003e \u003c/em\u003e\u003c/strong\u003e\u003cstrong\u003e\u003cem\u003eAIcon2abs\u003c/em\u003e\u003c/strong\u003e\u003cstrong\u003e\u003cem\u003ea suitable method to support a diverse audience of different ages and backgrounds in understanding the concept of machine learning in its most basic form in an uncomplicated and satisfactory way?\u003c/em\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e1.2. Objectives\u003c/p\u003e\n\u003cp\u003e1.2.1. General objective\u003c/p\u003e\n\u003cp\u003eThe primary aim of this study is to evaluate the AIcon2abs method in order to assess its potential strengths and weaknesses as a tool for helping an audience of eight years old and over to understand satisfactorily, in an uncomplicated way, the concept of machine learning in its most basic form. The concept of \u003cem\u003esatisfactory solution\u003c/em\u003e, as defined by Simon (1996), refers to a solution that adequately addresses a given problem. In this context, a satisfactory solution may be achieved in two ways: (i) through consensus among the stakeholders involved in the problem or (ii) by advancing the current solution relative to previous solutions provided by earlier artifacts (Dresch et al., 2015).\u003c/p\u003e\n\u003cp\u003e1.2.2. Specific objectives\u003c/p\u003e\n\u003cp\u003eThe specific objectives of this research are as follows: (i) to evaluate the artifacts developed as part of the AIcon2abs method, ensuring they meet the specified requirements; (ii) to assess the validity of the theoretical conjectures formulated in Subsection 1.4; and (iii) to formalize the construction and contingency heuristics of the AIcon2abs method (see Subsections 4.3.4 and 4.3.5), which constitute the primary research contributions outlined by the Design Science Research (DSR) methodology proposed by Dresch et al. (2015) (Section 2).\u003c/p\u003e\n\u003cp\u003e1.3. AIcon2abs method requirements\u003c/p\u003e\n\u003cp\u003eThe AIcon2abs method was developed with the intention of fulfilling the following requirements (Queiroz et al., 2021):\u003c/p\u003e\n\u003col class=\"decimal_type\"\u003e\n\u003cli\u003eIt must be suitable for individuals both with and without fully developed abstraction capabilities.\u003c/li\u003e\n\u003cli\u003eIt must incorporate a machine learning algorithm that facilitates the easy visualization and manipulation of its underlying logic through interactive visual elements.\u003c/li\u003e\n\u003cli\u003eIt must provide a block-based programming environment that enables non-expert users (including children) to create basic programs capable of learning. \u003cem\u003eThe program environment must include the chosen learning algorithm\u0026apos;s instantiation, training, and classification commands among its primitives\u003c/em\u003e, and it must be compatible with low-cost robotic platforms.\u003c/li\u003e\n\u003cli\u003eThe method should adopt Active Learning methodologies, including narratives, challenges, interactive activities, and games.\u003c/li\u003e\n\u003cli\u003e\u003cem\u003eIt must be designed to operate effectively on low-cost computers which do not require a Graphics Processing Unit (GPU) or an internet connection.\u003c/em\u003e\u003c/li\u003e\n\u003c/ol\u003e\n\u003cp\u003e1.4. AIcon2abs theoretical conjectures\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eTheoretical conjectures\u003c/em\u003e represent assumptions regarding human or social behavior, grounded in theories, models, and constructs from various disciplines. Within the framework of Design Science Research (DSR), these conjectures guide the design of the artifact, and the artifact is then employed to test the validity of these conjectures (Pimentel et al., 2019). The core theoretical conjecture that informed the design and development of the AIcon2abs method is as follows:\u003c/p\u003e\n\u003col\u003e\n\u003cli\u003e\u003cstrong\u003e\u003cem\u003eLearning abstract concepts from computer science can be facilitated through practices that highlight observable aspects of these concepts in the concrete world.\u003c/em\u003e\u003c/strong\u003e\u003c/li\u003e\n\u003c/ol\u003e\n\u003cp\u003eAdditional theoretical conjectures that contributed to the design of the AIcon2abs method include:\u003c/p\u003e\n\u003col class=\"decimal_type\" start=\"2\"\u003e\n\u003cli\u003eAlgorithms with explicit visual representations enhance understanding for non-expert audiences, including children.\u003c/li\u003e\n\u003cli\u003eA machine learning algorithm capable of online learning (interleaving training and classification) and generalizing from only one or a few trained examples facilitates the understanding of the machine learning process described by Mitchell (1997).\u003c/li\u003e\n\u003cli\u003eGeneral users, including children, can grasp the distinction between conventional computer programs and those capable of learning by: (i) integrating the instantiation of the machine learning algorithm, along with training and classification primitives, into basic programming constructs; (ii) adopting an agent-based approach; and (iii) utilizing the machine learning process framework articulated by Mitchell (1997).\u003c/li\u003e\n\u003cli\u003eThe essence of the Imitation Game (Turing, 1950) provides a valuable tool for comprehending AI, particularly by comparing the performance of a human and a machine in tasks requiring human-like intelligence. \u003c/li\u003e\n\u003c/ol\u003e\n\u003cp\u003e1.5. Expected research subjects\u0026rsquo; performance\u003c/p\u003e\n\u003cp\u003eBy adhering to the theoretical conjectures and fulfilling the outlined requirements, the AIcon2abs method is expected to enable its target audience to achieve the following outcomes:\u003c/p\u003e\n\u003col start=\"1\" type=\"1\"\u003e\n\u003cli\u003eDevelop the ability to create programs utilizing basic programming structures such as loops and conditionals.\u003c/li\u003e\n\u003cli\u003e\u003cem\u003eReplicate, through ludic activities, the specific steps performed by the machine learning algorithm used in the programs they develop\u003c/em\u003e.\u003c/li\u003e\n\u003c/ol\u003e\n\u003col start=\"3\"\u003e\n\u003cli\u003e\u003cstrong\u003e\u003cem\u003eUnderstand the fundamental process of teaching a machine to learn from examples using machine learning\u003c/em\u003e\u003c/strong\u003e\u003cstrong\u003e\u003cem\u003e.\u003c/em\u003e\u003c/strong\u003e\u003c/li\u003e\n\u003c/ol\u003e"},{"header":"2. Research methodology and text structure","content":"\u003cp\u003eTo address our research question, we employed the Design Science Research (DSR) method outlined by Dresch et al. (2015), incorporating adaptations from the DSR map elements developed by Pimentel et al. (2019). Notably, the DSR map by Pimentel et al. (2019) does not constitute a research method per se; rather, it serves as a framework to support DSR-oriented investigations, irrespective of the specific DSR method adopted. By integrating these approaches, we established a process comprising 14 steps organized into six clusters (see Figure 1). This article is structured in alignment with the DSR steps depicted in the diagram presented in Fig. 1, with a particular focus on the evaluation of the AIcon2abs method and the communication of findings, covering steps 10 through 14.\u003c/p\u003e\n\u003cp\u003eSection 1 provides the context for this study, presenting an overview of the AIcon2abs method\u0026rsquo;s conception, structure, and the objectives of this research. Section 2 delineates the research methodology, followed by Section 3, which details the design, development, and implementation of the AIcon2abs standardized user interface (UI). Sections 4 and 5 then concentrate on the evaluation of the AIcon2abs method. A comprehensive discussion of steps 1 through 9 can be found in Queiroz et al. (2021) and Queiroz (2023), the latter being Queiroz\u0026apos;s Doctoral Thesis.\u003c/p\u003e\n\u003cp\u003eDresch et al. (2015) suggest that DSR-based research may integrate data collection and analysis techniques from other methodological frameworks, such as case studies and experimental designs. In this study, we adopted a pre-experimental design alongside a phenomenological approach to data collection and analysis. A pre-experiment involves limited control over the variables involved in the process, employing a single group (the experimental group) where the intervention is implemented without a control group (Cohen et al., 2002). For the phenomenological component, we applied the approach suggested by Creswell (2016), which frames phenomenology as a qualitative method to explore the shared meanings individuals ascribe to their lived experiences regarding a specific phenomenon (Van Manen, 1990).\u003c/p\u003e"},{"header":"3. AIcon2abs instructional unit","content":"\u003cp\u003eFigure 2 illustrates a comprehensive diagram of the AIcon2abs method\u0026apos;s components. The standardized AIcon2abs interface (IU) was implemented as the primary tool in the empirical study to rigorously assess and validate the method. For additional information on the BlockWiSARD block-based programming environment, Ludic Activities related to the WiSARD Weightless Artificial Neural Network (WANN), and AIcon2abs Educational Robotics Materials, refer to Queiroz et al. (2021). \u003c/p\u003e\n\u003cp\u003e3.1. AIcon2abs IU components\u003c/p\u003e\n\u003cp\u003eThe development of the AIcon2abs IU adhered to the ADDIE instructional design model (Analyze, Design, Develop, Implement, and Evaluate) as advocated by Branch (2009) and Schlegel (2012). Among various instructional design frameworks, ADDIE is recognized for its efficacy in addressing complex educational scenarios, making it well-suited for the creation of comprehensive learning resources (Branch, 2009).