Co-Producing AI: Toward an Augmented, Participatory Lifecycle

Participatory involvement throughout the entire AI lifecycle—not just at the end—is essential for addressing algorithmic bias. The paper proposes five interconnected phases (co-framing, co-design, co-implementation, co-deployment, and co-maintenance) where affected communities have input from problem definition through ongoing maintenance, rather than being consulted only after systems are built.

Current approaches to responsible AI, including ethical guidelines and technical fairness solutions, are insufficient on their own. The authors argue that these existing efforts have failed to prevent AI algorithms from disproportionately impacting culturally marginalized groups, indicating that a more fundamental restructuring of how AI is produced is needed.

Diverse perspectives and multidisciplinary collaboration must be embedded into AI development teams. The research, grounded in four multidisciplinary workshops, emphasizes that centering diversity, equity, and inclusion (DEI) throughout the production pipeline—rather than as an afterthought—is critical for identifying and mitigating harms before they occur.

"Distributed authority" and shared decision-making power should replace traditional top-down AI development models. The proposed lifecycle is built on principles of distributed authority and iterative knowledge exchange, suggesting that communities affected by AI systems should have genuine influence over decisions, not merely provide feedback.

Scaling participatory AI governance remains an open challenge with significant research questions still unanswered. The authors acknowledge that while the framework is promising, key questions about how to implement and scale this participatory approach across different contexts and organizations require further investigation.