How public involvement can improve the science of AI

Area of Science:

• Computer Science
• Social Science
• Participatory Science

Background:

• AI systems are increasingly deployed, necessitating trustworthy evaluations of their safety and reliability.
• Affected communities demand involvement in AI evaluation processes.
• Sociotechnical challenges arise from AI systems adapting to social contexts.

Purpose of the Study:

• To explore the value of public involvement in the science of AI.
• To offer guidance on engaging lived-experience experts in AI evaluation.
• To address concerns regarding participatory methods in AI research.

Main Methods:

• Review of sociotechnical challenges in AI evaluation.
• Summarization of literature on participatory science.
• Discussion of case studies in AI for healthcare, policing, and social media.
• Analysis of five key stages of quantitative AI evaluation.

Main Results:

• Public participation can improve AI evaluations at multiple stages: equipoise, explanation, measurement, inference, and interpretation.
• Engaging lived-experience experts addresses concerns about generalizability, subjectivity, and reliability.
• Participatory methods offer practical logistics for inclusive AI research.

Conclusions:

• Integrating public involvement strengthens the scientific rigor of AI evaluations.
• Participatory science is crucial for developing trustworthy AI systems.
• Collaborative approaches foster more reliable and socially relevant AI assessments.