Adoption of Artificial Intelligence in the Judiciary: A Comparison of 28 Advanced Democracies

Adoption of Artificial Intelligence in the Judiciary: A Comparison of 28 Advanced Democracies | 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 Short Report Adoption of Artificial Intelligence in the Judiciary: A Comparison of 28 Advanced Democracies Clement Guitton, Vlada Druta, Markus Hinterleitner, Aurelia Tamò-Larrieux, and 1 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-5881593/v1 This work is licensed under a CC BY 4.0 License Status: Published Journal Publication published 23 Jul, 2025 Read the published version in Discover Artificial Intelligence → Version 1 posted 12 You are reading this latest preprint version Abstract Artificial Intelligence (AI) is increasingly used worldwide to make decisions, be it by public administrations, industry, banks, or insurers. One area with a particularly high impact on citizens is its use in judiciary processes. To this day, there has been too little investigation into what drives the willingness of a country to present itself as adopting AI in its judiciary. In this article, we show that the mere promise of efficiency gains is not enough of an explanation for the claim of adopting AI within judiciary processes. We show that the administrative burden (as measured by the number of days to trial), a government’s leaning within the political spectrum (namely towards the left), and the level of adoption of technology by governments in bordering neighboring countries predict countries’ announcement to adopt AI in their judiciary. Together, our findings from a data analysis of 28 advanced democracies form a theory of adoption respectively around functionality, politics, and diffusion of AI in the judiciary. Figures Figure 1 Figure 2 Figure 3 Introduction Automation within public administration has gained global momentum (Madan & Ashok, 2023), with artificial intelligence (AI) technologies being increasingly implemented to streamline government processes that directly impact citizens (Tangi et al., 2022; Grimmelikhuijsen & Tangi, 2024), and with keen interest from legislators on the topic (Schiff & Schiff, 2023). One area drawing significant attention is the judiciary, where AI is beginning to support various stages of legal proceedings (OECD, 2024). As the judiciary plays an important role in deciding citizen disputes and thereby significantly interferes in social relations, it is a particularly interesting use case to study the motivations behind AI adoption (Goel & Nelson, 2023). The integration of AI technologies into legal proceedings constitutes a far-reaching regulatory reform that may fundamentally alter traditional judicial practices, changing the ways policies are made and decisions are taken in a political system. Such reforms generally do not come lightly and are believed to be not only driven by utility considerations but by a broader range of political, functional, and/or ideational motivations (Knill & Tosun 2020). While the promise of efficiency gains and, thereby, a reduction of current backlogs in courts has driven much of the initial interest, this premise alone is unlikely to account for the complexities surrounding the integration of AI in judicial systems (Ingrams et. al., 2020). This article is organised into four key sections. Section 2 outlines the theoretical framework, positioning AI adoption in the judiciary as a form of regulatory reform whose adoption can be explained by looking at political, functional, and ideational factors. Section 3 describes the methodology employed. Section 4 presents the findings, showing that all three factors are statistically significant in influencing AI adoption in judicial systems. Section 5 provides a broader interpretation of these results, situating them within the context of judicial reform while also addressing the study's limitations and avenues for future research. Together, these sections build an exploratory understanding of the factors shaping AI’s integration into judiciary processes. Theorising Adoption There are various frameworks that aim to explain technology adoption with the most cited ones the Technology Acceptance Model (TAM; Davis, 1989), the Unified Theory of Acceptance and Use of Technology (UTAUT; Venkatesh et. al., 2003), and respective iterations on them (Venkatesh et. al., 2016). Recent years have witnessed a proliferation of analyses of these frameworks across many domains, most of them focusing on adoption of users and by the private sector (Emon, 2023). Additionally, with the emergence of more advanced technology, more studies have emerged specifically focusing on AI adoption but have, once more, largely focused on user adoptions in the private sector (Kelly et al., 2023). The findings of a review of 60 studies indicate that the most influential factors in AI adoption are perceived usefulness, performance expectancy, attitudes, trust, and effort expectancy, mirroring that the most used and cited theory for user acceptance was the extended Technology Acceptance Model (TAM) (Kelly et al., 2023). Some of these frameworks have also been applied to explore technology adoption in the public sector, uncovering the most prominent factors influencing adoption. With regard to the judiciary, notably, studies have highlighted that technology acceptance is primarily linked to perceptions of usefulness (Barysė, 2022; Ahmed et. al., 2024), utility of technology (Nguyen et. al., 2024) and perception of trust (Xu and Wang, 2021). The utility factor in technology adoption is further supported by numerous studies that emerged during the context of COVID-19; one study highlights that the legal profession has embraced technology as an important tool for enhancing access to justice (Muigua, 2020). Once again, a host of studies has looked at drivers behind user adoption of e-government services (Rana et al., 2017; Verkijika & De Wet, 2018)—a question which is beyond the scope of this research. A few others have focused on the drivers behind how the transition of public services to the digital realm occurs, highlighting the role of: the need to streamline administrative tasks and enhance accessibility for users (Ilieva et. al., 2024), the level of democracy and regional competition (Ingrams et. al., 2020; Bussell, 2011), perceived organisational benefits, in-house capacities and clear strategies (Grimmelikhuijsen & Tangi, 2024). At a very basic level, it is important to note that the capacity to pursue digital transformation in the public sector—particularly in sensitive domains such as the judiciary—is shaped by broader institutional conditions. In particular, high levels of institutional quality, such as those discussed in the literature on the “quality of government” (La Porta et al., 1999), appear to be a necessary precondition for governments to undertake and implement far-reaching reforms involving AI. This literature underscores the importance of effective, impartial, and corruption-free institutions in enabling states to carry out complex regulatory changes—conditions that are typically met in advanced democracies and that set the stage for the adoption of AI in judicial systems. At the same time, it is important to emphasise that the adoption of e-government services, in particular those involving AI components, constitute far-reaching regulatory reforms that face significant political, ideational, and functional hurdles before their adoption (Knill & Tosun 2020). The reason is that such reforms are likely to alter decision-making processes within government and that those involved in these processes may fear to lose out from the reforms. Especially in advanced democracies, which are very path-dependent (Pierson 2004) due to the presence of multiple veto points that can block substantial change (Tsebelis 2002), institutional reforms are unlikely to be the result of utility considerations alone. Even governments intent on doing the “right” thing often fail to do so because of structural or political constraints or because they are simply unaware of how to match problems to potential solutions. For example, when examining legal technology in contemporary USA and China, Wang (2020) highlights that the characteristics of legal tech are closely tied to each country’s political landscape (centralized versus decentralized governance), legal system objectives (whether aimed at increasing efficiency or enhancing justice), and judicial structure. Another study on e-justice in Switzerland and Brazil emphasises the interplay of individual factors, innovation features (e.g., usability), organisational aspects (team dynamics), and environmental influences (e.g., centralisation, federalism, and legislation) (Sousa et. al., 2022). Additionally, a study on implementing technology in the Canadian justice sector highlights that learning from the mistakes and successes of others plays an important role in fostering adoption (Bailey and Burkell, 2013). With these insights in mind, we develop a theoretical framework for explaining AI adoption in the judiciary that accounts for functional, political, and ideational factors. These factors have been shown to drive the adoption of other regulatory reforms in advanced democracies (Knill & Tosun 2020; Steinebach et al. 2024) and are hence well suited for an exploratory analysis in our area of interest. Moreover, our framework is based on the well-established insight that governments may embark on regulatory reforms for both substantive reason (i.e., they want a reform to address a particular problem) and for more symbolic reasons (i.e., they are less interested in the reform’s actual effects and more in the political message it sends; see e. g. Hacker & Pierson, 2014; Strøm, 1990). To encapsulate these ideas in a theoretically parsimonious way, we focus on three concrete factors: (i) the degree of overburdening of the judiciary, (ii) the political position of the government, and (iii) diffusion effects. In the following, we discuss these factors in detail and formulate expectations on how they should influence AI adoption in the judiciary. Functional pressures: Overburdening of the judiciary The proliferation of rules and regulations has been identified as a ubiquitous feature of modern democracies (Hinterleitner et al., 2024). As the administrative capacities necessary to implement ever more rules and regulations have not grown to the same extent, administrative overburdening has become a widespread phenomenon in advanced democracies (Fernandez-i-Marin et al., 2024). While this finding applies to public administrations in general, it has also been observed in the judiciary. Many judiciaries (Martinuzzi, 2017) currently work at or beyond their capacity limit and struggle with ever-growing case backlogs and longer durations to trial. Administrative overburdening creates pressure for governments to look for ways to slow down workload growth and allow administrations to shoulder greater workloads by streamlining and automating procedures. One example are regulatory offsetting schemes that seek to compel governments to compensate for the production of additional rules by simultaneously abolishing existing provisions and associated burdens (Steinebach et al. 2024). We thus expect that countries with overburdened judiciaries are more likely to adopt an AI project than countries whose judiciary faces no such problem. H1 : The average duration of a case to trial has a positive effect on the decision to adopt AI in the judiciary. Political factors: The position of the government Political parties are known to be both vote and policy seekers (Strøm, 1990). They compete for office by championing issues that their supporters care about; and, once in government, they seek to adopt policies that benefit their supporters and align with their own ideological orientations. Left-leaning parties are generally known to advocate for an expansive and influential government and broad-based access to public services. Left-leaning parties have historically championed a comprehensive welfare state, supported government oversight of markets (Jakobsen & Mortensen, 2015), and focused on social protection and human capital creation (Bremer & McDaniel, 2020; Häusermann et al., 2022). Right-leaning parties, on the contrary, have been shown to champion a smaller state that restricts access to social services, contingent on specific criteria. In the US, for example, Republican governments, which traditionally oppose generous welfare programs, have repeatedly managed to put a high administrative burden on welfare recipients so as to make it harder for them to claim the benefits they are entitled to (Herd & Moynihan 2019). We thus expect that left-leaning parties are generally more likely to adopt regulatory reforms that promise to make the state more responsive to its citizens. Since many AI projects in the judiciary seek to increase access to judicial services that citizens are entitled to (e.g., by making the law more transparent through anonymized case publications, by making the law more understandable to laypeople by adapting the language to the vernacular, see for more examples OECD, 2024; Tangi et al., 2022), we expect left-leaning parties to be positively associated with AI adoption while this should not be the case for right-leaning parties. [1] H2 : Left-leaning governments are more likely to decide to adopt AI in the judiciary than right-leaning governments. Matching solutions to problems: The role of diffusion effects Governments rarely act entirely independently when attempting to tackle specific political challenges. Instead, they often draw inspiration from the experiences and policy solutions of other governments (Marsh & Sharman, 2009; Meseguer, 2005). This tendency is even more pronounced when there are multiple regulatory options available, allowing governments to choose among several potential solutions rather than being limited to a single approach. Policy diffusion is especially likely between neighboring countries that are closely interlinked through trade relationships and political alliances, as these connections facilitate the sharing and adaptation of ideas and policies (Maggetti & Gilardi, 2016). Neighboring countries are especially prone to drawing insights from one another's experiences with various policies and to adopting each other’s attempts at problem-solving (Gilardi, 2016; Simmons & Elkins, 2004). Based on these insights, we expect that the level of adoption of technology in neighboring governments positively impacts a government’s decision to adopt AI in the judiciary. H3 : The level of adoption of e-government in bordering neighbors has a positive effect on the decision to adopt AI in the judiciary. [1] We note that right-wing parties are generally pro law and order (Baker et al, 2015; Ren et al., 2008). However, as described below in the section Methodology, the projects captured do not solely focus on seeking to enhance law and order, and in fact, very few even focus on criminal matters at all (see the Discussion section for this). Methodology Dependent variable We are interested in capturing