\u003c/p\u003e\n\u003cp\u003eTo facilitate this empirical evaluation, forty-one video lessons and fourteen interactive ludic activities were developed in Portuguese for the AIcon2abs IU. These learning resources were organized into seven instructional units as follows: (1) Introduction; (2) Introducing the Story of Zrow; (3) Transforming Your Computer into a Zrow Prototype; (4) Exploring BlockWiSARD \u0026ndash; Zrow\u0026apos;s Programming Environment; (5) Teaching Zrow about Star and Flower Shapes; (6) WiSARD \u0026ndash; The Learning Engine of Zrow; and (7) Zrow\u0026apos;s Departure. All AIcon2abs reference materials (in Portuguese) are available in Queiroz (2023). The Portuguese video lessons and ludic activities can also be accessed via the following YouTube\u003csup\u003e8\u003c/sup\u003e playlists:\u003c/p\u003e\n\u003col\u003e\n\u003cli\u003eInvitation to the AIcon2abs Course (https://www.youtube.com/playlist?list=PLGOt2oAva9YDbXosELwFw_tv3CHL4kkOg)\u003c/li\u003e\n\u003cli\u003eAI for the General Public: Introduction (https://www.youtube.com/playlist?list=PLGOt2oAva9YCHZaaYQeVq2D8TiasBnwc-)\u003c/li\u003e\n\u003cli\u003eAI for the General Public: Introducing the Story of Zrow (https://www.youtube.com/playlist?list=PLGOt2oAva9YCjFb3Of5NrZdN2PJV1SWn6)\u003c/li\u003e\n\u003c/ol\u003e\n\u003col start=\"4\"\u003e\n\u003cli\u003eAIcon2abs Course Videos (https://www.youtube.com/playlist?list=PLGOt2oAva9YB-nlAZgqpeuHp53uQLTE7x)\u003c/li\u003e\n\u003c/ol\u003e\n\u003cp\u003eExamples of interactive activities in English are also available at:\u003c/p\u003e\n\u003col start=\"1\" type=\"1\"\u003e\n\u003cli\u003eThe Digitized Game (https://scratch.mit.edu/projects/453213702)\u003c/li\u003e\n\u003cli\u003eYou Are a WiSARD! (https://scratch.mit.edu/projects/440606183)\u003c/li\u003e\n\u003c/ol\u003e\n\u003col start=\"3\"\u003e\n\u003cli\u003eYou Are a WiSARD with Bleaching! (https://scratch.mit.edu/projects/440606183)\u003c/li\u003e\n\u003c/ol\u003e\n\u003cp\u003eThe instructional design included the formation of three Google Classroom groups segmented by age: children (8\u0026ndash;12 years), adolescents (12\u0026ndash;17 years), and adults (18+ years), in compliance with the Brazilian National Research Ethics Committee (Comiss\u0026atilde;o Nacional de \u0026Eacute;tica em Pesquisa \u0026ndash; CONEP\u003csup\u003e9\u003c/sup\u003e) guidelines. All groups accessed the same learning content and environment. Prior to deployment, navigability testing was conducted: all video lessons were viewed, and all activities were completed to identify and rectify any potential issues.\u003c/p\u003e\n\u003cp\u003eTo promote the course, materials were created in three formats\u0026mdash;text, image, and video\u0026mdash;and disseminated via YouTube, Facebook\u003csup\u003e10\u003c/sup\u003e, WhatsApp\u003csup\u003e11\u003c/sup\u003e, and Instagram\u003csup\u003e12\u003c/sup\u003e. Video-based student guides were also developed, providing instructions on course progression and usage of the virtual learning platform. \u003cstrong\u003ePre-test and post-test questionnaires\u003c/strong\u003e were prepared and diostributed through Google Forms. \u003c/p\u003e\n\u003cp\u003e3.2. AIcon2abs IU implementation\u003c/p\u003e\n\u003cp\u003eFollowing the development and testing of the AIcon2abs standardized IU, the instructional materials created for this method were delivered through an online course titled \u0026quot;AIcon2abs\u0026quot; (\u003cem\u003eAI from Concrete to Abstract, Demystifying Artificial Intelligence to the General Public\u003c/em\u003e). Details of this process are provided in Subsections 3.2.1 and 3.2.2. To access the AIcon2abs course for the method evaluation, users can follow the steps below using a Google account:\u003c/p\u003e\n\u003col\u003e\n\u003cli\u003e\u003cstrong\u003eUse the invitation link:\u003c/strong\u003e https://classroom.google.com/c/NjE3NDgyNjEwMzE3?cjc=6bndekk\u003c/li\u003e\n\u003c/ol\u003e\n\u003cp\u003eor\u003c/p\u003e\n\u003col class=\"decimal_type\"\u003e\n\u003cli\u003eAccess \u003cem\u003e\u003cu\u003eGoogle Classroom\u003c/u\u003e\u003c/em\u003e.\u003c/li\u003e\n\u003cli\u003eClick on the (+) icon in the upper corner of the page.\u003c/li\u003e\n\u003cli\u003eChoose the option \u003cem\u003eJoin Class\u003c/em\u003e.\u003c/li\u003e\n\u003cli\u003eEnter the class code \u003cstrong\u003e6bndekk \u003c/strong\u003eand click \u003cem\u003eJoin\u003c/em\u003e.\u003c/li\u003e\n\u003cli\u003eFinally, click on the classwork tab. \u003c/li\u003e\n\u003c/ol\u003e\n\u003cp\u003e3.1.1. Initiation, conduction, and completion of instruction\u003c/p\u003e\n\u003cp\u003eParticipants had the flexibility to engage with the course activities at their convenience, allowing for asynchronous participation and independent task completion. This design permitted participants to join the course at different times and progress at their own pace. Any questions or feedback from participants were managed through messages within the platform.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eThe AIcon2abs course was completed by thirty-four participants aged 8 to 72\u003c/strong\u003e. This cohort included twenty-four adults (aged 21\u0026ndash;72), five adolescents (aged 12\u0026ndash;17), and five children (aged 8\u0026ndash;11). Of the participants, 53% had previous programming experience, while 23% had prior knowledge of AI or machine learning. Participants\u0026rsquo; educational backgrounds were diverse, with 32% in Social, Biological, and Human Sciences, 38% in the exact sciences and engineering, and 29% in basic education. Geographically, 47% resided in Southeast Brazil, and 53% in the Southern region of the country.\u003c/p\u003e\n\u003cp\u003e3.1.2. Data collection instruments\u003c/p\u003e\n\u003cp\u003eThe primary instrument for data collection was a set of \u003cstrong\u003epre-test and post-test questionnaires\u003c/strong\u003e comprising both open-ended and closed questions. Closed questions included nominal (yes/no and multiple-choice) and ordinal (Likert scale) formats. The pre-test questionnaire contained 11 items, while the post-test questionnaire consisted of 15 items. These questionnaires are available in Queiroz (2023). Secondary data, including public posts, private messages within the learning platform, and email communications, also contributed to the data collection process.\u003c/p\u003e"},{"header":"4. Results","content":"\u003cp\u003eAs outlined in Section 2, data from the pre-test and post-test questionnaires, along with secondary data sources, were analyzed from two perspectives: (i) \u003cstrong\u003eas part of a mixed-methods pre-experiment\u0026nbsp;\u003c/strong\u003e(\u003cstrong\u003eincluding exploratory data analysis and hypothesis testing\u003c/strong\u003e) and (ii) \u003cstrong\u003efrom\u003c/strong\u003e\u003cstrong\u003e\u0026nbsp;a\u0026nbsp;\u003c/strong\u003e\u003cstrong\u003ephenomenological perspective\u003c/strong\u003e (a qualitative examination capturing the essence of participants\u0026apos; experiences with the AIcon2abs course).\u003c/p\u003e\n\u003cp\u003eThis section presents findings related to participants\u0026apos; comprehension of the machine learning process and their evaluation of the usability and clarity of the AIcon2abs method components (see Table 1). The questions used in this analysis, available in Queiroz (2023), focused on participants\u0026apos; basic understanding of machine learning (\u003cstrong\u003ein its most basic form\u003c/strong\u003e) before and after the course, as well as their insights regarding the course materials and their overall experience in the AIcon2abs course.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eTable 1\u0026nbsp;List of questions used for the data analysis presented in this article\u003c/p\u003e\n\u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\" width=\"605\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 76px;\"\u003e\n \u003cp\u003eKind\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 57px;\"\u003e\n \u003cp\u003eID\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 378px;\"\u003e\n \u003cp\u003eQuestion\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 94px;\"\u003e\n \u003cp\u003eData type\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 76px;\"\u003e\n \u003cp\u003ePre-post\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 57px;\"\u003e\n \u003cp\u003eQ03\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 378px;\"\u003e\n \u003cp\u003eIf you believe these devices can learn new things, tell me how you think they are taught\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 94px;\"\u003e\n \u003cp\u003eCategorized\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 76px;\"\u003e\n \u003cp\u003ePre-post\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 57px;\"\u003e\n \u003cp\u003eQ04\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 378px;\"\u003e\n \u003cp\u003eIf you believe that the computer can recognize an image it has never \u0026quot;seen\u0026quot; but is similar to other photos it has \u0026quot;seen,\u0026quot; tell me: how do you think it can do that?\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 94px;\"\u003e\n \u003cp\u003eCategorized\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 76px;\"\u003e\n \u003cp\u003ePre-post\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 57px;\"\u003e\n \u003cp\u003eQ07\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 378px;\"\u003e\n \u003cp\u003eWhat does it take to \u0026quot;transform\u0026quot; a conventional computer program into a program capable of learning?\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 94px;\"\u003e\n \u003cp\u003eCategorized\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 76px;\"\u003e\n \u003cp\u003ePre-post\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 57px;\"\u003e\n \u003cp\u003eQ10\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 378px;\"\u003e\n \u003cp\u003eWhat is machine learning?\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 94px;\"\u003e\n \u003cp\u003eCategorized\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 76px;\"\u003e\n \u003cp\u003ePost\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 57px;\"\u003e\n \u003cp\u003eQ14\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 378px;\"\u003e\n \u003cp\u003eHas your understanding of machine learning changed after you took the course?\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 94px;\"\u003e\n \u003cp\u003eCategorical\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 76px;\"\u003e\n \u003cp\u003ePost\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 57px;\"\u003e\n \u003cp\u003eQ15\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 378px;\"\u003e\n \u003cp\u003eWhat has changed?\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 94px;\"\u003e\n \u003cp\u003eOpen\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 76px;\"\u003e\n \u003cp\u003ePost\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 57px;\"\u003e\n \u003cp\u003eQ18\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 378px;\"\u003e\n \u003cp\u003eDid you understand the programs made with BlockWiSARD?\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 94px;\"\u003e\n \u003cp\u003eOrdinal\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 76px;\"\u003e\n \u003cp\u003ePost\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 57px;\"\u003e\n \u003cp\u003eQ20\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 378px;\"\u003e\n \u003cp\u003eDid you understand how WiSARD works by doing the Scratch\u003ca href=\"#_ftn1\" name=\"_ftnref1\" title=\"\"\u003e\u003c/a\u003e\u003csup\u003e13\u003c/sup\u003e [interactive ludic] activities?\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 94px;\"\u003e\n \u003cp\u003eOrdinal\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 76px;\"\u003e\n \u003cp\u003ePost\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 57px;\"\u003e\n \u003cp\u003eQ23\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 378px;\"\u003e\n \u003cp\u003eDid you like to take the AIcon2abs course? Were you satisfied with the things you learned during the process?