A ChatGPT-Generated Taxonomy of Risky Real-Life Situations

With the recent emergence of large language models (e.g., ChatGPT) trained on enormous amounts of information regarding human affairs it is now possible to easily explore naturalistic environments by digital means. Here, this technology was put to to use by having ChatGPT 4.0 identify everyday decisions that can involve risk. Risk was defined according to four different and commonly used definitions of risk (Hansson, 2023); (i) the cause of an unwanted event that may or may not occur, (ii) an unwanted event that may or may not occur, (iii) the probability of an unwanted event that may or may not occur, and (iv) the statistical expectation value of an unwanted event that may or may not occur. After, several conversation reiterations over 1,000 decisions were identified, and after redundancy actions (vectorization, word embedding techniques, and manual review) over 350 situations remained .