data on the first AI project implemented by a country in the judiciary to assess the hypotheses; this corresponds to data for years 1997-2024 (1997 being the earliest year for which we found a project for the countries under study, and 2024 corresponding to when we conducted the research). From this, one component to capture is already evident: the projects are in different stages (we categorised them between not started, pilot / ongoing, and implemented). Furthermore, in order to analyse H2, it is important to capture when the project was initiated. We also sought to apply the typology from Guitton et al. (2022) to capture a project’s aim if it is geared towards increasing the accessibility to the law or its efficiency, as well as the area of implementation (courts, prosecutors, notaries) and what purpose exactly the project seeks to achieve (anonymisation, transcription, information extraction, etc.). We therefore captured 5 variables: year, stage, aim, area, and purpose. The sources for the data were twofold. First, we manually looked for information for each country for which sources for the independent variables existed (see below). Second, we cross-checked and complemented this information with data from the Council of Europe’s resource on “cyberjustice and artificial intelligence” (Council of Europe, 2024). We assume that countries directly provided data for this latter source and that a certain bias to put the countries in the best of light could have resulted as a consequence. It is also noteworthy that we only relied on secondary sources about the existence and data of each programme. In order to minimise errors, two of the authors independently went through the data collection. The list of 28 countries was primarily derived from the intersection of countries for which we could find information for the different independent variables. We sought to maximise this dataset and, in a second step, eliminated countries whose socio-political structures were judged to be too different from the rest of the dataset, particularly in terms of democratic institutions (e.g. Azerbaijan, Morocco), since our theory is specifically concerned with advanced democracies. The criterion for inclusion was for a country itself to label the technology that they were using as “AI”. Regarding the lack of accepted consensus around a definition of AI [2] , this approach suited our research goal, as we are interested in why countries would seek to present themselves as (early) adopters of AI. Following this, two specific types of projects were excluded, as they were systematically not presented under the “AI” label: projects related to risk-assessment-instruments, and those related to the automatic assignment of cases. The result of the data collection shows a rather homogeneous dataset, with only 5 countries with no implementation at all. 86% of the projects collected were in the context of courts, 10% for adjudication, and 5% for the overall justice system (note that one project was for both court and adjudication within the administration). Furthermore, most projects focused on anonymising court records for their release to the general public; the second most frequent category of projects concerned automatic speech-to-text transcription of court hearings (see Figure 1). Independent variable We used the three different metrics and sources for each hypothesis, as listed in Table 1: Table 1: Sources and metrics for hypotheses. H# Years Metric Source H1 2022 Disposition time: average duration to trial (in days) for criminal, civil, and administrative trials. European Commission for the Efficiency of Justice (2024) H2 1997*-2024 Political position of a government, measured as “the proportion of cabinet posts of right-wing parties in relation to total cabinet posts, weighted by the number of days in office in a given year” (Armingeon et al., 2024). Comparative Political Data Set (Armingeon et al., 2024) H3 2022 E-government development index: Difference between a country’s index and the average across its neighbouring countries. United Nations (2022) *. 1997 corresponds to the first year that an AI project surfaces in the subset of the 28 selected countries. The dataset for H2 only included data up to 2022. As we carried out a survival analysis (see below), this was sufficient for all countries whose adoption of AI in the judiciary came before or in 2022. For 3 other countries (namely: Romania, Slovenia, Spain), we completed the dataset by adding the composition of right-wing members in the government manually following Armingeon et al. (2024) for the years 2023-2024. We took the subset of countries for which data was available for each of the three introduced operationalisations. This led to slightly different datasets for H1/H3 and H2, with 25 countries considered in H1/H3 and 28 for H2 (the complete list is provided in Appendix A). Data analysis For H1 and H3, we carried out separate simple linear regressions including each of the five dependent variables, as this is a standard method within policy/political science to attempt at identifying potential relations between different sets of variables and is in line with how we posited the hypotheses (Chatterjee & Wiseman, 1983). This yielded a set of 5 linear regressions, with b 0 the intercept and b i>0 the linear regression coefficient. In all models, we also control for the quality of government (World Bank, 2022) and GDP per capita (World Bank, 2025) by taking 2022 data to take into account previous evidence on the factors that may drive AI adoption in government (La Porta et al., 1999; Goel & Nelson, 2023): Y year = b 0 + b 1 X H1 + b 2 X H2 + b 3 X H3 + b GovernmentQuality X GovernmentQuality + b GDPperCapita X GDPperCapita Y stage = b 0 + b 1 X H1 + b 2 X H2 + b 3 X H3 + b GovernmentQuality X GovernmentQuality + b GDPperCapita X GDPperCapita Y aim = b 0 + b 1 X H1 + b 2 X H2 + b 3 X H3 + b GovernmentQuality X GovernmentQuality + b GDPperCapita X GDPperCapita Y area = b 0 + b 1 X H1 + b 2 X H2 + b 3 X H3 + b GovernmentQuality X GovernmentQuality + b GDPperCapita X GDPperCapita Y purpose = b 0 + b 1 X H1 + b 2 X H2 + b 3 X H3 + b GovernmentQuality X GovernmentQuality + b GDPperCapita X GDPperCapita Furthermore, we conducted one-way ANOVA tests on the stage of the projects to try to evidence differences between the different groups, and in a second step, we merged the stage of projects so that they could fit within two groups only: those countries with no project started vs., those with an ongoing project or implemented one. This allowed us to carry out one-tail t-tests. For H2, further in line with the hypothesis, we carried out an event history analysis, which is a well-suited form of analysis when the phenomenon under a study has a duration with an occurrence at a discrete point in time (so called “time-to-event data”), and it allows us to identify a likelihood of a variable to trigger an “event” (see for a similar data analysis method Steinbach et al., 2024). Traditionally, scholars have used it to investigate survivability (i.e., “impact of treatments on the event of death”), hence the function bearing the name of “survival function”. In our case, every country is “at risk” of starting an AI project in its judiciary and this represents therefore the event under consideration; once this has occurred, the country is then “censored” (removed) from the dataset for the subsequent years. The unit of analysis is therefore the country level, time periods are years, and all data are continuous as explained in the section above. We take as a starting year 1995, slightly before the year of the first project on the compiled list (1997 in Austria). This allows us to compute the probability of survival to the time t Pr(T i ≥ t), with the following model description: y t,c : binary component that captures whether a country has started an AI project in the judiciary t: time expressed in year c: country We then compute a univariate Cox’s regression on the survival function for its statistical significance. In a second step, we repeated the event history analysis by adding as a treatment the aim followed behind kickstarting a project (whether for efficiency gain or to promote accessibility to the law). [2] https://cacm.acm.org/opinion/between-the-booms-ai-in-winter/ Results H1 and H3 We only find statistical significant results when considering the stage of the project; the other dependent variables did not result in statistically significant relations even if a Principal Component Analysis showed them as rather close together (see Figure 2), notably with stage very close to the area of implementation. When controlling for quality of government and GDP per capita, none of them have a significant effect on the dependent variables. Furthermore, within H1, we only obtained statistically significant results when considering duration of civil trials (see Table 2) and using one-tail t-tests with the stage of the projects clustered into 2 groups. Duration to criminal and administrative trials did not result in statistically significant results (see in Appendix B Table B.4), as well as a one-way ANOVA test was not statistically significant ( F =1.946, p =0.161, df=24). Table 2: Results are statistically significant for H1 and H3 when tested against the dependent variable “stage of the project”. N=25 H1 (duration to trial of civil cases) H3 (adoption in bordering countries) Linear regression coefficient against stage (p<0.01) 0.003 (p<0.01, R 2 =0.28) 8.051 (p<0.02, R 2 =0.21) T-stat (df=24) 3.206 (p<0.03) 1.904 (p<0.04) Therefore, we can accept H1 and H3 within this set of conditions: the average number of days to trial in a country as well as the level of e-government adoption of bordering neighboring countries both significantly influence the decision of the government of a country to present itself as starting projects on AI in the judiciary. H2 The survival function for the risk of starting an AI project obtained on N= 28 countries is shown in Figure 3; the Cox’s regression returns a statistically significant relation (coefficient=0.277, p<0.01). We can therefore accept H2: the left-leaning governments are more likely to present their country as starting projects on AI in the judiciary. However, we did not find that the aim of the project (efficiency or accessibility) was statistically relevant in the likelihood of a left-wing (or right-wing) to initiate an AI project in the judiciary (see Appendix B, Table B.5). Discussion We can therefore draw three conclusions about the drivers behind the decision to implement AI in the judiciary in the context of advanced democracies. First, we find that the duration to trial of civil cases plays a role in deciding to turn to AI to alleviate the burden. More specifically, the average number of days to trial in civil cases for countries with an AI implementation averaged 310 whereas it averaged only 126 for those countries with such implementation. This confirms the perception of the role of technology to remove some of the burden on the state as it seeks to find room to operate within existing constraints (financial resources, political capital). Second, we find that the leaning of the government in power does have an influence on the starting AI projects in the judiciary, with left-wing governments more likely to do so than right-wing ones. Of the 23 countries with an implementation, only six were kickstarted by a right-wing government (Canada, Czech Republic, Estonia, France, Latvia, Switzerland) and conversely, and 17 by a left-wing government (Austria, Croatia, Germany, Spain, Malta, Romenia, Slovenia, Denmark, Finland, Italy, Luxembourg, Netherlands, Poland, Portugal, Sweden, UK, USA). Also strengthening the conclusion is that of the five countries without an implementation (Belgium, Cyprus, Hungary, Lithuania, Norway), all but one country (Hungary) did go through governments with a different majority across the 26 years under study. A closer look at the countries in our sample reveals no consistent pattern in socio-political or economic structures that would account for the observed variation. Both lists—those with and without AI initiatives in the judiciary—include a mix of social democracies and economically liberal states, centralised and federal systems, high- and low-GDP countries, large and small populations, as well as EU and non-EU members. This diversity suggests that these structural factors are unlikely to explain the difference. Instead, the pattern appears to align more closely with an ideological inclination, specifically a greater propensity among left-leaning governments to pursue such initiatives. We therefore interpret this result in light of the agenda of left-wing governments to give more access to a country’s residents to state services, even beyond the judiciary, regardless of the aim when starting the AI project. The mixture of projects started actually falls under both aims of making the state more efficient, and to promote access to the law (e.g. with the single highest number of projects in the category of automatically anonymising court proceedings to publish them online, often explicitly to foster legal access), and this explains why the addition of the aim as a treatment in the event history analysis did not play a role. Yet, we contend that even projects geared to make the judiciary more efficient contributes to a more responsive state, and hence to one that can better fulfil its promise and mandate towards a country’s inhabitants. Our third and last finding concerns the role of governments in bordering neighbours in bearing an influence, although the argument may be a bit circular. Those countries with a judicial AI implementation have, in average, a positive difference between their e-government index and their neighbours’; those other countries without a judicial AI implementation have, in average, a negative difference. In other words, that means that those which have implemented judicial AI tools have generally a better readiness to implement access to public services via digital means, and those without lag behind their border’s peers when it comes to e-government. Limitations and Conclusion In conducting this study, we have faced a number of methodological hurdles probably due to the exploratory nature of the research question. With no clear framework to follow, and with no clear set of countries to focus on, we have iterated a number of times between frameworks, country sets, dependent/independent variables and data sources available before we could find meaningful and statistically relevant results (a few of the negative results of our investigation can be found in Appendix B). Our initial methodology was more traditional to comparative politics by picking three countries and qualitatively analysing them in-depth: two countries with judicial AI implementation, and a counter-case of a country without one. However, by seeking to generalise the hypotheses to a larger N , we did not succeed. In line with these first negative results, we would like to caution on the interpretation of our findings notably when it comes to extrapolating them outside of