\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 94px;\"\u003e\n \u003cp\u003eOrdinal\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 76px;\"\u003e\n \u003cp\u003ePost\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 57px;\"\u003e\n \u003cp\u003eQ25\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 378px;\"\u003e\n \u003cp\u003eWhat did you learn by taking this course?\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 94px;\"\u003e\n \u003cp\u003eOpen\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 76px;\"\u003e\n \u003cp\u003ePost\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 57px;\"\u003e\n \u003cp\u003eQ26\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 378px;\"\u003e\n \u003cp\u003eDid the course bring you any reflections? Which?\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 94px;\"\u003e\n \u003cp\u003eOpen\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003eFor this analysis, we employed two types of categorical data: (i) \u003cstrong\u003einherently categorical data and\u003c/strong\u003e (ii) \u003cstrong\u003ederived categorical data\u003c/strong\u003e. Inherently categorical data were obtained from responses to questions with predefined options, such as Yes/No and multiple-choice answers. Derived categorical data, on the other hand, were generated by categorizing responses to open-ended questions. To avoid influencing participants\u0026apos; responses, some questions were intentionally left open-ended without predefined options. Categorization was applied when responses could be grouped by thematic similarity, enabling both visualization and quantification of the data. Consequently, wherever feasible, answers to open-ended questions were systematically categorized.\u003c/p\u003e\n\u003cp\u003eTable 2 presents examples of categorized responses to the question: \u0026quot;\u003cem\u003eIf you believe these devices can learn new things, explain how you think they are taught\u003c/em\u003e.\u0026quot; A comprehensive categorization of all open-ended responses is available in AIcon2abs (2023). The categorization and scoring process for these responses adhered to the methodologies proposed by Smith (2000) and Woike (2007). Categorical and categorized data visualization and analysis are presented for the full participant dataset, with certain visualizations organized by age group where relevant. Since the analyzed data are exclusively categorical or ordinal, calculations and visualization tools appropriate solely for interval and ratio data in exploratory analysis were not employed in this study.\u003c/p\u003e\n\u003cp\u003eTable 2 Examples of categorization of open-ended answers\u003c/p\u003e\n\u003ctable border=\"0\" cellspacing=\"0\" cellpadding=\"0\" width=\"100%\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 79px;\"\u003e\n \u003cp\u003eOriginal answer\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 20px;\"\u003e\n \u003cp\u003eCategorization\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 79px;\"\u003e\n \u003cp\u003e\u003cem\u003eThey are taught through the presentation of some examples of a particular class so that, later, reading, and interpreting patterns.\u003c/em\u003e\u003c/p\u003e\u0026nbsp;\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 20px;\"\u003e\n \u003cp\u003eWith examples\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 79px;\"\u003e\n \u003cp\u003e\u003cem\u003eI\u0026apos;m not sure, but I think these devices learn from training. For example, a robot can be \u0026quot;trained\u0026quot; to recognize facial expressions\u003c/em\u003e.\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 20px;\"\u003e\n \u003cp\u003eTrained by humans\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 79px;\"\u003e\n \u003cp\u003e\u003cem\u003eThey are taught with programs and images\u003c/em\u003e.\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 20px;\"\u003e\n \u003cp\u003eComputer programs\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 79px;\"\u003e\n \u003cp\u003e\u003cem\u003eWith data provided by different users over time\u003c/em\u003e.\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 20px;\"\u003e\n \u003cp\u003eCollecting data\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 79px;\"\u003e\n \u003cp\u003e\u003cem\u003eMachine Learning\u003c/em\u003e.\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 20px;\"\u003e\n \u003cp\u003eMachine learning\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 79px;\"\u003e\n \u003cp\u003e\u003cem\u003eDo not learn\u003c/em\u003e.\u003c/p\u003e\u0026nbsp;\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 20px;\"\u003e\n \u003cp\u003eDo not learn\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003eIn the case of pre- and post-test questions, a modified application and interpretation of the confusion matrix was utilized as a data visualization tool, which we termed the \u003cstrong\u003ePre-Posttest Matrix\u003c/strong\u003e. Originating with Pearson (1904), the confusion matrix, or contingency table, is defined as follows: \u0026quot;A confusion matrix for an N-way classification task is an N by N matrix where the cell (x, y) contains the number of times an item with correct classification x was classified by the model as y\u0026quot; (Jurafsky \u0026amp; Martin, 2008, p.35). Correspondingly, \u003cstrong\u003ea Pre-posttest Matrix for an \u003cem\u003eN\u003c/em\u003e-possible-answers pretest-posttest question \u003cem\u003eQ\u003c/em\u003e is an \u003cem\u003eN\u003c/em\u003e by N matrix where the cell (\u003cem\u003ex\u003c/em\u003e, \u003cem\u003ey\u003c/em\u003e) contains the number of times a question \u003cem\u003eQ\u003c/em\u003e answer changed its value from \u003cem\u003ex\u003c/em\u003e in the pre-test to \u003cem\u003ey\u003c/em\u003e in the post-test\u003c/strong\u003e\u003cstrong\u003e. The Pre-Posttest Matrix is designed to illustrate more explicitly the shifts in participants\u0026rsquo; responses to a given question between the pre-test and post-test\u003c/strong\u003e.\u003c/p\u003e\n\u003cp\u003eIt is also noteworthy that, prior to the implementation of the AIcon2abs course, a beta test\u003ca href=\"#_ftn2\" name=\"_ftnref2\" title=\"\"\u003e\u003c/a\u003e\u003csup\u003e14\u003c/sup\u003e was conducted to evaluate the ludic activities centered around the WiSARD model. This test was administered to undergraduate students enrolled in an introductory computer programming course. After engaging in these activities, the students successfully implemented fully functional WiSARDs based solely on the knowledge they acquired from the exercises. All participants rated the activities involving the WiSARD model as accessible and engaging. Informed consent was obtained from all students prior to their participation in the beta test.\u003c/p\u003e\n\u003cp\u003e4.1. Data analysis from a pre-experiment perspective\u003c/p\u003e\n\u003cp\u003eThis section provides an exploratory analysis and \u003cstrong\u003ehypothesis testing\u003c/strong\u003e using a mixed-methods approach, based on data collected and analyzed within a pre-experimental design framework. For the purpose of this analysis, the adult participant group was subdivided into two distinct age categories: (i) Adults, comprising individuals aged 21 to 59, and (ii) Seniors, comprising individuals aged 60 to 72. This age-based division aligns with the United Nations guidelines\u003csup\u003e15\u003c/sup\u003e\u003csup\u003e.\u003c/sup\u003e.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e4.1.1. \u003cstrong\u003ePre-posttest questions exploratory analysis\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eQ03-Pre-posttest:\u003c/strong\u003e If you believe these devices can learn new things, tell me how you think they are taught. (Categorized)\u003c/p\u003e\n\u003cp\u003eThe bar chart in Figure 3a illustrates that, during the pre-test phase (labeled as Q1), none of the participants expressed the concept of using examples to teach a computer. Additionally, only three participants identified training as a method for teaching a computer. However, in the post-test phase (labeled as Q2), the number of participants who verbalized these concepts increased to 16, indicating an 80% rise. Notably, of the six participants who either did not know or did not respond to the question during the pre-test, only one retained this non-response in the post-test. The stacked bar charts in Figure 3b and Figure 3c further highlight a perceptible improvement in the quality of responses to this question across all age groups from pre-test to post-test.\u003c/p\u003e\n\u003cp\u003eThe Pre-Post Test Matrix presented in Figure 4 enables the observation of specific response changes among research participants. For instance, four participants who initially responded with \u0026quot;Collecting data\u0026quot; in the pre-test modified their answer to \u0026quot;With examples\u0026quot; in the post-test. Additionally, the matrix reveals the distribution of responses provided in the post-test by the 11 participants who had initially answered \u0026quot;Other\u0026quot; in the pre-test, as well as by the 6 participants who had either not responded or indicated a lack of knowledge regarding this question prior to participating in the course.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eQ04-Pre-posttest:\u003c/strong\u003e If you believe that the computer can recognize an image it has never \u0026quot;seen\u0026quot; but is similar to other photos it has \u0026quot;seen,\u0026quot; tell me: how do you think it can do that? (Categorized)\u003c/p\u003e\n\u003cp\u003eThe bar chart in Figure 5a indicates that approximately 44% of participants (35.3% + 8.8%) who did not verbalize concepts related to training, classification, and generalization for classifying new observations in the pre-test articulated these concepts in the post-test. Furthermore, the stacked bar charts in Figure 5b and Figure 5c demonstrate that this shift in responses occurred consistently across all age groups.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eQ07-Pre-posttest\u003c/strong\u003e: What is required to \u0026quot;transform\u0026quot; a conventional computer program into a program capable of learning? (Categorized)\u003c/p\u003e\n\u003cp\u003eA comparison of the pre-test and post-test responses to Q07 (as shown in Figure 6) reveals that numerous participants completed the course by providing answers aligned with the material presented in the AIcon2abs course. Ideas such as \u0026quot;Include a learning mechanism\u0026quot; and \u0026quot;Train with examples,\u0026quot; which were absent in the pre-test responses, were verbalized by many participants in the post-test.