Michael’s talk was based on two of his recent publications. He began by offering some highlights from his paper titled ‘Understanding Accountability in Algorithmic Supply Chains’ (co-authored with Jennifer Cobbe and Jatinder Singh). More specifically, Michael spoke about how the production, deployment, and use of AI tools are split between various actors, all of which influence these tools’ functionality. As these actors bear distributed responsibility for AI systems yet have a limited overview of the overall supply chain, the emerging legal accountability frameworks may be ineffective in holding relevant entities responsible for the outcomes of their operations. Then, Michael presented a summary of his paper titled 'Moderating Model Marketplaces: Platform Governance Puzzles for AI Intermediaries' (co-authored with Robert Gorwa), which outlines how platforms hosting user-generated AI models and training data aim to address their use in an illegal or harmful manner. As seen from three case studies involving Hugging Face, GitHub and Civitai, AI marketplaces have been developing some basic moderation strategies, such as licensing or access and use restrictions. However, Michael contended that effective AI model moderation calls for a significant increase in the analytic capacity which can be achieved through the (partial) externalisation of moderation tasks.

The discussion that followed Michael’s talk revolved around the the conceptual underpinnings of supply chains, the agency of technology, and the implications of AI model moderation for freedom of expression.