the dataset of countries under consideration. The sheer number and complexity of drivers possibly involved means that generalisation should be refrained. Similarly, the results reported here have a certain number of limitations per hypothesis too. First regarding H1, we could not find any statistically significant result when looking at the waiting time to trial in the cases of criminal and administrative cases (even when trying with different statistical tools, from Pearson’s correlation, to Principal Component Analysis, to one-way ANOVA). There might be a rationale explanation to this: when collecting the data, it has come to our attention that a few countries have explicitly referred to focusing on implementation of AI tools in the civil law domain, and it is hence sensical that we find this mirrored as to why they chose to focus on this sub-domain. Furthermore, countries’ focus on civil rather than criminal cases, especially as a first foray to deploy AI tools in the judiciary may be explained by how resolving procedural issues or small claims is generally less sensitive than criminal proceedings (Bell et. Al., 2022, Fine & Marsh, 2024). The higher stakes in criminal proceedings could therefore play a role, an argument which is also valid when looking at administrative justice where significant ethical and legal risks also arise, including concerns about the dehumanization of justice, real threats to the independence and impartiality of justice (Nouri et. Al., 2024). The many scandals which have arisen concerning using AI in public administration to make decisions quicker about social benefits (see from the Netherlands, to France, to the US, Tamò-Larrieux et al., 2025) appear to give credence to the argument about stakes at play—all the while highlighting the many pitfalls which exist when deploying AI tools in general (Guitton et al., 2024) and in the judiciary in particular (John et al, 2023). It remains that future qualitative studies using process tracing and (semi-) structured interviews with key decision-makers would be needed to confirm how the decision process was shaped Second, regarding H2, it may be argued that we only looked at the leaning of the majority in government as opposed to looking at the leaning of the minister of justice most likely in charge. The different datasets we considered (e.g. the Manifesto Project Dataset, Lehmann et al., 2024) also did not include this information, and we accept that this may skew the results, albeit only to a certain limited extent. Also noteworthy concerning H2 is that we could not find any relation when running the historical event analysis with as additional covariate the aim of the project (efficiency / accessibility); we also found no relation when looking at odds-ratio and regressions. We cannot therefore conclude that the leaning of the party in government has an influence on the type of project started, and we can “only” fail to accept H2 without making a distinction for the aims of projects. Lastly, regarding H3, we note that the operationalisation matters. We found no statistical significance in the following cases: when we only took the average (or standard deviation) between a country and its bordering neighbours’ e-government index, when we took only the e-government index without consideration of neighbours (the e-government index is supposed to be a relative assessment of countries, not an absolute measure), or when we defined neighbours more coarsely in terms of sub-continent (Northern Europe, Northern America, etc.). While we posit that our current operationalisation is the one that makes the most sense (by looking at the difference, we actually look at how the country differs in relation to its neighbours), we also acknowledge the difficulty in the fragility of the conclusion dependent on the definition of the metric used to assess it. Yet another challenge that we have experienced—on top of defining a subset of country and data availability—concerned the selection of the independent variable for a model. We started with a lot more hypotheses, but for the sake of coherence and shortness, we decided to present a model based only on the three hypotheses reported here. Other hypotheses for which we collected data and found no meaningful relations were: trust in AI, R&D spending, the degree of access to AI expertise, regulation concerning AI in the judiciary, the independence and corruption of the judiciary, and distance to power. We notably acknowledge that our initial thinking was, if not that these variables correlate to some extent with AI adoption in the judiciary, that they might be otherwise confounding variables. For each of them, however, we have either found no relation worth reporting, or we faced challenges to match with our subset of countries at the intersection of our other datasets. Despite these limitations, we still posit that the findings offer insights into AI adoption in the judiciary amidst the hype and craze that surrounds the topic of AI. By looking at data, we obtain different insights than what qualitative interviews can return on both adoption or non adoption (Druță et al., 2024) and offer a complementary picture. Whether decision-makers are aware that they follow certain drivers—be it on their perception of the state capacity as overburdened, of their own political party’s agenda, or of how competitive they want to be against their neighbours—, the research teases out the relevance of the factors nonetheless. Future research could consider finding relations with further factors—although it may be well noted again to consider that we did not find any when looking at trust in AI, R&D spending, the degree of access to AI expertise, regulation concerning AI in the judiciary, corruption, independence of the judiciary, and distance to power. Regarding the wealth of possible factors playing a role (e.g. role of international organisations, trust in the police vs trust in the courts, etc.), and as already highlighted in the literature review section (“Theorising Adoption”), there should not be a lack of material to collect and test. Future research should also further investigate whether these findings apply to AI use in other branches of government, as current research in this area remains limited. For example, research on public perceptions of AI in policing indicates the influence of political partisanship: Republican members of the public tend to favor predictive policing over automated internal reviews, whereas Democrats show higher support for both (Schiff et al., 2025). Additionally, a few studies have similarly suggested how e-government initiatives could support AI adoption for crime management (Gummadidala et al., 2020), and when fighting corruption and the shadow economy (Goel & Saunoris, 2016). It would be valuable to further explore how a country's political ideology influences AI implementation in law enforcement—not only in the judiciary—and examine whether neighboring e-government initiatives similarly impact AI adoption. Investigating these factors across different government branches, such as policing, could offer deeper insights into the key drivers of AI adoption and the role of context in shaping policies. Declarations Ethics, Consent to Participate, and Consent to Publish declarations Not applicable. Author contributions statement CG wrote the main manuscript and carried out the data analysis; VD and MH wrote the theory and hypotheses; CG and VD collected the data; ATM and SM gave directions to the research; all contributed to the conceptualisation of the research and all reviewed the manuscript. Funding declaration Not applicable. References Ahmed, M. A., Kaya, T., & Karanfiller, T. (2024). Evaluating E-court systems in regional governments in developing countries using technology acceptance model. Information Development , https://doi.org/10.1177/02666669241229176 Armingeon, Klaus, Engler, S., Leemann, L., & Weisstanner, D. (2024). Comparative Political Data Set 1960-2022. Zurich/Lueneburg/Lucerne: University of Zurich, Leuphana University Lueneburg, and University of Lucerne. Bailey, J., & Burkell, J. (2013). Implementing technology in the justice sector: A Canadian perspective. Canadian Journal of Law and Technology , 11(2). Baker, T., Falco Metcalfe, C., Berenblum, T., Aviv, G., & Gertz, M. (2015). Examining Public Preferences for the Allocation of Resources to Rehabilitative Versus Punitive Crime Policies. Criminal Justice Policy Review , 26(5), 448-462. Barysė, D. (2022). People’s attitudes towards technologies in courts. Laws , 11(5), 71. Bell, F., Moses, L.B., Legg, M., Silove, J., & Zalnieriute, M. (2022). AI decision-making and the courts: A guide for judges, tribunal members and court administrators. Sydney: Australasian Institute of Judicial Administration Incorporated. Bremer, B., & McDaniel, S. (2020). The ideational foundations of social democratic austerity in the context of the Great Recession. Socio-Economic Review , 18 (2), 439–463. Bussell, J. (2011). Explaining cross-national variation in government adoption of new technologies. International Studies Quarterly , 55 (1), 267-280. Chatterjee, S., & Wiseman, F. (1983). Use of regression diagnostics in political science research. American Journal of Political Science , 27 (3), 601-613. Council of Europe (2024). Resource Centre Cyberjustice and AI. https://public.tableau.com/app/profile/cepej/viz/ResourceCentreCyberjusticeandAI/AITOOLSINITIATIVESREPORT?publish=yes Davis, F. D. (1989). Technology acceptance model: TAM. Al-Suqri, MN, Al-Aufi, AS: Information Seeking Behavior and Technology Adoption , 205 , 219. Druță, V., Tamo-Larrieux, A., Guitton, C., & van Dijck, G. (2024, August). From Analog Risk Assessment Instruments to Digital Ones: Mapping the Concerns. In Proceedings of Ongoing Research, Practitioners, Posters, Workshops, and Projects of the International Conference EGOV-CeDEM-ePart 2024 . Emon, M. M. H. (2023). Insights Into Technology Adoption: A Systematic Review of Framework, Variables and Items. Information Management and Computer Science , 6 (2), 27-33 European Commission for the Efficiency of Justice (2024). https://public.tableau.com/app/profile/cepej/viz/EfficiencyEN/Efficiency Fernández‐i‐Marín, X., Hinterleitner, M., Knill, C., & Steinebach, Y. (2024a). “Bureaucratic Overburdening in Advanced Democracies.” Public Administration Review , 84 (4), 696-709. Fine, A., & Marsh, S. (2024). Judicial leadership matters (yet again): the association between judge and public trust for artificial intelligence in courts. Discover Artificial Intelligence , 4 (1), 44. Gilardi, F. (2016). “Four Ways we Can Improve Policy Diffusion Research.” State Politics & Policy Quarterly 16(1): 8–21. Goel, R. K., & Nelson, M. A. (2023). Awareness of artificial intelligence: Diffusion of information about AI versus ChatGPT in the United States (No. 2259). Kiel Working Paper. Goel, R. K., & Saunoris, J. W. (2016). Virtual versus physical government decentralization: Effects on corruption and the shadow economy. Public Budgeting & Finance , 36 (4), 68-93. Grimmelikhuijsen, S., & Tangi, L. (2024). What factors influence perceived artificial intelligence adoption by public managers?, European Commission , JRC138684. https://data.europa.eu/doi/10.2760/0179285. Gummadidala, P. R., Karippur, N. K., & Koilakuntla, M. (2020). Analysis of factors influencing the adoption of artificial intelligence for crime management. In Re-imagining Diffusion and Adoption of Information Technology and Systems: A Continuing Conversation: IFIP WG 8.6 International Conference on Transfer and Diffusion of IT, TDIT 2020, Tiruchirappalli, India, December 18–19, 2020, Proceedings, Part I (pp. 3-9). Springer International Publishing. Guitton, C., Mayer, S., Tamo-Larrieux, A., Van Landuyt, D., Fosch-Villaronga, E., Kamara, I., & Pałka, P. (2024). Responsible automatically processable regulation. AI & SOCIETY , 1-16. Guitton, C., Tamò-Larrieux, A., & Mayer, S. (2022). A typology of automatically processable regulation. Law, Innovation and Technology , 14 (2), 267-304. Hacker, JS., & Pierson, P. (2014). “After the ‘Master Theory’: Downs, Schattschneider, and the Rebirth of Policy-Focused Analysis.” Perspectives on Politics , 12(03), 643–662. Häusermann, S., Garritzmann, J., & Palier, B. (2022). The politics of social investment. In Julian L. Garritzmann, Silja Häusermann, and Bruno Palier (Eds.), The world politics of social investment . Oxford University Press. Herd, P., & Moynihan, DP (2019). Administrative burden: Policymaking by other means . Russell Sage Foundation. Hinterleitner, M., Knill, C., & Steinebach, Y. (2024). “The Growth of Policies, Rules, and Regulations: A Review of the Literature and Research Agenda”, Regulation & Governance , 18 (2), 637-654. Ilieva, G., Yankova, T., Ruseva, M., Dzhabarova, Y., Zhekova, V., Klisarova-Belcheva, S., ... & Dimitrov, A. (2024). Factors Influencing User Perception and Adoption of E-Government Services. Administrative Sciences , 14 (3), 54. Ingrams, A., Manoharan, A., Schmidthuber, L., & Holzer, M. (2020). Stages and determinants of e-government development: A twelve-year longitudinal study of global cities. International Public Management Journal, 23 (6), 731-769. Jakobsen, MLF., and Mortensen, PB (2015). “How politics shapes the growth of rules.” Governance , 28 , 497–515. John, A. M., Aiswarya, M. U., & Panachakel, J. T. (2023, October). Ethical challenges of using artificial intelligence in judiciary. In 2023 IEEE International Conference on Metrology for eXtended Reality, Artificial Intelligence and Neural Engineering (MetroXRAINE) (pp. 723-728). IEEE. Kelly, S., Kaye, S. A., & Oviedo-Trespalacios, O. (2023). What factors contribute to the acceptance of artificial intelligence? A systematic review. Telematics and Informatics , 77 , 101925. Knill, C., & Tosun, J. (2020). Public policy: A new introduction . Bloomsbury Publishing. Legal technology acceptance in Vietnam’s courts. Cogent Business & Management, 11(1), 2321955. Lehmann, P., Franzmann, S., Al-Gaddooa, D., Burst, T., Ivanusch, C., Regel, S., Riethmüller, F., Volkens, A., Weßels, B., Zehnter, L. (2024): The Manifesto Data Collection. Manifesto Project (MRG/CMP/MARPOR). Version 2024a. Berlin: Wissenschaftszentrum Berlin für Sozialforschung (WZB) / Göttingen: Institut für Demokratieforschung (IfDem). https://doi.org/10.25522/manifesto.mpds.2024a Madan, R., & Ashok, M. (2023). AI adoption and diffusion in public administration: A systematic literature review and future research agenda. Government Information Quarterly , 40 (1), 101774. Maggetti, M., & Gilardi, F. (2016). “Problems (and solutions) in the measurement of policy diffusion mechanisms.” Journal of Public Policy , 36(1), 87-107. Marsh, D., & Sharman, JC. (2009). “Policy diffusion and policy transfer.” Policy Studies , 30(3), 269-288. Martinuzzi, A. (2017). Taking Justice Seriously: The Problem of Courts Overload and the New Model of Judicial Process (April 6, 2017). Civil Procedure Review , 8(1), 65-106. Meseguer, C.. 2005. “Policy learning, policy diffusion, and the making of a new order.” The Annals of the American Academy of Political and