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eQ10-Pre-posttest:\u003c/strong\u003e What is machine learning? (Categorized)\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eThe analysis of the bar chart related to responses to question Q10 (Figure 7) reveals that approximately 32% of participants (26.5% + 5.9%) who did not initially articulate the concepts of \u0026quot;Teaching the computer with examples\u0026quot; or \u0026quot;Humans training computers\u0026quot; as definitions of machine learning in the pre-test expressed these ideas in the post-test. Additionally, all participants who were unsure or did not respond to the pre-test question (35.3%) provided answers in the post-test. Notably, eleven participants (approximately 32% of the total) began the course by offering responses already consistent with the course material.\u003c/p\u003e\n\u003cp\u003eThe Pre-Posttest Matrix in Figure 8 further indicates that among those participants who did not answer or stated they did not know the answer in the pre-test, three changed their responses to \u0026quot;Teaching the computer with examples,\u0026quot; one to \u0026quot;Pattern recognition,\u0026quot; two to \u0026quot;Humans training computers,\u0026quot; two to \u0026quot;A computer program,\u0026quot; while two did not incorporate any of the machine learning concepts covered in the course into their post-test responses. Additionally, of the twelve individuals who provided alternative responses in the pre-test, one adjusted their answer to \u0026quot;The machine generalizing what was taught to it,\u0026quot; five to \u0026quot;Teaching the computer with examples,\u0026quot; and two to \u0026quot;A computer program.\u0026quot; Four participants concluded the course without demonstrating clear concepts of machine learning in their responses.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eQ14-Pos-test:\u003c/strong\u003e Has your understanding of machine learning changed after you took the course? (Categorical) / Q15-Pos-test: What has changed? (Open)\u003c/p\u003e\n\u003cp\u003eThe bar chart displaying responses to Question 14 (Figure 9) suggests that the course contributed significantly to altering the participants\u0026apos; understanding of machine learning. Illustrative responses to the open-ended question \u0026ldquo;What has changed in your understanding of machine learning?\u0026rdquo; (Q15) include: (i) \u0026ldquo;I now understand how this learning process functions internally,\u0026rdquo; (ii) \u0026ldquo;I have learned that machines acquire knowledge through examples,\u0026rdquo; (iii) \u0026ldquo;I realized that machines can be continuously trained and improve their accuracy over time,\u0026rdquo; and (iv) \u0026ldquo;I previously believed there was only one approach to learning\u0026rdquo;.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e\u003cem\u003eQ18-Pos-test:\u003c/em\u003e\u003c/strong\u003e Did you understand the programs made with BlockWiSARD Been 1 = \u0026quot;I did not understand\u0026quot; and 5 = \u0026quot;I totally understand\u0026quot; how would you rate your understanding of the programs? (Ordinal) / Q20-Posttest: Did you understand how WiSARD works by doing the Scratch [interactive ludic] activities? Choose rate from 1 to 5, where 1 = \u0026quot;I cannot understand\u0026quot; and 5 = \u0026quot;I fully understand\u0026quot;. (Ordinal)\u003c/p\u003e\n\u003cp\u003eBar chart in Figure 10 shows that most of the research subjects could understand the programming challenges performed with BlockWiSARD. Regarding the Scratch ludic activities about the WiSARD model, the results indicate a good understanding of most participants. As shown in Figure 11, scores 4 and 5 represent m\u003ca href=\"#_ftn1\" name=\"_ftnref1\" title=\"\"\u003e\u003c/a\u003e\u003csup\u003e16\u003c/sup\u003eost of the scores used to evaluate the method.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eOnce the empirical evaluation was conducted remotely (Due to the COVID-19 pandemic), the IU utilized in the method evaluation did not include the AIcon2abs robotics materials. However, Alpha15 tests were performed with these materials, as shown in the links below\u003c/p\u003e\n\u003col class=\"decimal_type\"\u003e\n \u003cli\u003eA Raspberry Pi wheeled robot that can sense, learn, and act (https://youtu.be/KfEOnvSr_kU)\u003c/li\u003e\n \u003cli\u003eA wheeled robot (made from simple robotics) sensing, learning, and acting through a program developed with BlockWiSARD (https://youtu.be/GK6GHzuix10)\u003c/li\u003e\n \u003cli\u003eBlockWiSARD working alongside an Arduino board (https://youtu.be/xOLXF5LvziQ)\u003c/li\u003e\n\u003c/ol\u003e\n\u003cp\u003e4.1.2. \u003cstrong\u003eHypothesis testing (Inferential analysis)\u003c/strong\u003e\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eBeyond exploratory data analysis, \u003cstrong\u003enon-parametric hypothesis tests\u003c/strong\u003e (Siegel, 1957) were conducted on the responses to pre- and post-test questions Q03, Q04, Q07, and Q10. Two non-parametric tests were utilized: the \u003cstrong\u003eWilcoxon signed-rank test\u003c/strong\u003e and the \u003cstrong\u003eSign test\u003c/strong\u003e (Cohen et al., 2002; Kirk, 1978; Pett, 2015; Siegel, 1957). The hypothesis testing described in this section was designed to evaluate the following hypotheses:\u003c/p\u003e\n\u003cul\u003e\n \u003cli\u003e\u003cstrong\u003e\u003cem\u003eH0\u003c/em\u003e\u003c/strong\u003e (\u003cstrong\u003eNull Hypothesis\u003c/strong\u003e):\u0026nbsp;The AIcon2abs course had no effect on participants\u0026apos; understanding of the content covered regarding machine learning concepts and processes.\u003c/li\u003e\n \u003cli\u003e\u003cstrong\u003e\u003cem\u003eH1\u003c/em\u003e\u003c/strong\u003e\u003cem\u003e\u0026nbsp;\u003c/em\u003e(\u003cstrong\u003eAlternative hypothesis\u003c/strong\u003e):\u0026nbsp;The AIcon2abs course had a positive effect on participants\u0026apos; understanding of the content covered regarding machine learning concepts and processes.\u003c/li\u003e\n\u003c/ul\u003e\n\u003cp\u003eTo conduct these hypothesis tests, scores were assigned to each response category defined for the pre- and post-test questions Q03, Q04, Q07, and Q10 (see introductory paragraphs of Section 4). Table 3 presents the scoring scale applied to attribute scalar values to the four categorized questions related to machine learning concepts and processes addressed in the AIcon2abs course. All hypothesis testing procedures were carried out using IBM\u0026reg; SPSS\u0026reg; Statistics\u003csup\u003e\u0026nbsp;\u003c/sup\u003e\u003ca href=\"#_ftn1\" name=\"_ftnref1\" title=\"\"\u003e\u003c/a\u003e\u003csup\u003e17\u003c/sup\u003e.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eTable 3\u003c/strong\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003eCriteria adopted to score the categorized answers\u003c/p\u003e\n\u003cdiv align=\"\"\u003e\n \u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 76px;\"\u003e\n \u003cp\u003eScore\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 529px;\"\u003e\n \u003cp\u003eAnswer\u0026apos;s coherence level regarding the content covered in the course\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 76px;\"\u003e\n \u003cp\u003e5\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 529px;\"\u003e\n \u003cp\u003eAnswer consistent with the course content\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 76px;\"\u003e\n \u003cp\u003e4\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 529px;\"\u003e\n \u003cp\u003eAnswer relatively consistent with the course content\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 76px;\"\u003e\n \u003cp\u003e3\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 529px;\"\u003e\n \u003cp\u003eAnswer slightly consistent with the course content\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 76px;\"\u003e\n \u003cp\u003e2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 529px;\"\u003e\n \u003cp\u003eAnswer with little consistency in relation to the course content\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 76px;\"\u003e\n \u003cp\u003e1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 529px;\"\u003e\n \u003cp\u003eAnswer not consistent with the course content\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 76px;\"\u003e\n \u003cp\u003e0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 529px;\"\u003e\n \u003cp\u003eAnswer equal to \u0026ldquo;Do not know\u0026rdquo; or \u0026ldquo;Blank answer\u0026rdquo;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n \u003c/table\u003e\n\u003c/div\u003e\n\u003cp\u003e\u003cstrong\u003eHypothesis test part 1\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eFor the Wilcoxon signed-rank test to yield robust and reliable results, a primary requirement is that the distribution of the differences between the two dependent samples should be approximately symmetrical. This condition was met between the two dependent samples when considering the total score across all four questions for all 34 research participants. Table 4 and Table 5 display the results of the Wilcoxon signed-rank test and the Sign test for questions Q03, Q04, Q07, and Q10. As indicated, both tests produced identical results.\u003c/p\u003e\n\u003cp\u003eTable 4 Wilcoxon signed-rank test results for Q03+Q04+Q07+Q10\u003c/p\u003e\n\u003ctable border=\"0\" cellspacing=\"0\" cellpadding=\"0\" width=\"605\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 142px;\"\u003e\n \u003cp\u003eScores\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 76px;\"\u003e\n \u003cp\u003eSignificance\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 76px;\"\u003e\n \u003cp\u003eDecision\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 71px;\"\u003e\n \u003cp\u003eN\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 80px;\"\u003e\n \u003cp\u003e(+) dif\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 80px;\"\u003e\n \u003cp\u003e(-) dif\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 80px;\"\u003e\n \u003cp\u003eTais\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 142px;\"\u003e\n \u003cp\u003eQ03 + Q04 + Q07 + Q10\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 76px;\"\u003e\n \u003cp\u003e\u0026lt;,001.\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 76px;\"\u003e\n \u003cp\u003e\u003cem\u003eReject H0\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 71px;\"\u003e\n \u003cp\u003e34\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 80px;\"\u003e\n \u003cp\u003e28\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 80px;\"\u003e\n \u003cp\u003e2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 80px;\"\u003e\n \u003cp\u003e4\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003eTable 5\u0026nbsp;Sign test results for Q03+Q04+Q07+Q10\u003c/p\u003e\n\u003ctable border=\"0\" cellspacing=\"0\" cellpadding=\"0\" width=\"605\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 142px;\"\u003e\n \u003cp\u003eScores\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 76px;\"\u003e\n \u003cp\u003eSignificance\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 76px;\"\u003e\n \u003cp\u003eDecision\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 71px;\"\u003e\n \u003cp\u003eN\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 80px;\"\u003e\n \u003cp\u003e(+) dif\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 80px;\"\u003e\n \u003cp\u003e(-) dif\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 80px;\"\u003e\n \u003cp\u003eTais\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 142px;\"\u003e\n \u003cp\u003eQ03 + Q04 + Q07 + Q10\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 76px;\"\u003e\n \u003cp\u003e\u0026lt;,001.