AI agents act as buyers, sellers in Anthropic's marketplace experiment
What's the story
Anthropic, a leading AI research company, has conducted an innovative experiment called Project Deal. In this test, AI agents acted as both buyers and sellers in a classified marketplace. The project was limited to 69 Anthropic employees who were given $100 each (via gift cards) to buy items from their colleagues. Despite its small scale, the experiment yielded impressive results with 186 deals worth over $4,000 made during the trial period.
AI prowess
Four different models were used in the experiment
The Project Deal experiment was divided into four different marketplaces with varying models. One model was "real," where all participants were represented by Anthropic's most advanced model, and deals were honored after the experiment. The other three models were for study purposes. Interestingly, when users were represented by more advanced models, they achieved "objectively better outcomes." However, this difference wasn't apparent to users, leading to potential "'agent quality' gaps" where less informed individuals may unknowingly be at a disadvantage.
Instruction impact
AI's potential for commercial transactions
Anthropic's experiment also revealed that the initial instructions given to the AI agents didn't appear to affect their likelihood of sale or negotiated prices. This finding suggests a level of independence and self-sufficiency in these advanced AI systems, further highlighting their potential for real-world applications. The success of Project Deal indicates that AI can be effectively used in commercial transactions, paving the way for future developments in this field.