Social Science , 598(1), 67-82. Muigua, K. (2020). Legal Practice and New Frontiers: Embracing Technology for Enhanced Efficiency and Access to Justice. ADR Journal, 11 (2). Nguyen, N. A. D., Nguyen, V. P., & Bui, K. H. (2024). Legal technology acceptance in Vietnam’s courts. Cogent Business & Management , 11 (1), 2321955. Nouri, Z., Salah, W. B., & AlOmran, N. (2024). Artificial Intelligence and Administrative Justice: An Analysis of Predictive Justice in France. Hasanuddin Law Review , 10 (2), 119-143. OECD (2024), “Governing with Artificial Intelligence: Are governments ready?”, OECD Artificial Intelligence Papers , No. 20, OECD Publishing, Paris, https://doi.org/10.1787/26324bc2-en. Pierson, P. (2004). Politics in Time . Princeton: Princeton University Press. Rana, N. P., Dwivedi, Y. K., Lal, B., Williams, M. D., & Clement, M. (2017). Citizens’ adoption of an electronic government system: towards a unified view. Information systems frontiers , 19 , 549-568. Ren, L., Zhao, J., and Lovrich, NP. (2008). Liberal versus conservative public policies on crime: What was the comparative track record during the 1990s?, Journal of Criminal Justice, 36 (4), 316-325. Schiff, K. J., Schiff, D. S., Adams, I. T., McCrain, J., & Mourtgos, S. M. (2025). Institutional factors driving citizen perceptions of AI in government: Evidence from a survey experiment on policing. Public Administration Review , 85 (2), 451-467. Schiff, Daniel S., & Schiff, KJ. (2023). “ Narratives and Expert Information in Agenda-setting: Experimental Evidence on State Legislator Engagement With Artificial Intelligence Policy.” Policy Studies Journal 51(4): 817–842. Simmons, Beth A., & Elkins, Z. (2004). “The Globalization of Liberalization: Policy Diffusion in the International Political Economy.” American Political Science Review 98(1): 171–189. Sousa, M., Kettiger, D., & Lienhard, A. (2022). E-justice in Switzerland and Brazil: Paths and Experiences. In IJCA (Vol. 13, p. 1). Steinebach, Y., Hinterleitner, M., & Fernández‐i‐Marín, X. (2024). Regulatory offsetting in advanced democracies. Public Administration Review . Strøm, Kare (1990). “A behavioral theory of competitive political parties.” American Journal of Political Science 34(2): 565–598. Tamò-Larrieux, A., Guitton, C., & Mayer, S. (2025). AI and Law: How Automation is Changing the Law . CRC Press. Tangi, L., Van Noordt, C., Combetto, M., Gattwinkel, D. and Pignatelli, F., (2022). AI Watch. European landscape on the use of Artificial Intelligence by the Public Sector, Publications Office of the European Union , doi:10.2760/39336, JRC129301. Tsebelis, G. (2002). Veto players: How political institutions work . Princeton University Press. United Nations (2022). E-Government Development Index. https://publicadministration.un.org/egovkb/en-us/About/Overview/-E-Government-Development-Index Venkatesh, V., Morris, M.G., Davis, G.B., Davis, F.D. (2003). User acceptance of information technology: Toward a unified view. MIS Quarterly , 27 (3), 425–478. Venkatesh, V., Thong, J.Y., Xu, X. (2016). Unified theory of acceptance and use of technology: A synthesis and the road ahead. J ournal of the association for Information Systems, 17(5), 328–376. Verkijika, S. F., & De Wet, L. (2018). E-government adoption in sub-Saharan Africa. Electronic Commerce Research and Applications , 30 , 83-93. Wang, R. (2020). Legal technology in contemporary USA and China. Computer Law & Security Review, 39 , 105459. World Bank (2022). Worldwide Governance Indicators, 2024 Update. https://www.worldbank.org/en/publication/worldwide-governance-indicators, accessed on March 25, 2025. World Bank (2025, March 24). GDP per capita (current US$). https://data.worldbank.org/indicator/NY.GDP.PCAP.CD, accessed on March 25, 2025. Xu, N., & Wang, K. J. (2021). Adopting robot lawyer? The extending artificial intelligence robot lawyer technology acceptance model for legal industry by an exploratory study. Journal of Management & Organization , 27 (5), 867-885. Additional Declarations No competing interests reported. Supplementary Files Appendix.docx Cite Share Download PDF Status: Published Journal Publication published 23 Jul, 2025 Read the published version in Discover Artificial Intelligence → Version 1 posted Editorial decision: Accepted 13 May, 2025 Editor assigned by journal 12 May, 2025 Reviewers agreed at journal 26 Apr, 2025 Reviews received at journal 25 Apr, 2025 Reviews received at journal 25 Apr, 2025 Reviewers agreed at journal 25 Apr, 2025 Reviews received at journal 25 Apr, 2025 Reviewers agreed at journal 25 Apr, 2025 Reviewers agreed at journal 25 Apr, 2025 Reviewers invited by journal 23 Apr, 2025 Submission checks completed at journal 22 Apr, 2025 First submitted to journal 03 Apr, 2025 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-5881593","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Short Report","associatedPublications":[],"authors":[{"id":446858402,"identity":"c0e0e9bf-d46b-4561-bcc0-ffc5260835ca","order_by":0,"name":"Clement Guitton","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAABBklEQVRIiWNgGAWjYDACdjiLsQFEysEYCTi1MINJA5CWxoYDDAzGpGgBKgVqSWyASuDUwt/MfHTDxx1/GOQjktsff6i4k77hdnPzxy8Mdnm4tEgcZku7OfOMAYPhjUSgw848y91w52CbtAxDcjEuLQbMPGa3eduAWmYAtRxsO5y74UZiG7MEwwG4CzG18H+7/RdJS7rBjcTmz/i18LDdZgRqkZeAaEkAammQ/IBHC9AvZjd724x5DHgeNs44c+aw4Uygw6QZDJJxauFvb35242ebnJx8e/qDDxUVh+X5bqQ//vijwg6nFhjgMTiAxGPmMSCgHgTkkQ1l/EGEjlEwCkbBKBgxAAAwkGLbD6YVpwAAAABJRU5ErkJggg==","orcid":"","institution":"University of St. Gallen","correspondingAuthor":true,"prefix":"","firstName":"Clement","middleName":"","lastName":"Guitton","suffix":""},{"id":446858403,"identity":"dfd757dc-59e8-4df9-94e8-382d87d1a73e","order_by":1,"name":"Vlada Druta","email":"","orcid":"","institution":"University of Lausanne","correspondingAuthor":false,"prefix":"","firstName":"Vlada","middleName":"","lastName":"Druta","suffix":""},{"id":446858404,"identity":"64ed1589-50a6-441c-92ee-799372dc71a2","order_by":2,"name":"Markus Hinterleitner","email":"","orcid":"","institution":"University of Lausanne","correspondingAuthor":false,"prefix":"","firstName":"Markus","middleName":"","lastName":"Hinterleitner","suffix":""},{"id":446858405,"identity":"119b788b-7a4d-488e-b752-17d87a5341ee","order_by":3,"name":"Aurelia Tamò-Larrieux","email":"","orcid":"","institution":"University of Lausanne","correspondingAuthor":false,"prefix":"","firstName":"Aurelia","middleName":"","lastName":"Tamò-Larrieux","suffix":""},{"id":446858406,"identity":"9b751059-1c45-4093-860a-95cf11713451","order_by":4,"name":"Simon Mayer","email":"","orcid":"","institution":"University of St. Gallen","correspondingAuthor":false,"prefix":"","firstName":"Simon","middleName":"","lastName":"Mayer","suffix":""}],"badges":[],"createdAt":"2025-01-22 14:23:05","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-5881593/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-5881593/v1","draftVersion":[],"editorialEvents":[{"content":"https://doi.org/10.1007/s44163-025-00311-y","type":"published","date":"2025-07-24T00:00:00+00:00"}],"editorialNote":"","failedWorkflow":false,"files":[{"id":81287592,"identity":"cdd89211-85b7-4067-a401-436569e61a05","added_by":"auto","created_at":"2025-04-24 11:08:53","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":31316,"visible":true,"origin":"","legend":"\u003cp\u003eFrequency of projects by types of application.\u003c/p\u003e","description":"","filename":"1.png","url":"https://assets-eu.researchsquare.com/files/rs-5881593/v1/190fa474188c3c4838fc85fe.png"},{"id":81287586,"identity":"26eb9580-feb5-4dff-9852-b4389ec13750","added_by":"auto","created_at":"2025-04-24 11:08:52","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":166573,"visible":true,"origin":"","legend":"\u003cp\u003ePrincipal Component Analysis for the dependent variables and H1/H3. As the vectors are not orthogonal, this can be taken as an indication of possible statistical relation between them.\u003c/p\u003e","description":"","filename":"2.png","url":"https://assets-eu.researchsquare.com/files/rs-5881593/v1/17d99cf8c503aff213b71d6f.png"},{"id":81287582,"identity":"e7151214-e255-4e15-b8cb-e1137c425924","added_by":"auto","created_at":"2025-04-24 11:08:52","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":46794,"visible":true,"origin":"","legend":"\u003cp\u003eThe survival function shows a higher probability of projects being started from a left-wing government majority, and this constantly over time.\u003c/p\u003e","description":"","filename":"3.png","url":"https://assets-eu.researchsquare.com/files/rs-5881593/v1/068cd68dc19ca15ed280fc5d.png"},{"id":87744574,"identity":"241d6009-ea99-48e9-8a34-a4f0475e4ea4","added_by":"auto","created_at":"2025-07-28 14:02:40","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":762515,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-5881593/v1/991d02a0-2544-467b-9a30-91f0e727a42a.pdf"},{"id":81287968,"identity":"dde86f55-d09a-46ef-a758-ce987922fb95","added_by":"auto","created_at":"2025-04-24 11:16:52","extension":"docx","order_by":1,"title":"","display":"","copyAsset":false,"role":"supplement","size":91683,"visible":true,"origin":"","legend":"","description":"","filename":"Appendix.docx","url":"https://assets-eu.researchsquare.com/files/rs-5881593/v1/549b4ccd045b2532d7cf3040.docx"}],"financialInterests":"No competing interests reported.","formattedTitle":"Adoption of Artificial Intelligence in the Judiciary: A Comparison of 28 Advanced Democracies","fulltext":[{"header":"Introduction","content":"\u003cp\u003eAutomation within public administration has gained global momentum (Madan \u0026amp; Ashok, 2023), with artificial intelligence (AI) technologies being increasingly implemented to streamline government processes that directly impact citizens (Tangi et al., 2022; Grimmelikhuijsen \u0026amp; Tangi, 2024), and with keen interest from legislators on the topic (Schiff \u0026amp; Schiff, 2023). One area drawing significant attention is the judiciary, where AI is beginning to support various stages of legal proceedings (OECD, 2024). As the judiciary plays an important role in deciding citizen disputes and thereby significantly interferes in social relations, it is a particularly interesting use case to study the motivations behind AI adoption (Goel \u0026amp; Nelson, 2023).\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eThe integration of AI technologies into legal proceedings constitutes a far-reaching regulatory reform that may fundamentally alter traditional judicial practices, changing the ways policies are made and decisions are taken in a political system. Such reforms generally do not come lightly and are believed to be not only driven by utility considerations but by a broader range of political, functional, and/or ideational motivations (Knill \u0026amp; Tosun 2020). While the promise of efficiency gains and, thereby, a reduction of current backlogs in courts has driven much of the initial interest, this premise alone is unlikely to account for the complexities surrounding the integration of AI in judicial systems (Ingrams et. al., 2020).\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eThis article is organised into four key sections. Section 2 outlines the theoretical framework, positioning AI adoption in the judiciary as a form of regulatory reform whose adoption can be explained by looking at political, functional, and ideational factors. Section 3 describes the methodology employed. Section 4 presents the findings, showing that all three factors are statistically significant in influencing AI adoption in judicial systems. Section 5 provides a broader interpretation of these results, situating them within the context of judicial reform while also addressing the study\u0026apos;s limitations and avenues for future research. Together, these sections build an exploratory understanding of the factors shaping AI\u0026rsquo;s integration into judiciary processes.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eTheorising Adoption\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThere are various frameworks that aim to explain technology adoption with the most cited ones the Technology Acceptance Model (TAM; Davis, 1989), the Unified Theory of Acceptance and Use of Technology (UTAUT; Venkatesh et. al., 2003), and respective iterations on them (Venkatesh et. al., 2016). Recent years have witnessed a proliferation of analyses of these frameworks across many domains, most of them focusing on adoption of \u003cem\u003eusers\u003c/em\u003e and by the \u003cem\u003eprivate sector\u003c/em\u003e (Emon, 2023). Additionally, with the emergence of more advanced technology, more studies have emerged specifically focusing on AI adoption but have, once more, largely focused on \u003cem\u003euser\u003c/em\u003e adoptions in the private sector (Kelly et al., 2023). The findings of a review of 60 studies indicate that the most influential factors in AI adoption are perceived usefulness, performance expectancy, attitudes, trust, and effort expectancy, mirroring that the most used and cited theory for \u003cem\u003euser\u003c/em\u003e acceptance was the extended Technology Acceptance Model (TAM) (Kelly et al., 2023).\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eSome of these frameworks have also been applied to explore technology adoption in the public sector, uncovering the most prominent factors influencing adoption. With regard to the judiciary, notably, studies have highlighted that technology acceptance is primarily linked to perceptions of usefulness (Barysė, 2022; Ahmed et. al., 2024), utility of technology (Nguyen et. al., 2024) and perception of trust (Xu and Wang, 2021). The utility factor in technology adoption is further supported by numerous studies that emerged during the context of COVID-19; one study highlights that the legal profession has embraced \u0026nbsp;technology as an important tool for enhancing access to justice (Muigua, 2020). Once again, a host of studies has looked at drivers behind \u003cem\u003euser\u003c/em\u003e adoption of e-government services (Rana et al., 2017; Verkijika \u0026amp; De Wet, 2018)\u0026mdash;a question which is beyond the scope of this research.\u0026nbsp;A few others have focused on the drivers behind how the transition of public services to the digital realm occurs, highlighting the role of: the need to streamline administrative tasks and enhance accessibility for users (Ilieva et. al., 2024), the level of democracy and regional competition (Ingrams et. al., 2020; Bussell, 2011), perceived organisational benefits, in-house capacities and clear strategies (Grimmelikhuijsen \u0026amp; Tangi, 2024).\u003c/p\u003e\n\u003cp\u003eAt a very basic level, it is important to note that the capacity to pursue digital transformation in the public sector\u0026mdash;particularly in sensitive domains such as the judiciary\u0026mdash;is shaped by broader institutional conditions. In particular, high levels of institutional quality, such as those discussed in the literature on the \u0026ldquo;quality of government\u0026rdquo; (La Porta et al., 1999), appear to be a necessary precondition for governments to undertake and implement far-reaching reforms involving AI. This literature underscores the importance of effective, impartial, and corruption-free institutions in enabling states to carry out complex regulatory changes\u0026mdash;conditions that are typically met in advanced democracies and that set the stage for the adoption of AI in judicial systems.