\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 76px;\"\u003e\n \u003cp\u003e\u003cem\u003eReject H0\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 71px;\"\u003e\n \u003cp\u003e34\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 80px;\"\u003e\n \u003cp\u003e28\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 80px;\"\u003e\n \u003cp\u003e2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 80px;\"\u003e\n \u003cp\u003e4\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003ch4\u003eHypothesis test part 2\u003c/h4\u003e\n\u003cp\u003eThe Wilcoxon signed-rank test assumption of symmetry was not met when analyzing the scores for each of the four questions individually (even without dividing the data by age groups). Consequently, only the Sign test was applied for these hypothesis tests. Table 6 presents the Sing tests results.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eTable 6\u003c/strong\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003eSign test results for Q03, Q04, Q07, and Q10 (separately) regarding the pre-posttest scores of the 34 research subjects\u003c/p\u003e\n\u003ctable border=\"0\" cellspacing=\"0\" cellpadding=\"0\" width=\"605\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 113px;\"\u003e\n \u003cp\u003eScores\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 82px;\"\u003e\n \u003cp\u003eSignificance\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 82px;\"\u003e\n \u003cp\u003eDecision\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 82px;\"\u003e\n \u003cp\u003eN\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 82px;\"\u003e\n \u003cp\u003e(+) dif\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 82px;\"\u003e\n \u003cp\u003e(-) dif\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 82px;\"\u003e\n \u003cp\u003eTais\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 113px;\"\u003e\n \u003cp\u003eQ03\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 82px;\"\u003e\n \u003cp\u003e,002\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 82px;\"\u003e\n \u003cp\u003e\u003cem\u003eReject H0\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 82px;\"\u003e\n \u003cp\u003e34\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 82px;\"\u003e\n \u003cp\u003e20\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 82px;\"\u003e\n \u003cp\u003e4\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 82px;\"\u003e\n \u003cp\u003e10\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 113px;\"\u003e\n \u003cp\u003eQ04\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 82px;\"\u003e\n \u003cp\u003e\u0026lt;,001\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 82px;\"\u003e\n \u003cp\u003e\u003cem\u003eReject H0\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 82px;\"\u003e\n \u003cp\u003e34\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 82px;\"\u003e\n \u003cp\u003e20\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 82px;\"\u003e\n \u003cp\u003e3\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 82px;\"\u003e\n \u003cp\u003e11\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 113px;\"\u003e\n \u003cp\u003eQ07\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 82px;\"\u003e\n \u003cp\u003e\u0026lt;,001\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 82px;\"\u003e\n \u003cp\u003e\u003cem\u003eReject H0\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 82px;\"\u003e\n \u003cp\u003e34\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 82px;\"\u003e\n \u003cp\u003e19\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 82px;\"\u003e\n \u003cp\u003e2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 82px;\"\u003e\n \u003cp\u003e13\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 113px;\"\u003e\n \u003cp\u003eQ10\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 82px;\"\u003e\n \u003cp\u003e,002\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 82px;\"\u003e\n \u003cp\u003e\u003cem\u003eReject H0\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 82px;\"\u003e\n \u003cp\u003e34\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 82px;\"\u003e\n \u003cp\u003e22\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 82px;\"\u003e\n \u003cp\u003e5\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 82px;\"\u003e\n \u003cp\u003e7\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003eThe p-value for statistical significance was set at 0.05 when defining test parameters in IBM\u0026reg; SPSS\u0026reg; Statistics. A p-value of 0.05 is a standard threshold in hypothesis testing, indicating a 5% probability that the observed results could occur purely by chance. However, in most cases, the statistical significance calculated by the tests yielded a p-value below 0.001. This level of significance implies that the likelihood of the observed results occurring by chance is approximately 1 in 1,000, or 0.1%.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eThe results obtained from the hypothesis tests (leading to a rejection of H0) support the positive effect of the AIcon2abs course on the research participants.\u003c/strong\u003e In addition to statistical significance, the marked prevalence of positive post-test outcomes compared to pre-test results is also noteworthy.\u003c/p\u003e\n\u003cp\u003e4.2. Data Analysis from a phenomenological perspective\u003c/p\u003e\n\u003cp\u003eThis study also employed a phenomenological analysis approach. As outlined in Section 2, we adopted the phenomenological methodology proposed by Creswell (2016), which involves describing the phenomenon experienced by research subjects and the context in which this experience unfolded. Additionally, in a few comprehensive paragraphs, the researcher presents a portrayal of the essence of participants\u0026apos; perceptions of their lived experience. This narrative is constructed by reducing and organizing participants\u0026apos; testimonies into significant quotations classified by thematic categories.\u003c/p\u003e\n\u003cp\u003e4.2.1. The lived experience and its context\u003c/p\u003e\n\u003cp\u003eSection 3 detailed the context in which participants experienced the observed phenomenon: an online course conducted as part of research on a method intended to provide insight into the inner workings of machine learning and AI. Participation was voluntary, and motivations varied, including (i) curiosity and interest in the topics covered (computer programming, AI, and machine learning), (ii) the desire to expand knowledge, and (iii) the opportunity to contribute to scientific research.\u003c/p\u003e\n\u003cp\u003eIn Subsection 4.2.2, titled \u003cem\u003eThe AIcon2abs Course\u003c/em\u003e, we present a narrative capturing the essence of participants\u0026apos; experiences. This narrative synthesizes perceptions derived from responses to post-test questions 23, 25, and 26 (Table 1), alongside additional comments on the learning platform. To develop the narrative, we organized participants\u0026apos; responses into four categories: (i) Feelings, (ii) Didactics and Content, (iii) Learning, and (iv) Reflections. The thematic categorization and original statements in Portuguese are available in AIcon2abs (2023).\u003c/p\u003e\n\u003cp\u003e4.2.2. The\u0026nbsp;AIcon2abs\u0026nbsp;course\u003c/p\u003e\n\u003cp\u003eThe following narrative serves as a composite testimonial reflecting the common feelings, perceptions, and reflections of the 34 participants regarding the AIcon2abs course:\u003c/p\u003e\n\u003cp\u003e\u003cem\u003e\u0026ldquo;I chose to enroll in the AIcon2abs course out of curiosity, hoping to learn more about artificial intelligence and machine learning. The course was offered online, featuring pre-recorded lessons, and began with the story of Zrow, a robot from another planet on a quest to learn about Earth. The narrative of Zrow\u0026rsquo;s adventures provided a stimulating framework for the course, evoking the dreams of my childhood. This method of presenting the course content was highly engaging. The course materials were of exceptional quality, both technically and pedagogically. The content was intuitive and well-explained, requiring no prior knowledge. This shows that complex topics, like AI and machine learning, can be made accessible to a broad audience. The themes were explored through interactive, simple, and creative activities, including Scratch-based games and animations, which were very engaging. The course was designed to accommodate various age groups, with thoughtful consideration for older adults and children alike.\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eThroughout the course, I gained insights into programming, machine learning, and AI. I learned that machines acquire knowledge through examples and that they operate based on programmed instructions. I also learned about WiSARD, which was surprisingly easy to understand. I gained a sense of the processes that allow machines to \u0026lsquo;learn\u0026rsquo; and discovered different methods for training AI. This course opened a new world to me, a realm in which I had been an outsider. Some activities, however, could benefit from additional time and examples. I also believe that face-to-face interaction could enhance learning, particularly for younger participants who may find even brief video content tiring.\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eThe course prompted me to reflect on the need for greater awareness of AI. The general public should understand AI to participate in discussions about its future. This is knowledge that should permeate our schools and academic institutions. The course raised several questions: What applications of AI influence our daily lives without our awareness? Will most people remain uninformed, leaving knowledge holders to shape decisions for us? I reflected on the implications of AI for free will and its potential for transforming our understanding of autonomy. This experience broadened my understanding of the world and affirmed the remarkable capacity of the human mind. Ultimately, I found the course to be a truly rewarding experience.