AI-Simulated Emotions for Lawmaking: A Useful Approach?

This paper addresses AI-simulated emotions related to those affected by a law reform aiming to tighten the requirements for permanent residence in Finland. Emotionally laden perspectives on the reform are derived from simulated consultations produced by generative artificial intelligence (GenAI). The rationale for this method is that those seeking permanent residence represent a group typically excluded from participatory approaches in lawmaking, despite being severely impacted by laws that define their status as residents. Addressing the impacts of such laws on their lives would ideally require carefully designed interviews; however, their execution would necessitate knowledge, skills, and resources that an average civil servant is unlikely to possess. Firstly, we study the potential usefulness of simulated consultations regarding the position of those affected: the nature of the insights received, and their significance as ex ante impact assessments compared to the contributions of actual open consultations. Secondly, we examine the epistemic nature of the simulated perspectives: the nature of the knowledge provided and its relation to epistemic injustice. The main data set derived from GenAI is based on thorough experimentation with different 1) LLM applications, 2) profiles of immigrants, and 3) designs of prompts, including key elements of the law proposal, questions asked in the simulation, and general instructions for the tone of the 'consultation.' After the experimentation, we constructed seven profiles of low-income jobs, as individuals in these roles are particularly affected by the reform. The profiles also include different genders, as well as various national and family backgrounds. The final data consists of six sets of simulations, each containing five sets of question-answer pairs from the simulated participants (210 question-answer adjacency pairs). A complementary data set comprises actual comments provided during the publicly held open consultations. The comparison of the two sets of data reveals that the simulated consultations produce more down-to-earth and emotionally laden perspectives on probable impacts: justifiable and credible fears, worries, and frustrations, as well as feelings of unjust treatment of people in low-income jobs compared to those in high-income jobs. In line with existing research, such perspectives can have a notable impact on the integration of immigrants. Thus, these emotionally laden perspectives are also linked to macro-level impacts. As to whether the perspectives derived are valid knowledge for impact assessment, it is important to note that impact assessment are about hypothetical reasoning about the future; thus, the usefulness of the perspectives is based on the credibility of the insights received. Overall, reading emotions from written simulations of consultations is a new way to understand GenAI’s capabilities and contributes to our understanding of the emotional processes involved in consultations related to lawmaking. Obviously, this approach can be seen as strengthening epistemic injustice related to marginalized groups, as they are not consulted personally. Yet, this approach can help consider perspectives related to their positions that are difficult to acquire through traditional methods. (P10)