\u003c/p\u003e\n\u003cp\u003eAt the same time, it is important to emphasise that the adoption of e-government services, in particular those involving AI components, constitute far-reaching regulatory reforms that face significant political, ideational, and functional hurdles before their adoption (Knill \u0026amp; Tosun 2020). The reason is that such reforms are likely to alter decision-making processes within government and that those involved in these processes may fear to lose out from the reforms. Especially in advanced democracies, which are very path-dependent (Pierson 2004) due to the presence of multiple veto points that can block substantial change (Tsebelis 2002), institutional reforms are unlikely to be the result of utility considerations alone. Even governments intent on doing the \u0026ldquo;right\u0026rdquo; thing often fail to do so because of structural or political constraints or because they are simply unaware of how to match problems to potential solutions. For example, when examining legal technology in contemporary USA and China, Wang (2020) highlights that the characteristics of legal tech are closely tied to each country\u0026rsquo;s political landscape (centralized versus decentralized governance), legal system objectives (whether aimed at increasing efficiency or enhancing justice), and judicial structure. Another study on e-justice in Switzerland and Brazil emphasises the interplay of individual factors, innovation features (e.g., usability), organisational aspects (team dynamics), and environmental influences (e.g., centralisation, federalism, and legislation) (Sousa et. al., 2022). Additionally, a study on implementing technology in the Canadian justice sector highlights that learning from the mistakes and successes of others plays an important role in fostering adoption (Bailey and Burkell, 2013). \u0026nbsp;\u003c/p\u003e\n\u003cp\u003eWith these insights in mind, we develop a theoretical framework for explaining AI adoption in the judiciary that accounts for functional, political, and ideational factors. These factors have been shown to drive the adoption of other regulatory reforms in advanced democracies (Knill \u0026amp; Tosun 2020; Steinebach et al. 2024) and are hence well suited for an exploratory analysis in our area of interest. Moreover, our framework is based on the well-established insight that governments may embark on regulatory reforms for both substantive reason (i.e., they want a reform to address a particular problem) and for more symbolic reasons (i.e., they are less interested in the reform\u0026rsquo;s actual effects and more in the political message it sends; see e. g. Hacker \u0026amp; Pierson, 2014; Str\u0026oslash;m, 1990). To encapsulate these ideas in a theoretically parsimonious way, we focus on three concrete factors: (i) the degree of overburdening of the judiciary, (ii) the political position of the government, and (iii) diffusion effects. In the following, we discuss these factors in detail and formulate expectations on how they should influence AI adoption in the judiciary.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFunctional pressures: Overburdening of the judiciary\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe proliferation of rules and regulations has been identified as a ubiquitous feature of modern democracies (Hinterleitner et al., 2024). As the administrative capacities necessary to implement ever more rules and regulations have not grown to the same extent, administrative overburdening has become a widespread phenomenon in advanced democracies (Fernandez-i-Marin et al., 2024). While this finding applies to public administrations in general, it has also been observed in the judiciary. Many judiciaries (Martinuzzi, 2017) currently work at or beyond their capacity limit and struggle with ever-growing case backlogs and longer durations to trial. Administrative overburdening creates pressure for governments to look for ways to slow down workload growth and allow administrations to shoulder greater workloads by streamlining and automating procedures. One example are regulatory offsetting schemes that seek to compel governments to compensate for the production of additional rules by simultaneously abolishing existing provisions and associated burdens (Steinebach et al. 2024). We thus expect that countries with overburdened judiciaries are more likely to adopt an AI project than countries whose judiciary faces no such problem.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eH1\u003c/strong\u003e: The average duration of a case to trial has a positive effect on the decision to adopt AI in the judiciary.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003ePolitical factors: The position of the government\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003ePolitical parties are known to be both vote and policy seekers (Str\u0026oslash;m, 1990). They compete for office by championing issues that their supporters care about; and, once in government, they seek to adopt policies that benefit their supporters and align with their own ideological orientations. Left-leaning parties are generally known to advocate for an expansive and influential government and broad-based access to public services. Left-leaning parties have historically championed a comprehensive welfare state, supported government oversight of markets (Jakobsen \u0026amp; Mortensen, 2015), and focused on social protection and human capital creation (Bremer \u0026amp; McDaniel, 2020; Häusermann et al., 2022). Right-leaning parties, on the contrary, have been shown to champion a smaller state that restricts access to social services, contingent on specific criteria. In the US, for example, Republican governments, which traditionally oppose generous welfare programs, have repeatedly managed to put a high administrative burden on welfare recipients so as to make it harder for them to claim the benefits they are entitled to (Herd \u0026amp; Moynihan 2019). We thus expect that left-leaning parties are generally more likely to adopt regulatory reforms that promise to make the state more responsive to its citizens. Since many AI projects in the judiciary seek to increase access to judicial services that citizens are entitled to (e.g., by making the law more transparent through anonymized case publications, by making the law more understandable to laypeople by adapting the language to the vernacular, see for more examples OECD, 2024; Tangi et al., 2022), we expect left-leaning parties to be positively associated with AI adoption while this should not be the case for right-leaning parties.\u003csup\u003e\u003csup\u003e[1]\u003c/sup\u003e\u003c/sup\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eH2\u003c/strong\u003e: Left-leaning governments are more likely to decide to adopt AI in the judiciary than right-leaning governments.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eMatching solutions to problems: The role of diffusion effects\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eGovernments rarely act entirely independently when attempting to tackle specific political challenges. Instead, they often draw inspiration from the experiences and policy solutions of other governments (Marsh \u0026amp; Sharman, 2009; Meseguer, 2005). This tendency is even more pronounced when there are multiple regulatory options available, allowing governments to choose among several potential solutions rather than being limited to a single approach. Policy diffusion is especially likely between neighboring countries that are closely interlinked through trade relationships and political alliances, as these connections facilitate the sharing and adaptation of ideas and policies (Maggetti \u0026amp; Gilardi, 2016). Neighboring countries are especially prone to drawing insights from one another\u0026apos;s experiences with various policies and to adopting each other\u0026rsquo;s attempts at problem-solving (Gilardi, 2016; Simmons \u0026amp; Elkins, 2004). Based on these insights, we expect that the level of adoption of technology in neighboring governments positively impacts a government\u0026rsquo;s decision to adopt AI in the judiciary.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eH3\u003c/strong\u003e: The level of adoption of e-government in bordering neighbors has a positive effect on the decision to adopt AI in the judiciary.\u003c/p\u003e\n\u003cp\u003e\u003csup\u003e\u003csup\u003e[1]\u003c/sup\u003e\u003c/sup\u003e We note that right-wing parties are generally pro law and order (Baker et al, 2015; Ren et al., 2008). However, as described below in the section Methodology, the projects captured do not solely focus on seeking to enhance law and order, and in fact, very few even focus on criminal matters at all (see the Discussion section for this).\u003c/p\u003e"},{"header":"Methodology","content":"\u003ch3\u003eDependent variable\u003c/h3\u003e\n\u003cp\u003eWe are interested in capturing data on the \u003cem\u003efirst AI project implemented by a country in the judiciary\u003c/em\u003e to assess the hypotheses; this corresponds to data for years 1997-2024 (1997 being the earliest year for which we found a project for the countries under study, and 2024 corresponding to when we conducted the research). From this, one component to capture is already evident: the projects are in different stages (we categorised them between not started, pilot / ongoing, and implemented). Furthermore, in order to analyse H2, it is important to capture when the project was initiated. We also sought to apply the typology from Guitton et al. (2022) to capture a project\u0026rsquo;s \u003cem\u003eaim\u003c/em\u003e if it is geared towards increasing the accessibility to the law or its efficiency, as well as the \u003cem\u003earea of implementation\u003c/em\u003e (courts, prosecutors, notaries) and what purpose exactly the project seeks to achieve (anonymisation, transcription, information extraction, etc.). We therefore captured 5 variables: year, stage, aim, area, and purpose.\u003c/p\u003e\n\u003cp\u003eThe sources for the data were twofold. First, we manually looked for information for each country for which sources for the independent variables existed (see below). Second, we cross-checked and complemented this information with data from the Council of Europe\u0026rsquo;s resource on \u0026ldquo;cyberjustice and artificial intelligence\u0026rdquo; (Council of Europe, 2024). We assume that countries directly provided data for this latter source and that a certain bias to put the countries in the best of light could have resulted as a consequence. It is also noteworthy that we only relied on secondary sources about the existence and data of each programme. In order to minimise errors, two of the authors independently went through the data collection.\u003c/p\u003e\n\u003cp\u003eThe list of 28 countries was primarily derived from the intersection of countries for which we could find information for the different independent variables. We sought to maximise this dataset and, in a second step, eliminated countries whose socio-political structures were judged to be too different from the rest of the dataset, particularly in terms of democratic institutions (e.g. Azerbaijan, Morocco), since our theory is specifically concerned with advanced democracies.\u003c/p\u003e\n\u003cp\u003eThe criterion for inclusion was for a country itself to label the technology that they were using as \u0026ldquo;AI\u0026rdquo;. Regarding the lack of accepted consensus around a definition of AI\u003csup\u003e\u003csup\u003e[2]\u003c/sup\u003e\u003c/sup\u003e, this approach suited our research goal, as we are interested in \u003cem\u003ewhy\u003c/em\u003e countries would seek to present themselves as (early) adopters of AI. Following this, two specific types of projects were excluded, as they were systematically not presented under the \u0026ldquo;AI\u0026rdquo; label: projects related to risk-assessment-instruments, and those related to the automatic assignment of cases.\u003c/p\u003e\n\u003cp\u003eThe result of the data collection shows a rather homogeneous dataset, with only 5 countries with no implementation at all. 86% of the projects collected were in the context of courts, 10% for adjudication, and 5% for the overall justice system (note that one project was for both court and adjudication within the administration). Furthermore, most projects focused on anonymising court records for their release to the general public; the second most frequent category of projects concerned automatic speech-to-text transcription of court hearings (see Figure 1).\u003c/p\u003e\n\u003ch3\u003eIndependent variable\u003c/h3\u003e\n\u003cp\u003eWe used the three different \u0026nbsp;metrics and sources for each hypothesis, as listed in Table 1:\u003c/p\u003e\n\u003cp\u003eTable 1: Sources and metrics for hypotheses.\u003c/p\u003e\n\u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\" width=\"601\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 47px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eH#\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 55px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eYears\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 232px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eMetric\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 267px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eSource\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 47px;\"\u003e\n \u003cp\u003eH1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 55px;\"\u003e\n \u003cp\u003e2022\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 232px;\"\u003e\n \u003cp\u003eDisposition time: average duration to trial (in days) for criminal, civil, \u0026nbsp; \u0026nbsp; and administrative trials.\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 267px;\"\u003e\n \u003cp\u003eEuropean Commission for the Efficiency of Justice (2024)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 47px;\"\u003e\n \u003cp\u003eH2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 55px;\"\u003e\n \u003cp\u003e1997*-2024\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 232px;\"\u003e\n \u003cp\u003ePolitical position of a government, measured as \u0026ldquo;the proportion of cabinet posts of right-wing parties in relation to total cabinet posts, weighted by the number of days in office in a given year\u0026rdquo; (Armingeon et al., 2024).\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 267px;\"\u003e\n \u003cp\u003eComparative Political Data Set (Armingeon et al., 2024)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 47px;\"\u003e\n \u003cp\u003eH3\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 55px;\"\u003e\n \u003cp\u003e2022\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 232px;\"\u003e\n \u003cp\u003eE-government development index: Difference between a country\u0026rsquo;s index and the average across its neighbouring countries.