\u0026rdquo;\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003e4.3. Discussion\u003c/p\u003e\n\u003cp\u003eThis section discusses the results through the lens of the DSR method as outlined by Dresch et al. (2015) and Pimentel et al. (2019), which guided this research (see Figure 1). We will first evaluate the three primary assessments expected in DSR-based research per Pimentel et al. (2019). Next, \u003cstrong\u003ewe will formalize the construction and contingency heuristics\u003c/strong\u003e, the core contributions of the DSR method as proposed by Dresch et al. (2015). Finally, we will address some limitations of this study.\u003c/p\u003e\n\u003cp\u003e4.3.1. \u003cstrong\u003eIs the artifact valid?\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eResults in Subsections 4.1 and 4.2 indicate that (i) the research objectives (Subsection 1.2) were satisfactorily achieved, (ii) the requirements (Subsection 1.3) were met, and (iii) the anticipated performance (Subsection 1.5) was realized. \u003cstrong\u003eThese outcomes support the conclusion that the AIcon2abs method is valid\u003c/strong\u003e, meeting the requirements of Evaluation 1 in the DSR map presented in Section 2.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e4.3.2. \u003cstrong\u003eWas the problem solved?\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe data indicate that most participants found the AIcon2abs course activities easy and that many developed a foundational understanding of machine learning in its most basic form. Furthermore, participants\u0026rsquo; satisfaction with the course suggests that learning through AIcon2abs was a rewarding experience. According to Simon (1996), consensus among participants serves as a criterion for determining whether a solution is satisfactory. \u003cstrong\u003eThe data presented here show that participants were generally satisfied, indicating that AIcon2abs is a suitable method for addressing its intended problem\u003c/strong\u003e.\u003c/p\u003e\n\u003cp\u003e4.3.3. \u003cstrong\u003eDo the theoretical conjectures seem to be valid?\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eIn Subsection 1.4, we outlined six theoretical conjectures guiding the development of the AIcon2abs method. The findings suggest the validity of these conjectures, \u003cstrong\u003eparticularly the central one, which posits that \u003cem\u003elearning abstract concepts from computer science can be facilitated through practices that highlight observable aspects of these concepts in the concrete world\u003c/em\u003e.\u0026nbsp;\u003c/strong\u003eIt is valid to mention that none of the research or approaches found in the Queiroz et al. (2021) literature review had the same target audience as the AIcon2abs method: the general public. Those studies were exclusively dedicated to a specific age group: (i) children, (ii) adolescents, or (iii) undergraduate course students. Therefore, \u003cstrong\u003eit is worth stating that this study revealed that this theoretical conjecture seems valid for a diverse audience of different ages and backgrounds (including children)\u003c/strong\u003e.\u003c/p\u003e\n\u003cp\u003eRegarding conjecture 5 (Subsection 1.4), after implementing the AIcon2abs course, we realized that the agent-based approach could be enhanced by integrating it with the \u003cem\u003eInput-Process-Output\u003c/em\u003e (IPO) model\u003ca href=\"#_ftn2\" name=\"_ftnref2\" title=\"\"\u003e\u003c/a\u003e\u003csup\u003e18\u003c/sup\u003e. This combination could further facilitate understanding the fundamental logic of machine learning in its most basic form.\u003c/p\u003e\n\u003cp\u003e4.3.4. \u003cstrong\u003eAIcon2abs\u0026nbsp;\u003c/strong\u003e\u003cstrong\u003econstruction heuristics\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eConstruction heuristics\u003c/em\u003e establish rules for the internal environment of the artifact and for its construction. When formalized, these heuristics can aid other researchers in developing artifacts to address similar issues. Construction heuristics represent a core contribution to knowledge within the DSR framework (Dresch et al., 2015; Gregor \u0026amp; Jones, 2007; Simon, 1996; Venable, 2006). Fig.12 presents the formalized construction heuristic derived from the AIcon2abs method\u0026rsquo;s design, development, implementation, and evaluation.\u003c/p\u003e\n\u003cp\u003e4.3.5. \u003cstrong\u003eAIcon2abs\u003c/strong\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003cstrong\u003econtingency heuristics\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eContingency heuristics specify, when applicable, the adjustments required to adapt an artifact for effective functioning in different external environments\u003c/strong\u003e (Dresch et al., 2015). For the AIcon2abs method, the external environment is defined as its target audience. During data analysis, some participants highlighted certain aspects of the AIcon2abs method that made it difficult to understand some specific content. Notably, these challenges were exclusively reported by adult participants. The most frequently cited issues included:\u003c/p\u003e\n\u003cul type=\"disc\"\u003e\n \u003cli\u003eSome content was perceived as overly \u0026quot;complex,\u0026quot; requiring more than a single engagement with the related activity for adequate comprehension.\u003c/li\u003e\n \u003cli\u003eActivities associated with more complex content had insufficient examples to support understanding.\u003c/li\u003e\n \u003cli\u003eThe recommended course duration was inadequate for consolidating certain content.\u003c/li\u003e\n\u003c/ul\u003e\n\u003cp\u003eThese findings represent contingencies that should be addressed to meet the learning objectives of the AIcon2abs method, particularly when applied to an audience with the specified learning needs. The identified contingencies focus on the difficulty of adequately understanding some themes worked on in the AIcon2abs IU. In alignment with the \u003cem\u003eDifferentiated Learning educational approach\u003c/em\u003e (Tomlinson, 1999), we propose the contingency heuristic illustrated in Figure 13. This heuristic is founded on established theoretical principles and pedagogical practices. However, a new Design Science Research (DSR) cycle, encompassing steps from Artifact Design to Evaluation (see Figure 1), is recommended to validate its efficacy.\u003c/p\u003e\n\u003cp\u003e4.3.6. \u003cstrong\u003eLimitations of the study\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eWhile efforts were made to address threats to the validity of this research, some limitations remain. Regarding \u003cstrong\u003eexternal validity\u003c/strong\u003e, despite the promising results observed for the AIcon2abs method within the context of this study, its effectiveness in other settings is not assured. The method was evaluated based on a relatively small sample of 34 participants. While the sample exhibited diversity in certain respects, it was specific in others: (i) all adults had either completed or were attending higher education, and (ii) all children and adolescents had access to quality education. Concerning \u003cstrong\u003einternal validity\u003c/strong\u003e, data were not collected from participants who withdrew from the AIcon2abs course, precluding comparisons between those who completed the course and those who did not. Furthermore, all participants were presumed to be prepared to engage with the course content, potentially influencing the outcomes. These considerations highlight the need for further evaluations of the AIcon2abs method across diverse contexts. Suggestions for future research and potential directions are presented in Subsection 5.2.\u003c/p\u003e"},{"header":"5. Conclusion","content":"\u003cp\u003e5.1. Concluding summary\u003c/p\u003e\n\u003cp\u003eThis study aimed to address the question:\u0026nbsp;\u003cstrong\u003e\u003cem\u003eIs\u0026nbsp;\u003c/em\u003e\u003c/strong\u003e\u003cstrong\u003e\u003cem\u003eAIcon2abs\u0026nbsp;\u003c/em\u003e\u003c/strong\u003e\u003cstrong\u003e\u003cem\u003ea suitable method to support a diverse audience of different ages and backgrounds in understanding the concept of machine learning in its most basic form in an uncomplicated and satisfactory way\u003c/em\u003e\u003c/strong\u003e\u003cem\u003e?\u003c/em\u003e The results presented here affirmatively answer this question. Moreover, they demonstrate that the AIcon2abs method effectively achieved its objectives by incorporating the following four strategies, which constitute the primary contribution of this research:\u003c/p\u003e\n\u003cul class=\"decimal_type\"\u003e\n \u003cli\u003eEmploying Active Learning strategies.\u003c/li\u003e\n \u003cli\u003ePresenting the fundamental concept of machine learning (in its most basic form) through concrete examples that are familiar to the target audience and aligned with the adopted machine learning model constructs.\u003c/li\u003e\n \u003cli\u003eIntegrating the learning algorithm instantiation, along with training and classification primitives, as basic commands within a conventional programming structure.\u003c/li\u003e\n \u003cli\u003eUtilizing a machine learning algorithm whose training and classification processes are readily observable and replicable, allowing learners to observe the internal actions of the machine during the learning process.\u003c/li\u003e\n\u003c/ul\u003e\n\u003cp\u003eAdditionally,\u0026nbsp;\u003cstrong\u003ethe formalization of construction and contingency heuristics\u003c/strong\u003e for the AIcon2abs method (outlined in Subsections 4.3.4 and 4.3.5) constitutes a valuable contribution of this study. Other notable contributions include: (i) the\u0026nbsp;\u003cstrong\u003ePre-Posttest Matrix data visualization tool\u003c/strong\u003e, an adapted use and interpretation of the Confusion Matrix (Section 4), and (ii)\u0026nbsp;\u003cstrong\u003ethe blending of the Design Science Research (DSR) method by Dresch et al. (2015) with elements from the DSR map proposed by Pimentel et al. (2019)\u003c/strong\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e(Section 2).\u003c/p\u003e\n\u003cp\u003eThe AIcon2abs components were highly rated by nearly all participants\u0026mdash;children, adolescents, and adults\u0026mdash;who took part in the empirical evaluation. Having been rigorously evaluated and validated\u003cstrong\u003e,\u0026nbsp;\u003c/strong\u003e\u003cstrong\u003ethe AIcon2abs method is now available to those interested in demystifying artificial intelligence, particularly machine learning,\u0026nbsp;\u003c/strong\u003e\u003cstrong\u003eto the general public.\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e5.2. Future work\u003c/p\u003e\n\u003cp\u003eTo further enhance the AIcon2abs method and broaden its applicability to include new audiences, the following future work would be of significant value:\u003c/p\u003e\n\u003cul\u003e\n \u003cli\u003eDevelop English versions of the instructional videos and playful activities using the WiSARD model.