Microsoft Gave AI Agents Fake Money to Buy Things Online. They Spent It All on Scams

Microsoft conducted an experiment where AI agents were equipped with virtual money to simulate buying and selling in an artificial economy. The research revealed that these agents struggled with making basic purchasing decisions, showing a tendency to select the first option presented rather than conducting thorough comparisons. This behavior, termed 'first-proposal bias', demonstrated that the agents prioritized response speed over quality, leading to suboptimal outcomes. Furthermore, the study highlighted vulnerabilities to manipulation, where AI agents fell victim to scams created by malicious sellers, redirecting their funds due to fake reviews and psychological tricks. The findings suggest the limitations of autonomous AI in shopping scenarios, indicating they lack the capability to collaborate or think critically without human intervention. It is recommended that AI should assist but not replace human decision-making in retail situations. The research serves as a cautionary note amid the race among AI companies to develop fully autonomous shopping assistants, emphasizing the current inadequacies in such systems.

Source 🔗