\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 267px;\"\u003e\n \u003cp\u003eUnited Nations (2022)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003e*. 1997 corresponds to the first year that an AI project surfaces in the subset of the 28 selected countries.\u003c/p\u003e\n\u003cp\u003eThe dataset for H2 only included data up to 2022. As we carried out a survival analysis (see below), this was sufficient for all countries whose adoption of AI in the judiciary came before or in 2022. For 3 other countries (namely: Romania, Slovenia, Spain), we completed the dataset by adding the composition of right-wing members in the government manually following Armingeon et al. (2024) for the years 2023-2024.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eWe took the subset of countries for which data was available for each of the three introduced operationalisations. This led to slightly different datasets for H1/H3 and H2, with 25 countries considered in H1/H3 and 28 for H2 (the complete list is provided in Appendix A).\u003c/p\u003e\n\u003ch3\u003eData analysis\u003c/h3\u003e\n\u003cp\u003eFor H1 and H3, we carried out separate simple linear regressions including each of the five dependent variables, as this is a standard method within policy/political science to attempt at identifying potential relations between different sets of variables and is in line with how we posited the hypotheses (Chatterjee \u0026amp; Wiseman, 1983). This yielded a set of 5 linear regressions, with b\u003csub\u003e0\u003c/sub\u003e the intercept and b\u003csub\u003ei\u0026gt;0\u003c/sub\u003e the linear regression coefficient. In all models, we also control for the quality of government (World Bank, 2022) and GDP per capita (World Bank, 2025) by taking 2022 data to take into account previous evidence on the factors that may drive AI adoption in government (La Porta et al., 1999; Goel \u0026amp; Nelson, 2023):\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eY\u003csub\u003eyear\u003c/sub\u003e = b\u003csub\u003e0\u003c/sub\u003e + b\u003csub\u003e1\u003c/sub\u003eX\u003csub\u003eH1\u003c/sub\u003e + b\u003csub\u003e2\u003c/sub\u003eX\u003csub\u003eH2\u003c/sub\u003e + b\u003csub\u003e3\u003c/sub\u003eX\u003csub\u003eH3\u003c/sub\u003e + b\u003csub\u003eGovernmentQuality\u003c/sub\u003eX\u003csub\u003eGovernmentQuality\u003c/sub\u003e + b\u003csub\u003eGDPperCapita\u003c/sub\u003eX\u003csub\u003eGDPperCapita\u0026nbsp;\u003c/sub\u003e\u003c/p\u003e\n\u003cp\u003eY\u003csub\u003estage\u003c/sub\u003e = b\u003csub\u003e0\u003c/sub\u003e + b\u003csub\u003e1\u003c/sub\u003eX\u003csub\u003eH1\u003c/sub\u003e + b\u003csub\u003e2\u003c/sub\u003eX\u003csub\u003eH2\u003c/sub\u003e + b\u003csub\u003e3\u003c/sub\u003eX\u003csub\u003eH3\u003c/sub\u003e + b\u003csub\u003eGovernmentQuality\u003c/sub\u003eX\u003csub\u003eGovernmentQuality\u003c/sub\u003e + b\u003csub\u003eGDPperCapita\u003c/sub\u003eX\u003csub\u003eGDPperCapita\u003c/sub\u003e\u003c/p\u003e\n\u003cp\u003eY\u003csub\u003eaim\u003c/sub\u003e = b\u003csub\u003e0\u003c/sub\u003e + b\u003csub\u003e1\u003c/sub\u003eX\u003csub\u003eH1\u003c/sub\u003e + b\u003csub\u003e2\u003c/sub\u003eX\u003csub\u003eH2\u003c/sub\u003e + b\u003csub\u003e3\u003c/sub\u003eX\u003csub\u003eH3\u003c/sub\u003e + b\u003csub\u003eGovernmentQuality\u003c/sub\u003eX\u003csub\u003eGovernmentQuality\u003c/sub\u003e + b\u003csub\u003eGDPperCapita\u003c/sub\u003eX\u003csub\u003eGDPperCapita\u003c/sub\u003e\u003c/p\u003e\n\u003cp\u003eY\u003csub\u003earea\u003c/sub\u003e = b\u003csub\u003e0\u003c/sub\u003e + b\u003csub\u003e1\u003c/sub\u003eX\u003csub\u003eH1\u003c/sub\u003e + b\u003csub\u003e2\u003c/sub\u003eX\u003csub\u003eH2\u003c/sub\u003e + b\u003csub\u003e3\u003c/sub\u003eX\u003csub\u003eH3\u003c/sub\u003e + b\u003csub\u003eGovernmentQuality\u003c/sub\u003eX\u003csub\u003eGovernmentQuality\u003c/sub\u003e + b\u003csub\u003eGDPperCapita\u003c/sub\u003eX\u003csub\u003eGDPperCapita\u003c/sub\u003e\u003c/p\u003e\n\u003cp\u003eY\u003csub\u003epurpose\u003c/sub\u003e = b\u003csub\u003e0\u003c/sub\u003e + b\u003csub\u003e1\u003c/sub\u003eX\u003csub\u003eH1\u003c/sub\u003e + b\u003csub\u003e2\u003c/sub\u003eX\u003csub\u003eH2\u003c/sub\u003e + b\u003csub\u003e3\u003c/sub\u003eX\u003csub\u003eH3\u003c/sub\u003e + b\u003csub\u003eGovernmentQuality\u003c/sub\u003eX\u003csub\u003eGovernmentQuality\u003c/sub\u003e + b\u003csub\u003eGDPperCapita\u003c/sub\u003eX\u003csub\u003eGDPperCapita\u003c/sub\u003e\u003c/p\u003e\n\u003cp\u003eFurthermore, we conducted one-way ANOVA tests on the stage of the projects to try to evidence differences between the different groups, and in a second step, we merged the stage of projects so that they could fit within two groups only: those countries with no project started vs., those with an ongoing project \u003cem\u003eor\u003c/em\u003e implemented one. This allowed us to carry out one-tail t-tests.\u003c/p\u003e\n\u003cp\u003eFor H2, further in line with the hypothesis, we carried out an event history analysis, which is a well-suited form of analysis when the phenomenon under a study has a duration with an occurrence at a discrete point in time (so called \u0026ldquo;time-to-event data\u0026rdquo;), and it allows us to identify a likelihood of a variable to trigger an \u0026ldquo;event\u0026rdquo; (see for a similar data analysis method Steinbach et al., 2024). Traditionally, scholars have used it to investigate survivability (i.e., \u0026ldquo;impact of treatments on the event of death\u0026rdquo;), hence the function bearing the name of \u0026ldquo;survival function\u0026rdquo;. In our case, every country is \u0026ldquo;at risk\u0026rdquo; of starting an AI project in its judiciary and this represents therefore the event under consideration; once this has occurred, the country is then \u0026ldquo;censored\u0026rdquo; (removed) from the dataset for the subsequent years. The unit of analysis is therefore the country level, time periods are years, and all data are continuous as explained in the section above. We take as a starting year 1995, slightly before the year of the first project on the compiled list (1997 in Austria). This allows us to compute the probability of \u003cem\u003esurvival\u003c/em\u003e to the time t Pr(T\u003csub\u003ei\u003c/sub\u003e \u0026ge; t), with the following model description:\u0026nbsp;\u003c/p\u003e\n\u003cp\u003ey\u003csub\u003et,c\u003c/sub\u003e: binary component that captures whether a country has started an AI project in the judiciary\u003c/p\u003e\n\u003cp\u003et: time expressed in year\u003c/p\u003e\n\u003cp\u003ec: country\u003c/p\u003e\n\u003cp\u003eWe then compute a univariate Cox\u0026rsquo;s regression on the survival function for its statistical significance. In a second step, we repeated the event history analysis by adding as a treatment the aim followed behind kickstarting a project (whether for efficiency gain or to promote accessibility to the law).\u003c/p\u003e\n\u003cp\u003e\u003csup\u003e\u003csup\u003e[2]\u003c/sup\u003e\u003c/sup\u003e https://cacm.acm.org/opinion/between-the-booms-ai-in-winter/\u0026nbsp;\u003c/p\u003e"},{"header":"Results","content":"\u003ch3\u003eH1 and H3\u003c/h3\u003e\n\u003cp\u003eWe only find statistical significant results when considering the \u003cem\u003estage\u003c/em\u003e of the project; the other dependent variables did not result in statistically significant relations even if a Principal Component Analysis showed them as rather close together (see Figure 2), notably with \u003cem\u003estage\u003c/em\u003e very close to the area of implementation. When controlling for quality of government and GDP per capita, none of them have a significant effect on the dependent variables.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eFurthermore, within H1, we only obtained statistically significant results when considering duration of civil trials (see Table 2) and using one-tail t-tests with the stage of the projects clustered into 2 groups. Duration to criminal and administrative trials did not result in statistically significant results (see in Appendix B Table B.4), as well as a one-way ANOVA test was not statistically significant (\u003cem\u003eF\u003c/em\u003e=1.946, \u003cem\u003ep\u003c/em\u003e=0.161, df=24).\u003c/p\u003e\n\u003cp\u003eTable 2: Results are statistically significant for H1 and H3 when tested against the dependent variable \u0026ldquo;stage of the project\u0026rdquo;.\u003c/p\u003e\n\u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\" width=\"602\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 201px;\"\u003e\n \u003cp\u003e\u003cem\u003eN=25\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 201px;\"\u003e\n \u003cp\u003eH1 (duration to trial of \u003cem\u003ecivil\u0026nbsp;\u003c/em\u003ecases)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 201px;\"\u003e\n \u003cp\u003eH3 (adoption in bordering countries)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 201px;\"\u003e\n \u003cp\u003eLinear regression coefficient against \u003cem\u003estage\u0026nbsp;\u003c/em\u003e(p\u0026lt;0.01)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 201px;\"\u003e\n \u003cp\u003e0.003 (p\u0026lt;0.01, R\u003csup\u003e2\u003c/sup\u003e=0.28)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 201px;\"\u003e\n \u003cp\u003e8.051 (p\u0026lt;0.02, R\u003csup\u003e2\u003c/sup\u003e=0.21)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 201px;\"\u003e\n \u003cp\u003eT-stat (df=24)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 201px;\"\u003e\n \u003cp\u003e3.206 (p\u0026lt;0.03)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 201px;\"\u003e\n \u003cp\u003e1.904 (p\u0026lt;0.04)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003eTherefore, we can accept H1 and H3 within this set of conditions: the average number of days to trial in a country as well as the level of e-government adoption of bordering neighboring countries both significantly influence the decision of the government of a country to present itself as starting projects on AI in the judiciary.\u003c/p\u003e\n\u003ch3\u003eH2\u003c/h3\u003e\n\u003cp\u003eThe survival function for the risk of starting an AI project obtained on \u003cem\u003eN=\u003c/em\u003e28 countries is shown in Figure 3; the Cox\u0026rsquo;s regression returns a statistically significant relation (coefficient=0.277, p\u0026lt;0.01).\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eWe can therefore accept H2: the left-leaning governments are more likely to present their country as starting projects on AI in the judiciary. However, we did not find that the aim of the project (efficiency or accessibility) was statistically relevant in the likelihood of a left-wing (or right-wing) to initiate an AI project in the judiciary (see Appendix B, Table B.5).\u0026nbsp;\u003c/p\u003e"},{"header":"Discussion","content":"\u003cp\u003eWe can therefore draw three conclusions about the drivers behind the decision to implement AI in the judiciary in the context of advanced democracies. First, we find that the duration to trial of civil cases plays a role in deciding to turn to AI to alleviate the burden. More specifically, the average number of days to trial in civil cases for countries with an AI implementation averaged 310 whereas it averaged only 126 for those countries with such implementation. This confirms the perception of the role of technology to remove some of the burden on the state as it seeks to find room to operate within existing constraints (financial resources, political capital).\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eSecond, we find that the leaning of the government in power does have an influence on the starting AI projects in the judiciary, with left-wing governments more likely to do so than right-wing ones. Of the 23 countries with an implementation, only six were kickstarted by a right-wing government (Canada, Czech Republic, Estonia, France, Latvia, Switzerland) and conversely, and 17 by a left-wing government (Austria, Croatia, Germany, Spain, Malta, Romenia, Slovenia, Denmark, Finland, Italy, Luxembourg, Netherlands, Poland, Portugal, Sweden, UK, USA). Also strengthening the conclusion is that of the five countries without an implementation (Belgium, Cyprus, Hungary, Lithuania, Norway), all but one country (Hungary) did go through governments with a different majority across the 26 years under study. A closer look at the countries in our sample reveals no consistent pattern in socio-political or economic structures that would account for the observed variation. Both lists\u0026mdash;those with and without AI initiatives in the judiciary\u0026mdash;include a mix of social democracies and economically liberal states, centralised and federal systems, high- and low-GDP countries, large and small populations, as well as EU and non-EU members. This diversity suggests that these structural factors are unlikely to explain the difference. Instead, the pattern appears to align more closely with an ideological inclination, specifically a greater propensity among left-leaning governments to pursue such initiatives.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eWe therefore interpret this result in light of the agenda of left-wing governments to give more access to a country\u0026rsquo;s residents to state services, even beyond the judiciary, regardless of the aim when starting the AI project. The mixture of projects started actually falls under both aims of making the state more efficient, \u003cem\u003eand\u003c/em\u003e to promote access to the law (e.g. with the single highest number of projects in the category of automatically anonymising court proceedings to publish them online, often explicitly to foster legal access), and this explains why the addition of the aim as a treatment in the event history analysis did not play a role. Yet, we contend that even projects geared to make the judiciary more efficient contributes to a more responsive state, and hence to one that can better fulfil its promise and mandate towards a country\u0026rsquo;s inhabitants.\u003c/p\u003e\n\u003cp\u003eOur third and last finding concerns the role of governments in bordering neighbours in bearing an influence, although the argument may be a bit circular. Those countries with a judicial AI implementation have, in average, a positive difference between their e-government index and their neighbours\u0026rsquo;; those other countries without a judicial AI implementation have, in average, a negative difference. In other words, that means that those which have implemented judicial AI tools have generally a better readiness to implement access to public services via digital means, and those without lag behind their border\u0026rsquo;s peers when it comes to e-government.