Abstract
The paper focuses on the challenge of providing contextualized, value-sensitive, and participatory Artificial Intelligence (AI) that meets societal needs. The "AI for social Assessment" (AI FORA) project combines empirical research, gamification, and agent-based models (ABM) to assess the fairness of AI-based distribution in different countries and propose improvements. The paper presents a case study, where ABM and serious games are used to identify more desirable social assessment routines. A machine learning (ML) system is developed. The workflow for prototyping AI social assessment systems involves creating a synthetic population, generating synthetic outcomes based on improved assessment rules, training a neural network, and evaluating the effectiveness of the AI system. By following this approach, the paper suggests that unintended consequences and ineffective systems can be avoided, saving on costly development. The aim is to ensure that AI for social service delivery is responsive, fair, and beneficial to society.