\u003c/li\u003e\n \u003cli\u003eIncorporate the Input-Process-Output (IPO) concept alongside the Intelligent Agents concept in the design of the AIcon2abs instructional unit, to reinforce understanding of the fundamental differences between traditional computer systems and those utilizing AI mechanisms.\u003c/li\u003e\n \u003cli\u003eProduce the additional learning materials proposed in the formalized contingency heuristics.\u003c/li\u003e\n \u003cli\u003eOffer a face-to-face version of the AIcon2abs course to: (i) apply the contingency heuristic developed in this study and (ii) incorporate educational robotics materials in the BlockWiSARD activities.\u003c/li\u003e\n \u003cli\u003eAdapt the AIcon2abs method for illiterate learners.\u003c/li\u003e\n \u003cli\u003eInvestigate the feasibility of adapting the AIcon2abs method for individuals with disabilities, such as the visually impaired and the deaf.\u003c/li\u003e\n \u003cli\u003eExplore the potential of using AIcon2abs construction heuristics to develop artifacts introducing other AI paradigms to the general public.\u003c/li\u003e\n \u003cli\u003eApply the AIcon2abs construction heuristics to design materials addressing other complex computing concepts, such as Object-Oriented Programming Paradigm.\u003c/li\u003e\n\u003c/ul\u003e"},{"header":"Declarations","content":"\u003cp\u003eEthics statement\u003c/p\u003e\n\u003cp\u003eThis research has been approved by the CEP/HUCFF/FM/UFRJ\u003ca href=\"#_ftn1\" name=\"_ftnref1\" title=\"\"\u003e\u003c/a\u003e\u003csup\u003e19\u003c/sup\u003e human research ethics committee: CAAE: 44444621.7.0000.5257, approval date: June 22, 2021, approval order number: 4.861.208, 2021. All adults, parents, or legal guardians gave their informed consent, and all minors gave their informed assent before participating in this study.\u003c/p\u003e\n\u003cp\u003eConflict of interest statement\u003c/p\u003e\n\u003cp\u003eThe authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.\u003c/p\u003e\n\u003cp\u003eData availability\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eThe datasets used for the data analysis presented in this paper are available at\u0026nbsp;AIcon2abs\u0026nbsp;(2023).\u003c/p\u003e\n\u003cp\u003eAcknowledgements\u003c/p\u003e\n\u003cp\u003eThis study was financed in part by the Coordena\u0026ccedil;\u0026atilde;o de Aperfei\u0026ccedil;oamento de Pessoal de N\u0026iacute;vel Superior-Brasil (CAPES)-Finance Code 001.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eFunding\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eCoordena\u0026ccedil;\u0026atilde;o de Aperfei\u0026ccedil;oamento de Pessoal de N\u0026iacute;vel Superior-Brasil (CAPES)- Finance Code 001. (\u003cem\u003eGraduate Fellowship\u003c/em\u003e).\u003c/p\u003e\u003ch2\u003eAuthor Contribution\u003c/h2\u003e\u003cp\u003eAuthors\u0026rsquo; ContributionsQUEIROZ, R.L.: Conceptualization, Data curation, Formal analysis, Investigation, Methodology, Project administration, Software, Validation, Visualization, Writing - original draft, review \u0026amp; editing. LIMA, P.M.V.: Conceptualization, Formal analysis, Funding acquisition, Investigation, Methodology, Project administration, Resources, Supervision, Validation, Visualization, and Writing, review \u0026amp; editing. LIMA, C.: Formal analysis, Project administration, Supervision, Validation, Writing - review \u0026amp; editing. SAMPAIO, F.F.: Formal analysis, Project administration, Supervision, Validation, Writing - review \u0026amp; editing.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n\u003cli\u003eAIcon2abs (2023). AIcon2abs Empirical Evaluation Data. \u003cem\u003eMendeley Data\u003c/em\u003e, V1. https://doi.org/10.17632/ch6ys3gz82.1\u003c/li\u003e\n\u003cli\u003eAleksander, I., De Gregorio, M., Fran\u0026ccedil;a, F.M.G., Lima, P.M.V., \u0026amp; Morton, H, (2009), A brief introduction to weightless neural systems. \u003cem\u003eProceedings of ESANN 2009\u003c/em\u003e, 299\u0026ndash;305. \u003c/li\u003e\n\u003cli\u003eAlmeida, V. (2018, August 8). \u003cem\u003eBrasil precisa de estrat\u0026eacute;gia para intelig\u0026ecirc;ncia artificial\u003c/em\u003e (Transl.: Brazil needs a strategy for artificial intelligence). Academia Brasileira de Ci\u0026ecirc;ncias. Retrieved January 28, 2024, from http://www.abc.org.br/2018/08/08/brasil-precisa-de-estrategia-para-inteligencia-artificial/\u003c/li\u003e\n\u003cli\u003eBanzi, M. (2012, June 27). \u003cem\u003eHow Arduino is open-sourcing imagination\u003c/em\u003e. TEDGlobal Edinburgh. Retrieved January 28, 2024, from https://www.ted.com/talks/ massimo_banzi_how_arduino_is_open_sourcing_imagination\u003cu\u003e. \u003c/u\u003e\u003c/li\u003e\n\u003cli\u003eBonwell, C.C., \u0026amp; Eison, J.A. (1991). \u003cem\u003eActive learning: Creating excitement in the classroom\u003c/em\u003e. ASHE-ERIC higher education reports. George Washington University. Retrieved January 28, 2024, from https://files.eric.ed.gov/fulltext/ ED336049.pdf. \u003c/li\u003e\n\u003cli\u003eBranch, R.M. (2009). \u003cem\u003eInstructional design: The ADDIE approach\u003c/em\u003e. Springer.\u003c/li\u003e\n\u003cli\u003eCaltech (1988). \u003cem\u003eRichard Feynman\u0026rsquo;s Blackboard at the Time of His Death. \u003c/em\u003eCaltech Archives. Retrieved January 28, 2024, from http://archives-dc.library.caltech.edu/ islandora/object/ct1%3A483 \u003c/li\u003e\n\u003cli\u003eCaputo, M. (2019, December 21) \u003cem\u003eMCTIC lan\u0026ccedil;a consulta p\u0026uacute;blica para a Estrat\u0026eacute;gia Brasileira de Intelig\u0026ecirc;ncia Artificial \u003c/em\u003e(Transl.: MCTIC launches public consultation for the Brazilian Artificial Intelligence Strategy). Academia Brasileira de Ci\u0026ecirc;ncias. Retrieved January 28, 2024, from http://www.abc.org.br/2019/12/21/mctic-lanca-consulta-publica-para-a-estrategia-brasileira-de-inteligencia-artificial/\u003c/li\u003e\n\u003cli\u003eCohen, L., Manion, L., \u0026amp; Morrison, K. (2002). \u003cem\u003eResearch methods in education\u003c/em\u003e. Routledge.\u003c/li\u003e\n\u003cli\u003eCreswell, J.W., \u0026amp; Poth, C.N. (2016).\u003cem\u003e Qualitative inquiry and research design: Choosing among five approaches\u003c/em\u003e. Sage publications.\u003c/li\u003e\n\u003cli\u003eDick, W., \u0026amp; Carey, L.M. (1977). Formative Evaluation. In: L.J. Briggs (Ed.), \u003cem\u003eInstructional design: Principles and applications\u003c/em\u003e. (pp. 311-333). Educational Technology Publications.\u003c/li\u003e\n\u003cli\u003eDash, A. (2023).The LLM Advantage: How to Unlock the Power of Language Models for Business Success: \u003cem\u003eA Practical Guide to Leveraging AI for Innovation, Growth, and Competitive Edge\u003c/em\u003e. United Kindon. Grazing Minds Publishing, \u003c/li\u003e\n\u003cli\u003eDresch, A., Lacerda, D.P., \u0026amp; Antunes Jr, J.A.V. (2015). \u003cem\u003eDesign science research\u003c/em\u003e. Springer. Retrieved January 28, 2024, from http://www.springer.com/gb/book/ 9783319073736\u003cem\u003e\u003cu\u003e. \u003c/u\u003e\u003c/em\u003e\u003c/li\u003e\n\u003cli\u003eEguchi, A. (2010). What is educational robotics? Theories behind it and practical implementation. \u003cem\u003eProceedings of the Society for Information Technology \u0026amp; Teacher Education International Conference\u003c/em\u003e, 4006\u0026ndash;4014.\u003c/li\u003e\n\u003cli\u003eFeynman, R.P. (1989). Feynman\u0026rsquo;s office: the last blackboards. \u003cem\u003ePhysics Today\u003c/em\u003e, 42 (2), 88. \u003cem\u003e\u003cu\u003ehttps://doi.org/10.1063/1.2810904\u003c/u\u003e\u003c/em\u003e\u003c/li\u003e\n\u003cli\u003eFleming, N.D. (1995). I\u0026apos;m different; not dumb. Modes of presentation (VARK) in the tertiary classroom. \u003cem\u003eProceedings of the 1995 Annual Conference of the Higher Education Research and Development Society of Australasia (HERDSA)\u003c/em\u003e, 308\u0026ndash;313.\u003c/li\u003e\n\u003cli\u003eGregor, S., \u0026amp; Jones, D. (2007). The anatomy of a design theory. \u003cem\u003eJournal of the Association for Information Systems,\u003c/em\u003e 8(5), 312-335. \u003cem\u003ehttps://doi.org/10.17705/1jais.00129\u003c/em\u003e\u003c/li\u003e\n\u003cli\u003eKalota, F. 2024. A Primer on Generative Artificial Intelligence. \u003cem\u003eEducation Sciences\u003c/em\u003e. 14 (2), 172. https://doi.org/10.3390/educsci1402017\u003c/li\u003e\n\u003cli\u003eKaner, C., Bach, J., \u0026amp; Pettichord, B. (2001). \u003cem\u003eLessons learned in software testing\u003c/em\u003e. John Wiley \u0026amp; Sons.\u003c/li\u003e\n\u003cli\u003eKirk, R.E. (1978). \u003cem\u003eIntroductory statistics\u003c/em\u003e. Brooks/Cole Publishing Company.\u003c/li\u003e\n\u003cli\u003eMaharaj, S., Koon, M.K., Gookool, C., Taylor, M., Birbal, R., \u0026amp; Mohammed, N. (2023). \u003cem\u003eCambridge Primary Computing Learner\u0026apos;s Book - Stage 5\u003c/em\u003e. Hodder Education.\u003c/li\u003e\n\u003cli\u003eMedina, E.M. (2004). Beyond the ballot box: Computer science education and social responsibility. \u003cem\u003eACM SIGCSE Bulletin\u003c/em\u003e, 36(4), 7\u0026ndash;10. https://doi.org/10.1145/1041624.1041626\u003c/li\u003e\n\u003cli\u003eMitchell, T.M. (1997). \u003cem\u003eMachine intelligence\u003c/em\u003e. McGraw-Hill.\u003c/li\u003e\n\u003cli\u003eMitchell, T.M. (2006). \u003cem\u003eThe discipline of machine learning\u003c/em\u003e. Carnegie Mellon University, School of Computer Science, Machine Learning Department.\u003c/li\u003e\n\u003cli\u003eOgawa, R.T., \u0026amp; Malen, B. (1991). Towards rigor in reviews of multivocal literatures: Applying the exploratory case study method. \u003cem\u003eReview of Educational Research\u003c/em\u003e, 61(3), 265\u0026ndash;286. https://doi.org/10.3102/ 00346543061003265\u003c/li\u003e\n\u003cli\u003ePapert, S. (1994). \u003cem\u003eThe Children\u0026rsquo;s Machine: Rethinking School in the Age of the Computer\u003c/em\u003e. Basic Books.\u003c/li\u003e\n\u003cli\u003ePapert, S.A. (1993). \u003cem\u003eMindstorms: Children, Computers, and Powerful Ideas\u003c/em\u003e. Basic Books.\u003c/li\u003e\n\u003cli\u003ePearson, K. (1904). \u003cem\u003eMathematical contributions to the theory of evolution, XIII: On the theory of contingency and its relation to association and normal correlation\u003c/em\u003e. Draper\u0026rsquo;s Co. Research memoirs in biometry series, 1. Reprinted in Pearson, E.A. (Ed). (1948). \u003cem\u003eKarl Pearson\u0026rsquo;s Early Statistical Papers\u003c/em\u003e, Cambridge University Press. \u003c/li\u003e\n\u003cli\u003ePett, M. A. (2015). \u003cem\u003eNonparametric statistics for health care research: Statistics for small samples and unusual distributions.\u003c/em\u003e Sage Publications.\u003c/li\u003e\n\u003cli\u003ePiaget, J. (1950). \u003cem\u003eThe psychology of intelligence\u003c/em\u003e. London Routledge \u0026amp; Kegan Paul.