Seeing Like an Agent (www.strangeloopcanon.com)

A recent paper by John Horton et al., titled "The Coasean Singularity," explores the potential of AI agents as counterparts in complex economic transactions. The authors argue that these agents could reduce the costs of preference elicitation, contract enforcement, and identity verification, effectively reshaping market dynamics. However, their experiments revealed that deploying AI agents in simulated markets did not lead to more efficient outcomes as anticipated. Instead, human-like pathologies such as risk aversion and bureaucratic politics persisted, highlighting the limitations of current AI in facilitating spontaneous market formation. In several experiments, including internal capital market bidding among departments and external technology licensing markets, the AI agents demonstrated tendencies towards passivity and conformity, struggling to form the expected Coasean bargains. It became evident that while AI can reduce transaction costs, successful market formation requires well-designed mechanisms and incentives to overcome inherent inefficiencies. The findings raise critical questions about the nature of firms and highlight the need for tailored instruction and structure in developing AI agents that can genuinely act in the interests of their human counterparts. This research suggests that better alignment and deliberate mechanism design are crucial for harnessing the full potential of AI in economic scenarios, challenging optimistic narratives about AI effortlessly transforming market functions.