Abstract
This paper introduces and illustrates a process for stakeholder-driven innovation in a highly contested domain: using artificial intelligence (AI) algorithms for social service delivery in national welfare systems. AI technologies are increasingly being applied because they are assumed to lead to efficiency gains. However, the use of AI is being challenged for its fairness. Existing biases and discrimination in service delivery appear to be perpetuated and cemented as a result of basing the AI on machine learning of past data. Fairness, however, is a dynamic cultural concept: its meaning in terms of values and beliefs, its implications for technology design, and the desired techno-futures need to be societally negotiated with all stakeholders, especially vulnerable groups suffering from current practices. The challenge is to provide contextualized, value-sensitive and participatory AI that is responsive to societal needs and change. The ‘AI for Assessment’ (AI FORA) project combines empirical research on AI-based social service delivery with gamification at community-based multi-stakeholder workshops and a series of case-specific agent-based models for assessing the status quo of AI-based distribution fairness in different countries, for simulating desired policy scenarios, and for generating an approach to ‘Better AI’. The paper is structured as follows: after introducing the participatory approach of AI FORA with its motivation and overall elements, the paper focuses on gamification and simulation as central components of the modelling strategy. Case-specific game design and ABMs are described and illustrated using the example of the AI FORA Spanish case study.