\u003c/p\u003e"},{"header":"Limitations and Conclusion","content":"\u003cp\u003eIn conducting this study, we have faced a number of methodological hurdles probably due to the exploratory nature of the research question. With no clear framework to follow, and with no clear set of countries to focus on, we have iterated a number of times between frameworks, country sets, dependent/independent variables and data sources available before we could find meaningful and statistically relevant results (a few of the negative results of our investigation can be found in Appendix B). Our initial methodology was more traditional to comparative politics by picking three countries and qualitatively analysing them in-depth: two countries with judicial AI implementation, and a counter-case of a country without one. However, by seeking to generalise the hypotheses to a larger \u003cem\u003eN\u003c/em\u003e, we did not succeed. In line with these first negative results, we would like to caution on the interpretation of our findings notably when it comes to extrapolating them outside of the dataset of countries under consideration. The sheer number and complexity of drivers possibly involved means that generalisation should be refrained.\u003c/p\u003e\n\u003cp\u003eSimilarly, the results reported here have a certain number of limitations \u003cem\u003eper hypothesis\u003c/em\u003e too. First regarding H1, we could not find any statistically significant result when looking at the waiting time to trial in the cases of criminal and administrative cases (even when trying with different statistical tools, from Pearson\u0026rsquo;s correlation, to Principal Component Analysis, to one-way ANOVA). There might be a rationale explanation to this: when collecting the data, it has come to our attention that a few countries have explicitly referred to focusing on implementation of AI tools in the civil law domain, and it is hence sensical that we find this mirrored as to why they chose to focus on this sub-domain. Furthermore, countries\u0026rsquo; focus on civil rather than criminal cases, especially as a first foray to deploy AI tools in the judiciary may be explained by how resolving procedural issues or small claims is generally less sensitive than criminal proceedings (Bell et. Al., 2022, Fine \u0026amp; Marsh, 2024). The higher stakes in criminal proceedings could therefore play a role, an argument which is also valid when looking at administrative justice where significant ethical and legal risks also arise, including concerns about the dehumanization of justice, real threats to the independence and impartiality of justice (Nouri et. Al., 2024). The many scandals which have arisen concerning using AI in public administration to make decisions quicker about social benefits (see from the Netherlands, to France, to the US, Tam\u0026ograve;-Larrieux et al., 2025) appear to give credence to the argument about stakes at play\u0026mdash;all the while highlighting the many pitfalls which exist when deploying AI tools in general (Guitton et al., 2024) and in the judiciary in particular (John et al, 2023). It remains that future qualitative studies using process tracing and (semi-) structured interviews with key decision-makers would be needed to confirm how the decision process was shaped\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eSecond, regarding H2, it may be argued that we only looked at the leaning of the majority in government as opposed to looking at the leaning of the minister of justice most likely in charge. The different datasets we considered (e.g. the Manifesto Project Dataset, Lehmann et al., 2024) also did not include this information, and we accept that this may skew the results, albeit only to a certain limited extent. Also noteworthy concerning H2 is that we could not find any relation when running the historical event analysis with as additional covariate the aim of the project (efficiency / accessibility); we also found no relation when looking at odds-ratio and regressions. We cannot therefore conclude that the leaning of the party in government has an influence on the type of project started, and we can \u0026ldquo;only\u0026rdquo; fail to accept H2 without making a distinction for the aims of projects.\u003c/p\u003e\n\u003cp\u003eLastly, regarding H3, we note that the operationalisation matters. We found no statistical significance in the following cases: when we only took the average (or standard deviation) between a country and its bordering neighbours\u0026rsquo; e-government index, when we took only the e-government index without consideration of neighbours (the e-government index is supposed to be a relative assessment of countries, not an absolute measure), or when we defined neighbours more coarsely in terms of sub-continent (Northern Europe, Northern America, etc.). While we posit that our current operationalisation is the one that makes the most sense (by looking at the difference, we actually look at how the country differs in relation to its neighbours), we also acknowledge the difficulty in the fragility of the conclusion dependent on the definition of the metric used to assess it.\u003c/p\u003e\n\u003cp\u003eYet another challenge that we have experienced\u0026mdash;on top of defining a subset of country and data availability\u0026mdash;concerned the selection of the independent variable for a model. We started with a lot more hypotheses, but for the sake of coherence and shortness, we decided to present a model based only on the three hypotheses reported here. Other hypotheses for which we collected data and found no meaningful relations were: trust in AI, R\u0026amp;D spending, the degree of access to AI expertise, regulation concerning AI in the judiciary, the independence and corruption of the judiciary, and distance to power. We notably acknowledge that our initial thinking was, if not that these variables correlate to some extent with AI adoption in the judiciary, that they might be otherwise confounding variables. For each of them, however, we have either found no relation worth reporting, or we faced challenges to match with our subset of countries at the intersection of our other datasets.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eDespite these limitations, we still posit that the findings offer insights into AI adoption in the judiciary amidst the hype and craze that surrounds the topic of AI. By looking at data, we obtain different insights than what qualitative interviews can return on both adoption or non adoption (Druță et al., 2024) and offer a complementary picture. Whether decision-makers are aware that they follow certain drivers\u0026mdash;be it on their perception of the state capacity as overburdened, of their own political party\u0026rsquo;s agenda, or of how competitive they want to be against their neighbours\u0026mdash;, the research teases out the relevance of the factors nonetheless. Future research could consider finding relations with further factors\u0026mdash;although it may be well noted again to consider that we did not find any when looking at trust in AI, R\u0026amp;D spending, the degree of access to AI expertise, regulation concerning AI in the judiciary, corruption, independence of the judiciary, and distance to power. Regarding the wealth of possible factors playing a role (e.g. role of international organisations, trust in the police vs trust in the courts, etc.), and as already highlighted in the literature review section (\u0026ldquo;Theorising Adoption\u0026rdquo;), there should not be a lack of material to collect and test.\u003c/p\u003e\n\u003cp\u003eFuture research should also further investigate whether these findings apply to AI use in other branches of government, as current research in this area remains limited. For example, research on public perceptions of AI in policing indicates the influence of political partisanship: Republican members of the public tend to favor predictive policing over automated internal reviews, whereas Democrats show higher support for both (Schiff et al., 2025). Additionally, a few studies have similarly suggested how e-government initiatives could support AI adoption for crime management (Gummadidala et al., 2020), and when fighting corruption and the shadow economy (Goel \u0026amp; Saunoris, 2016). It would be valuable to further explore how a country\u0026apos;s political ideology influences AI implementation in law enforcement\u0026mdash;not only in the judiciary\u0026mdash;and examine whether neighboring e-government initiatives similarly impact AI adoption. Investigating these factors across different government branches, such as policing, could offer deeper insights into the key drivers of AI adoption and the role of context in shaping policies.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003eEthics, Consent to Participate, and Consent to Publish declarations\u003c/p\u003e\n\u003cp\u003eNot applicable.\u003c/p\u003e\n\u003cp\u003eAuthor contributions statement\u003c/p\u003e\n\u003cp\u003eCG wrote the main manuscript and carried out the data analysis; VD and MH wrote the theory and hypotheses; CG and VD collected the data; ATM and SM gave directions to the research; all contributed to the conceptualisation of the research and all reviewed the manuscript.\u003c/p\u003e\n\u003cp\u003eFunding declaration\u003c/p\u003e\n\u003cp\u003eNot applicable.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n \u003cli\u003eAhmed, M. A., Kaya, T., \u0026amp; Karanfiller, T. (2024). Evaluating E-court systems in regional governments in developing countries using technology acceptance model. \u003cem\u003eInformation Development\u003c/em\u003e, https://doi.org/10.1177/02666669241229176\u003c/li\u003e\n \u003cli\u003eArmingeon, Klaus, Engler, S., Leemann, L., \u0026amp; Weisstanner, D. (2024). Comparative Political Data Set 1960-2022. Zurich/Lueneburg/Lucerne: University of Zurich, Leuphana University Lueneburg, and University of Lucerne.\u003c/li\u003e\n \u003cli\u003eBailey, J., \u0026amp; Burkell, J. (2013). Implementing technology in the justice sector: A Canadian perspective. \u003cem\u003eCanadian Journal of Law and Technology\u003c/em\u003e, 11(2).\u003c/li\u003e\n \u003cli\u003eBaker, T., Falco Metcalfe, C., Berenblum, T., Aviv, G., \u0026amp; Gertz, M. (2015). Examining Public Preferences for the Allocation of Resources to Rehabilitative Versus Punitive Crime Policies. \u003cem\u003eCriminal Justice Policy Review\u003c/em\u003e, 26(5), 448-462.\u003c/li\u003e\n \u003cli\u003eBarysė, D. (2022). People\u0026rsquo;s attitudes towards technologies in courts. \u003cem\u003eLaws\u003c/em\u003e, 11(5), 71.\u003c/li\u003e\n \u003cli\u003eBell, F., Moses, L.B., Legg, M., Silove, J., \u0026amp; Zalnieriute, M. (2022). \u003cem\u003eAI decision-making and the courts: A guide for judges, tribunal members and court administrators.\u003c/em\u003e Sydney: Australasian Institute of Judicial Administration Incorporated.\u003c/li\u003e\n \u003cli\u003eBremer, B., \u0026amp; McDaniel, S. (2020). The ideational foundations of social democratic austerity in the context of the Great Recession. \u003cem\u003eSocio-Economic Review\u003c/em\u003e, \u003cem\u003e18\u003c/em\u003e(2), 439\u0026ndash;463.\u003c/li\u003e\n \u003cli\u003eBussell, J. (2011). Explaining cross-national variation in government adoption of new technologies. \u003cem\u003eInternational Studies Quarterly\u003c/em\u003e, \u003cem\u003e55\u003c/em\u003e(1), 267-280.\u003c/li\u003e\n \u003cli\u003eChatterjee, S., \u0026amp; Wiseman, F. (1983). Use of regression diagnostics in political science research. \u003cem\u003eAmerican Journal of Political Science\u003c/em\u003e, \u003cem\u003e27\u003c/em\u003e(3), 601-613.\u003c/li\u003e\n \u003cli\u003eCouncil of Europe (2024). Resource Centre Cyberjustice and AI. https://public.tableau.com/app/profile/cepej/viz/ResourceCentreCyberjusticeandAI/AITOOLSINITIATIVESREPORT?publish=yes\u003c/li\u003e\n \u003cli\u003eDavis, F. D. (1989). Technology acceptance model: TAM. \u003cem\u003eAl-Suqri, MN, Al-Aufi, AS: Information Seeking Behavior and Technology Adoption\u003c/em\u003e, \u003cem\u003e205\u003c/em\u003e, 219.\u003c/li\u003e\n \u003cli\u003eDruță, V., Tamo-Larrieux, A., Guitton, C., \u0026amp; van Dijck, G. (2024, August). From Analog Risk Assessment Instruments to Digital Ones: Mapping the Concerns. In \u003cem\u003eProceedings of Ongoing Research, Practitioners, Posters, Workshops, and Projects of the International Conference EGOV-CeDEM-ePart 2024\u003c/em\u003e.\u003c/li\u003e\n \u003cli\u003eEmon, M. M. H. (2023). Insights Into Technology Adoption: A Systematic Review of Framework, Variables and Items. \u003cem\u003eInformation Management and Computer Science\u003c/em\u003e, \u003cem\u003e6\u003c/em\u003e(2), 27-33\u003c/li\u003e\n \u003cli\u003eEuropean Commission for the Efficiency of Justice (2024). https://public.tableau.com/app/profile/cepej/viz/EfficiencyEN/Efficiency\u003c/li\u003e\n \u003cli\u003eFern\u0026aacute;ndez‐i‐Mar\u0026iacute;n, X., Hinterleitner, M., Knill, C., \u0026amp; Steinebach, Y. (2024a). \u0026ldquo;Bureaucratic Overburdening in Advanced Democracies.\u0026rdquo; \u003cem\u003ePublic Administration Review\u003c/em\u003e, 84 (4), 696-709.\u003c/li\u003e\n \u003cli\u003eFine, A., \u0026amp; Marsh, S. (2024). Judicial leadership matters (yet again): the association between judge and public trust for artificial intelligence in courts. \u003cem\u003eDiscover Artificial Intelligence\u003c/em\u003e, \u003cem\u003e4\u003c/em\u003e(1), 44.\u003c/li\u003e\n \u003cli\u003eGilardi, F. (2016). \u0026ldquo;Four Ways we Can Improve Policy Diffusion Research.\u0026rdquo; \u003cem\u003eState Politics \u0026amp; Policy Quarterly\u003c/em\u003e 16(1): 8\u0026ndash;21.\u003c/li\u003e\n \u003cli\u003eGoel, R. K., \u0026amp; Nelson, M. A. (2023). \u003cem\u003eAwareness of artificial intelligence: Diffusion of information about AI versus ChatGPT in the United States\u003c/em\u003e (No. 2259). Kiel Working Paper.\u003c/li\u003e\n \u003cli\u003eGoel, R. K., \u0026amp; Saunoris, J. W. (2016). Virtual versus physical government decentralization: Effects on corruption and the shadow economy. \u003cem\u003ePublic Budgeting \u0026amp; Finance\u003c/em\u003e, \u003cem\u003e36\u003c/em\u003e(4), 68-93.\u003c/li\u003e\n \u003cli\u003eGrimmelikhuijsen, S., \u0026amp; Tangi, L. (2024). What factors influence perceived artificial intelligence adoption by public managers?, \u003cem\u003eEuropean Commission\u003c/em\u003e, JRC138684. https://data.europa.eu/doi/10.2760/0179285.\u003c/li\u003e\n \u003cli\u003eGummadidala, P. R., Karippur, N. K., \u0026amp; Koilakuntla, M. (2020). Analysis of factors influencing the adoption of artificial intelligence for crime management. In \u003cem\u003eRe-imagining Diffusion and Adoption of Information Technology and Systems: A Continuing Conversation: IFIP WG 8.6 International Conference on Transfer and Diffusion of IT, TDIT 2020,\u0026nbsp;\u003c/em\u003eTiruchirappalli, India, December 18\u0026ndash;19, 2020, Proceedings, Part I (pp. 3-9). Springer International Publishing.\u003c/li\u003e\n \u003cli\u003eGuitton, C., Mayer, S., Tamo-Larrieux, A., Van Landuyt, D., Fosch-Villaronga, E., Kamara, I., \u0026amp; Pałka, P. (2024). Responsible automatically processable regulation. \u003cem\u003eAI \u0026amp; SOCIETY\u003c/em\u003e, 1-16.