Public Opinion and its Critical Role for Highlighting Lived Experiences in Risk Identification and Risk Evaluation Practices for Risk Management Approaches.

As the impact of AI on individuals, organizations, and society increases, the EU has adopted risk-based approaches, such as the DSA and the EU AI Act, to guide implementation processes. These approaches aim to assess and mitigate fundamental AI risks while protecting societal values. However, the effectiveness of these approaches depends on the methods used to identify, rate, and manage risks. By analysing the regulatory landscape and existing risk management frameworks, we identified critical gaps in these approaches. Notably, current risk assessment and management approaches lack the inclusion of non-expert voices, i.e., the public's perspective on AI risks. These perspectives are important not only for identifying risks but also for navigating critical trade-offs between potential benefits and risks. Consequently, excluding citizens from most risk assessment practices creates a democratic accountability gap. We argue that incorporating public perspectives, based on citizens' lived experiences, can enhance expert-driven risk assessments by: i) Identifying and describing overlooked or neglected risks, ii) Putting risks into real-world contexts that highlight the socio-technical elements of risks, and iii) Providing estimates of risk severity and magnitude as well as risk acceptance. To address these shortcomings, we advocate for public opinion studies using mixed-methods approaches that aim to i) qualitatively describe and identify risks from a citizen-centric socio-technical perspective, and ii) quantify the severity, and magnitude of those risks, and tapping into risk-benefit trade-offs. To put this into practice, we conducted two empirical studies focusing on the impact of generative AI on legal conflict resolution. In Study 1, we surveyed EU citizens (n=25) and legal professionals (n=25) about the potential impact of generative AI (genAI) on legal conflict resolution using scenario writing. Specifically, we qualitatively analysed the prevalence of risk and benefit themes, as well as the types of anticipated legal tasks. We also described the emerging trade-offs that will affect decision-makers in the legal field. In Study 2, we adapted a scenario from Study 1 detailing a common legal conflict with two legal tasks (legal consultation and legal mediation) as stimulus material for a public opinion survey of a representative sample of German citizens (N=488). Specifically, we surveyed respondents about risk and benefit factors for both legal tasks and their acceptance of risk for the use of genAI for these tasks. Thus, Study 2 quantifies citizen-centred risk acceptance of the highly contested use of genAI. Together, both studies provide important insights into citizens' perspectives on risk governance approaches.