\u003c/li\u003e\n\u003cli\u003ePimentel, M., Filippo, D., \u0026amp; Santoro, F.M. (2019). \u003cem\u003eDesign Science Research: fazendo pesquisas cient\u0026iacute;ficas rigorosas atreladas ao desenvolvimento de artefatos computacionais projetados para a educa\u0026ccedil;\u0026atilde;o \u003c/em\u003e(Transl.: Design Science Research: carrying out rigorous scientific research linked to developing computational artifacts designed for education). Metodologia de Pesquisa Em Inform\u0026aacute;tica Na Educa\u0026ccedil;\u0026atilde;o: Concep\u0026ccedil;\u0026atilde;o Da Pesquisa. SBC. Retrieved January 28, 2024, from https://metodologia.ceie-br.org/wp-content/uploads/2018/10/cap1_5.pdf\u003c/li\u003e\n\u003cli\u003eQueiroz, R.L. (2023). AI from concrete to abstract: demystifying artificial intelligence to the general public. \u003cem\u003eFederal University of Rio de Janeiro \u003c/em\u003e(UFRJ). Retrieved January 28, 2024, from https://bit.ly/4cYUuXK \u003c/li\u003e\n\u003cli\u003eQueiroz, R.L., Sampaio, F.F., Lima, C., \u0026amp; Lima, P.M.V. (2021). AI from concrete to abstract: demystifying artificial intelligence to the general public. \u003cem\u003eAI \u0026amp; SOCIETY,\u003c/em\u003e 36(3), 877-893. https://doi.org/10.1007/s00146-021-01151-x\u003c/li\u003e\n\u003cli\u003eResnick, M., Maloney, J., Monroy-Hern\u0026aacute;ndez, A., Rusk, N., Eastmond, E., Brennan, K., Millner, A., Rosenbaum, E., Silver, J., \u0026amp; Silverman, B. (2009). Scratch: programming for all. \u003cem\u003eCommunications of the ACM\u003c/em\u003e, 52(11), 60\u0026ndash;67.\u003cem\u003e\u003cu\u003e \u003c/u\u003e\u003c/em\u003ehttps://doi.org/10.1145/1592761.1592779\u003c/li\u003e\n\u003cli\u003eRussell, S.J., \u0026amp; Norvig, P. (2020). \u003cem\u003eArtificial intelligence: a modern approach. \u003c/em\u003ePrentice Hall.\u003c/li\u003e\n\u003cli\u003eSakulkueakulsuk, B., Witoon, S., Ngarmkajornwiwat, P., Pataranutaporn. P., Surareungchai, W., Pataranutaporn, P., \u0026amp; Subsoontorn, P. (2018). Kids making AI: Integrating machine learning, gamification, and social context in STEM education. \u003cem\u003eProceedings of the 2018 IEEE International Conference on Teaching, Assessment, and Learning for Engineering (TALE),\u003c/em\u003e 1005\u0026ndash;1010. https://doi.org/10.1109/ TALE.2018.8615249\u003c/li\u003e\n\u003cli\u003eSchlegel, M.J. (1995). A Handbook of Instructional and Training Program Design. ERIC - Institute of Education Sciences. Retrieved January 28, 2024, from \u003cu\u003ehttps://files.eric.ed.gov/fulltext/ED383281.pdf\u003c/u\u003e\u003c/li\u003e\n\u003cli\u003eSeverance, C. (2013). Eben Upton: Raspberry Pi. \u003cem\u003eComputer,\u003c/em\u003e 46(10), 14-16. https://doi.org/10.1109/MC.2013.349\u003c/li\u003e\n\u003cli\u003eSiegel, S. (1957). Nonparametric statistics. \u003cem\u003eThe American Statistician\u003c/em\u003e, 11(3), 13\u0026ndash;19. https://doi.org/10.1080/00031305 .1957.10501091\u003c/li\u003e\n\u003cli\u003eSimon, H.A. (1996). \u003cem\u003eThe sciences of the artificial\u003c/em\u003e. MIT press.\u003c/li\u003e\n\u003cli\u003eTessmer, M. (2013). \u003cem\u003ePlanning and conducting formative evaluations\u003c/em\u003e. Routledge.\u003c/li\u003e\n\u003cli\u003eTomlinson, C.A. (1999). Mapping a route toward differentiated instruction. \u003cem\u003eEducational leadership\u003c/em\u003e, 57(1), 12-17. https://education.illinoisstate.edu/downloads/linc/linccurriculummodule/Tomlinson.pdf. Accessed 28 January 2024\u003c/li\u003e\n\u003cli\u003eVan Manen, M. (1990). Beyond assumptions: Shifting the limits of action research. \u003cem\u003eTheory into Practice\u003c/em\u003e, 29 (3), 152\u0026ndash;157. \u003cu\u003ehttps://doi.org/10.1080/00405849009543448\u003c/u\u003e\u003c/li\u003e\n\u003cli\u003eVenable, J. (2006). The role of theory and theorizing in design science research. \u003cem\u003eProceedings of the 1st International Conference on Design Science in Information Systems and Technology (DESRIST 2006),\u003c/em\u003e 1\u0026ndash;18.\u003c/li\u003e\n\u003cli\u003eYogeshwar, R. (2018, October 18). \u003cem\u003eOs dados que voc\u0026ecirc; entrega acabar\u0026atilde;o por estrangul\u0026aacute;-lo no sistema\u003c/em\u003e (Transl.: The data you deliver will eventually strangle you in the system). El Pa\u0026iacute;s. Retrieved January 28, 2024, from\u003c/li\u003e\n\u003c/ol\u003e"},{"header":"Footnotes","content":"\u003col\u003e\n\u003cli\u003eSimply put, generative AI generates content, which could be text, images, or multimedia.\u003c/li\u003e\n\u003cli\u003e\u003csup\u003e \u003c/sup\u003eChat GPT is an AI Large Language Model (LLM) developed by Open AI to follow instructions in a prompt and provide detailed responses. https://openai.com/blog/chatgpt\u003c/li\u003e\n\u003cli\u003e\u0026ldquo;LLMs (Large Language Models) are a type of artificial intelligence (AI) that can generate, understand, and interact with natural language, such as text and speech. Natural language is the way we humans communicate with each other, using words, sentences and grammar\u0026rdquo; (Dash, 2023, p.1)\u003c/li\u003e\n\u003cli\u003e\u0026ldquo;Generative AI is a form of artificial intelligence that utilizes techniques from machine learning and deep learning to generate original data\u0026rdquo; (Kalota, 2024) \u003c/li\u003e\n\u003cli\u003eA Multivocal Literature Review (MLR) comprises all accessible writings on a usual, often contemporary topic (Ogawa \u0026amp; Malen, 1991). \u003c/li\u003e\n\u003cli\u003eArduino is a prototyping platform simple enough to be used by any student, including children. The board\u0026apos;s design allows people to \u0026quot;build things that work\u0026quot; without attending a complete electronics course (Banzi, 2012; Severance, 2013). Its low-cost, quality, flexibility, and ease of use have made Arduino electronic prototyping boards an excellent option for developing projects on Educational Robotics http://www.arduino.cc/\u003c/li\u003e\n\u003cli\u003eRaspberry Pi is a low-cost single-board computer provided with a set of General-Purpose Input/Output pins where one can connect robotics devices. Roughly speaking, we can say that the Raspberry Pi is a single-board computer with a microcontroller (like an Arduino) built into it. The flexibility of the Raspberry Pi allows one to use it for the most diverse purposes, such as experimentation, residential automation, or simply fun. (Severance, 2013). https://www.raspberrypi.org/\u003c/li\u003e\n\u003cli\u003ehttps://www.youtube.com/\u003c/li\u003e\n\u003cli\u003ehttp://conselho.saude.gov.br/comissoes-cns/conep/\u003c/li\u003e\n\u003cli\u003ehttps://www.facebook.com/\u003c/li\u003e\n\u003cli\u003ehttps://www.whatsapp.com/\u003c/li\u003e\n\u003cli\u003ehttps://www.instagram.com/\u003c/li\u003e\n\u003cli\u003ehttps://scratch.mit.edu/\u003c/li\u003e\n\u003cli\u003eA type of user testing that uses testers who are NOT part of your organization and who are members of your product\u0026rsquo;s target market. The product under test is typically very close to completion.\u0026rdquo; (kaner et al., 2008, p.34)\u003c/li\u003e\n\u003cli\u003ehttps://www.un.org/development/desa/pd/sites/www.un.org.development.desa.pd/files/unpd_egm _201902_s1_sergeischerbov.pdf \u003c/li\u003e\n\u003cli\u003e\u0026ldquo;In-house testing performed by the test team (and possibly other interested, friendly insiders).\u0026rdquo; (Kaner et al., 2008, p.34)\u003c/li\u003e\n\u003cli\u003ehttps://www.ibm.com/spss\u003c/li\u003e\n\u003cli\u003eA computer works based on a cycle defined as Input-Process-Output (IPO). This is an elementary concept widely used in introductory computing and can be summarized as: \u003cem\u003eInput (Anything that is put into a system) - \u0026gt; Process (The steps or actions taken by the system) -\u0026gt; Output (The result of the process) \u003c/em\u003e(Maharaj et al., 2023, p.147).\u003c/li\u003e\n\u003cli\u003ehttps://www.hucff.ufrj.br/comite-de-etica-em-pesquisa/\u003c/li\u003e\n\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":true,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":true,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"Demystifying AI with AIcon2abs, Machine -learning Education, WiSARD algorithm, Accessible AI tools, Hands-on learning, Easy machine learning visualization","lastPublishedDoi":"10.21203/rs.3.rs-5728177/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-5728177/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eThis study is an expansion of a previous work aiming to evaluate the AIcon2abs method (\u003cem\u003eAI from Concrete to Abstract: Demystifying Artificial Intelligence to the general public\u003c/em\u003e), an innovative method aimed at increasing the public (including children) understanding of machine learning (ML). The approach employs the WiSARD algorithm, a weightless neural network known for its simplicity, and user accessibility. WiSARD does not require Internet, making it ideal for non-technical users and resource-limited environments. This method enables participants to intuitively visualize and interact with ML processes through engaging, hands-on activities, as if they were the algorithms themselves. The method allows users to intuitively visualize and understand the internal processes of training and classification through practical activities. Once WiSARD's functionality does not require an Internet connection, it can learn effectively from a minimal dataset, even from a single example. This feature enables users to observe how the machine improves its accuracy incrementally as it receives more data. Moreover, WiSARD generates \u003cem\u003emental images\u003c/em\u003e representing what it has learned, highlighting essential features of the classified data. AIcon2abs was tested through a six-hour remote course with 34 Brazilian participants, including 5 children (8\u0026ndash;11 yo), 5 adolescents (12\u0026ndash;17 yo), and 24 adults (21\u0026ndash;72 yo). Data analysis was conducted from two perspectives: a mixed-method pre-experiment (including hypothesis testing), and a qualitative phenomenological analysis. Nearly 100% of participants rated AIcon2abs positively, with the results demonstrating a high degree of satisfaction in achieving the intended outcomes. This research was approved by the CEP/HUCFF/UFRJ Research Ethics Committee.\u003c/p\u003e","manuscriptTitle":"How do machines learn? Evaluating the AIcon2abs method","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-01-06 14:49:09","doi":"10.21203/rs.3.rs-5728177/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"a8deab18-cf1f-4b2c-b794-17e9842bc3ff","owner":[],"postedDate":"January 6th, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[],"tags":[],"updatedAt":"2025-04-15T22:38:19+00:00","versionOfRecord":[],"versionCreatedAt":"2025-01-06 14:49:09","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-5728177","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-5728177","identity":"rs-5728177","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}