\u003c/li\u003e\n \u003cli\u003eGuitton, C., Tam\u0026ograve;-Larrieux, A., \u0026amp; Mayer, S. (2022). A typology of automatically processable regulation. \u003cem\u003eLaw, Innovation and Technology\u003c/em\u003e, \u003cem\u003e14\u003c/em\u003e(2), 267-304.\u003c/li\u003e\n \u003cli\u003eHacker, JS., \u0026amp; Pierson, P. (2014). \u0026ldquo;After the \u0026lsquo;Master Theory\u0026rsquo;: Downs, Schattschneider, and the Rebirth of Policy-Focused Analysis.\u0026rdquo; \u003cem\u003ePerspectives on Politics\u003c/em\u003e, 12(03), 643\u0026ndash;662.\u003c/li\u003e\n \u003cli\u003eHäusermann, S., Garritzmann, J., \u0026amp; Palier, B. (2022). The politics of social investment. In Julian L. Garritzmann, Silja Häusermann, and Bruno Palier (Eds.), \u003cem\u003eThe world politics of social investment\u003c/em\u003e. Oxford University Press.\u003c/li\u003e\n \u003cli\u003eHerd, P., \u0026amp; Moynihan, DP (2019). \u003cem\u003eAdministrative burden: Policymaking by other means\u003c/em\u003e. Russell Sage Foundation.\u003c/li\u003e\n \u003cli\u003eHinterleitner, M., Knill, C., \u0026amp; Steinebach, Y. (2024). \u0026ldquo;The Growth of Policies, Rules, and Regulations: A Review of the Literature and Research Agenda\u0026rdquo;, \u003cem\u003eRegulation \u0026amp; Governance\u003c/em\u003e, 18 (2), 637-654.\u003c/li\u003e\n \u003cli\u003eIlieva, G., Yankova, T., Ruseva, M., Dzhabarova, Y., Zhekova, V., Klisarova-Belcheva, S., ... \u0026amp; Dimitrov, A. (2024). Factors Influencing User Perception and Adoption of E-Government Services. \u003cem\u003eAdministrative Sciences\u003c/em\u003e, \u003cem\u003e14\u003c/em\u003e(3), 54.\u003c/li\u003e\n \u003cli\u003eIngrams, A., Manoharan, A., Schmidthuber, L., \u0026amp; Holzer, M. (2020). Stages and determinants of e-government development: A twelve-year longitudinal study of global cities.\u003cem\u003e\u0026nbsp;International Public Management Journal, 23\u003c/em\u003e(6), 731-769.\u003c/li\u003e\n \u003cli\u003eJakobsen, MLF., and Mortensen, PB (2015). \u0026ldquo;How politics shapes the growth of rules.\u0026rdquo; \u003cem\u003eGovernance\u003c/em\u003e, \u003cem\u003e28\u003c/em\u003e, 497\u0026ndash;515.\u003c/li\u003e\n \u003cli\u003eJohn, A. M., Aiswarya, M. U., \u0026amp; Panachakel, J. T. (2023, October). Ethical challenges of using artificial intelligence in judiciary. In \u003cem\u003e2023 IEEE International Conference on Metrology for eXtended Reality, Artificial Intelligence and Neural Engineering (MetroXRAINE)\u003c/em\u003e (pp. 723-728). IEEE.\u003c/li\u003e\n \u003cli\u003eKelly, S., Kaye, S. A., \u0026amp; Oviedo-Trespalacios, O. (2023). What factors contribute to the acceptance of artificial intelligence? A systematic review. \u003cem\u003eTelematics and Informatics\u003c/em\u003e, \u003cem\u003e77\u003c/em\u003e, 101925.\u003c/li\u003e\n \u003cli\u003eKnill, C., \u0026amp; Tosun, J. (2020). \u003cem\u003ePublic policy: A new introduction\u003c/em\u003e. Bloomsbury Publishing.\u003c/li\u003e\n \u003cli\u003eLegal technology acceptance in Vietnam\u0026rsquo;s courts. Cogent Business \u0026amp; Management, 11(1), 2321955.\u003c/li\u003e\n \u003cli\u003eLehmann, P., Franzmann, S., Al-Gaddooa, D., Burst, T., Ivanusch, C., Regel, S., Riethm\u0026uuml;ller, F., Volkens, A., We\u0026szlig;els, B., Zehnter, L. (2024): The Manifesto Data Collection. Manifesto Project (MRG/CMP/MARPOR). Version 2024a. Berlin: Wissenschaftszentrum Berlin f\u0026uuml;r Sozialforschung (WZB) / G\u0026ouml;ttingen: Institut f\u0026uuml;r Demokratieforschung (IfDem). https://doi.org/10.25522/manifesto.mpds.2024a\u003c/li\u003e\n \u003cli\u003eMadan, R., \u0026amp; Ashok, M. (2023). AI adoption and diffusion in public administration: A systematic literature review and future research agenda. \u003cem\u003eGovernment Information Quarterly\u003c/em\u003e, \u003cem\u003e40\u003c/em\u003e(1), 101774.\u003c/li\u003e\n \u003cli\u003eMaggetti, M., \u0026amp; Gilardi, F. (2016). \u0026ldquo;Problems (and solutions) in the measurement of policy diffusion mechanisms.\u0026rdquo; \u003cem\u003eJournal of Public Policy\u003c/em\u003e, 36(1), 87-107.\u003c/li\u003e\n \u003cli\u003eMarsh, D., \u0026amp; Sharman, JC. (2009). \u0026ldquo;Policy diffusion and policy transfer.\u0026rdquo; \u003cem\u003ePolicy Studies\u003c/em\u003e, 30(3), 269-288.\u003c/li\u003e\n \u003cli\u003eMartinuzzi, A. (2017). Taking Justice Seriously: The Problem of Courts Overload and the New Model of Judicial Process (April 6, 2017). \u003cem\u003eCivil Procedure Review\u003c/em\u003e, 8(1), 65-106.\u003c/li\u003e\n \u003cli\u003eMeseguer, C.. 2005. \u0026ldquo;Policy learning, policy diffusion, and the making of a new order.\u0026rdquo; \u003cem\u003eThe Annals of the American Academy of Political and Social Science\u003c/em\u003e, 598(1), 67-82.\u003c/li\u003e\n \u003cli\u003eMuigua, K. (2020). Legal Practice and New Frontiers: Embracing Technology for Enhanced Efficiency and Access to Justice. \u003cem\u003eADR Journal, 11\u003c/em\u003e(2).\u003c/li\u003e\n \u003cli\u003eNguyen, N. A. D., Nguyen, V. P., \u0026amp; Bui, K. H. (2024). Legal technology acceptance in Vietnam\u0026rsquo;s courts. \u003cem\u003eCogent Business \u0026amp; Management\u003c/em\u003e, \u003cem\u003e11\u003c/em\u003e(1), 2321955.\u003c/li\u003e\n \u003cli\u003eNouri, Z., Salah, W. B., \u0026amp; AlOmran, N. (2024). Artificial Intelligence and Administrative Justice: An Analysis of Predictive Justice in France. \u003cem\u003eHasanuddin Law Review\u003c/em\u003e, \u003cem\u003e10\u003c/em\u003e(2), 119-143.\u003c/li\u003e\n \u003cli\u003eOECD (2024), \u0026ldquo;Governing with Artificial Intelligence: Are governments ready?\u0026rdquo;, \u003cem\u003eOECD Artificial Intelligence Papers\u003c/em\u003e, No. 20, OECD Publishing, Paris, https://doi.org/10.1787/26324bc2-en.\u003c/li\u003e\n \u003cli\u003ePierson, P. (2004). \u003cem\u003ePolitics in Time\u003c/em\u003e. Princeton: Princeton University Press.\u003c/li\u003e\n \u003cli\u003eRana, N. P., Dwivedi, Y. K., Lal, B., Williams, M. D., \u0026amp; Clement, M. (2017). Citizens\u0026rsquo; adoption of an electronic government system: towards a unified view. \u003cem\u003eInformation systems frontiers\u003c/em\u003e, \u003cem\u003e19\u003c/em\u003e, 549-568.\u003c/li\u003e\n \u003cli\u003eRen, L., Zhao, J., and Lovrich, NP. (2008). Liberal versus conservative public policies on crime: What was the comparative track record during the 1990s?, \u003cem\u003eJournal of Criminal Justice, 36\u003c/em\u003e(4), 316-325.\u003c/li\u003e\n \u003cli\u003eSchiff, K. J., Schiff, D. S., Adams, I. T., McCrain, J., \u0026amp; Mourtgos, S. M. (2025). Institutional factors driving citizen perceptions of AI in government: Evidence from a survey experiment on policing. \u003cem\u003ePublic Administration Review\u003c/em\u003e, \u003cem\u003e85\u003c/em\u003e(2), 451-467.\u003c/li\u003e\n \u003cli\u003eSchiff, Daniel S., \u0026amp; Schiff, KJ. (2023). \u0026ldquo; Narratives and Expert Information in Agenda-setting: Experimental Evidence on State Legislator Engagement With Artificial Intelligence Policy.\u0026rdquo; \u003cem\u003ePolicy Studies Journal\u003c/em\u003e 51(4): 817\u0026ndash;842.\u003c/li\u003e\n \u003cli\u003eSimmons, Beth A., \u0026amp; Elkins, Z. (2004). \u0026ldquo;The Globalization of Liberalization: Policy Diffusion in the International Political Economy.\u0026rdquo; \u003cem\u003eAmerican Political Science Review\u003c/em\u003e 98(1): 171\u0026ndash;189.\u003c/li\u003e\n \u003cli\u003eSousa, M., Kettiger, D., \u0026amp; Lienhard, A. (2022). E-justice in Switzerland and Brazil: Paths and Experiences. In \u003cem\u003eIJCA\u003c/em\u003e (Vol. 13, p. 1).\u003c/li\u003e\n \u003cli\u003eSteinebach, Y., Hinterleitner, M., \u0026amp; Fern\u0026aacute;ndez‐i‐Mar\u0026iacute;n, X. (2024). Regulatory offsetting in advanced democracies. \u003cem\u003ePublic Administration Review\u003c/em\u003e.\u003c/li\u003e\n \u003cli\u003eStr\u0026oslash;m, Kare (1990). \u0026ldquo;A behavioral theory of competitive political parties.\u0026rdquo; \u003cem\u003eAmerican Journal of Political Science\u003c/em\u003e 34(2): 565\u0026ndash;598.\u003c/li\u003e\n \u003cli\u003eTam\u0026ograve;-Larrieux, A., Guitton, C., \u0026amp; Mayer, S. (2025). \u003cem\u003eAI and Law: How Automation is Changing the Law\u003c/em\u003e. CRC Press.\u003c/li\u003e\n \u003cli\u003eTangi, L., Van Noordt, C., Combetto, M., Gattwinkel, D. and Pignatelli, F., (2022). AI Watch. European landscape on the use of Artificial Intelligence by the Public Sector, \u003cem\u003ePublications Office of the European Union\u003c/em\u003e, doi:10.2760/39336, JRC129301.\u003c/li\u003e\n \u003cli\u003eTsebelis, G. (2002). \u003cem\u003eVeto players: How political institutions work\u003c/em\u003e. Princeton University Press.\u003c/li\u003e\n \u003cli\u003eUnited Nations (2022). E-Government Development Index. https://publicadministration.un.org/egovkb/en-us/About/Overview/-E-Government-Development-Index\u003c/li\u003e\n \u003cli\u003eVenkatesh, V., Morris, M.G., Davis, G.B., Davis, F.D. (2003). User acceptance of information technology: Toward a unified view. \u003cem\u003eMIS Quarterly\u003c/em\u003e, \u003cem\u003e27\u003c/em\u003e(3), 425\u0026ndash;478.\u003c/li\u003e\n \u003cli\u003eVenkatesh, V., Thong, J.Y., Xu, X. (2016). Unified theory of acceptance and use of technology: A synthesis and the road ahead. J\u003cem\u003eournal of the association for Information Systems,\u003c/em\u003e 17(5), 328\u0026ndash;376.\u003c/li\u003e\n \u003cli\u003eVerkijika, S. F., \u0026amp; De Wet, L. (2018). E-government adoption in sub-Saharan Africa. \u003cem\u003eElectronic Commerce Research and Applications\u003c/em\u003e, \u003cem\u003e30\u003c/em\u003e, 83-93.\u003c/li\u003e\n \u003cli\u003eWang, R. (2020). Legal technology in contemporary USA and China. \u003cem\u003eComputer Law \u0026amp; Security Review, 39\u003c/em\u003e, 105459.\u003c/li\u003e\n \u003cli\u003eWorld Bank (2022). Worldwide Governance Indicators, 2024 Update. https://www.worldbank.org/en/publication/worldwide-governance-indicators, accessed on March 25, 2025.\u003c/li\u003e\n \u003cli\u003eWorld Bank (2025, March 24). GDP per capita (current US$). https://data.worldbank.org/indicator/NY.GDP.PCAP.CD, accessed on March 25, 2025.\u003c/li\u003e\n \u003cli\u003eXu, N., \u0026amp; Wang, K. J. (2021). Adopting robot lawyer? The extending artificial intelligence robot lawyer technology acceptance model for legal industry by an exploratory study. \u003cem\u003eJournal of Management \u0026amp; Organization\u003c/em\u003e, \u003cem\u003e27\u003c/em\u003e(5), 867-885.\u003cstrong\u003e\u003c/strong\u003e\u003c/li\u003e\n\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":true,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"discover-artificial-intelligence","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"diai","sideBox":"Learn more about [Discover Artificial Intelligence](https://www.springer.com/44163)","snPcode":"","submissionUrl":"","title":"Discover Artificial Intelligence","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Discover Series","inReviewEnabled":true,"inReviewRevisionsEnabled":true},"keywords":"","lastPublishedDoi":"10.21203/rs.3.rs-5881593/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-5881593/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eArtificial Intelligence (AI) is increasingly used worldwide to make decisions, be it by public administrations, industry, banks, or insurers. One area with a particularly high impact on citizens is its use in judiciary processes. To this day, there has been too little investigation into what drives the willingness of a country to present itself as adopting AI in its judiciary. In this article, we show that the mere promise of efficiency gains is not enough of an explanation for the claim of adopting AI within judiciary processes. We show that the administrative burden (as measured by the number of days to trial), a government\u0026rsquo;s leaning within the political spectrum (namely towards the left), and the level of adoption of technology by governments in bordering neighboring countries predict countries\u0026rsquo; announcement to adopt AI in their judiciary. Together, our findings from a data analysis of 28 advanced democracies form a theory of adoption respectively around functionality, politics, and diffusion of AI in the judiciary.\u003c/p\u003e","manuscriptTitle":"Adoption of Artificial Intelligence in the Judiciary: A Comparison of 28 Advanced Democracies","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-04-24 11:08:43","doi":"10.21203/rs.3.rs-5881593/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"decision","content":"Accepted","date":"2025-05-13T11:37:30+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2025-05-12T07:01:57+00:00","index":"","fulltext":""},{"type":"reviewerAgreed","content":"126126333235233011464588882283896453701","date":"2025-04-26T14:44:59+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2025-04-25T17:47:38+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2025-04-25T15:48:08+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"300745236679042074101503011548902378957","date":"2025-04-25T15:44:06+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2025-04-25T14:32:21+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"102100476319596674134432661428602634414","date":"2025-04-25T14:28:22+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"180016544556102034617160858983415814444","date":"2025-04-25T13:36:22+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2025-04-23T10:31:01+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2025-04-22T08:43:27+00:00","index":"","fulltext":""},{"type":"submitted","content":"Discover Artificial Intelligence","date":"2025-04-03T20:27:12+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"discover-artificial-intelligence","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"diai","sideBox":"Learn more about [Discover Artificial Intelligence](https://www.springer.com/44163)","snPcode":"","submissionUrl":"","title":"Discover Artificial Intelligence","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Discover Series","inReviewEnabled":true,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"452d9bf2-f029-481a-b289-fe89798b998e","owner":[],"postedDate":"April 24th, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"published-in-journal","subjectAreas":[],"tags":[],"updatedAt":"2025-07-28T14:02:35+00:00","versionOfRecord":{"articleIdentity":"rs-5881593","link":"https://doi.org/10.1007/s44163-025-00311-y","journal":{"identity":"discover-artificial-intelligence","isVorOnly":false,"title":"Discover Artificial Intelligence"},"publishedOn":"2025-07-24 00:00:00","publishedOnDateReadable":"July 24th, 2025"},"versionCreatedAt":"2025-04-24 11:08:43","video":"","vorDoi":"10.1007/s44163-025-00311-y","vorDoiUrl":"https://doi.org/10.1007/s44163-025-00311-y","workflowStages":[]},"version":"v1","identity":"rs-5881593","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-5881593","identity":"rs-5881593","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}