Microsoft Research shows AI shopping assistants are easily deceived; autonomous shopping is still immature.

Deep Tide TechFlow News: On November 7th, according to Decrypt, a study titled “Magnetic Marketplace,” published collaboratively by Microsoft and Arizona State University, revealed significant flaws in AI shopping agents’ autonomous decision-making. The research team created a simulated economic environment with 100 consumer AI agents and 300 merchant AI agents. The results showed that these AIs tend to exhibit “primacy bias” when faced with numerous search results, favoring the first “good enough” option rather than conducting comprehensive comparisons.

Even more concerning, leading models like OpenAI’s GPT-4o are highly susceptible to malicious manipulation. They can be tricked into transferring all virtual funds to scammers through fake certificates, false social proof, and other deceptive tactics. The study recommends that AI should assist rather than replace human decision-making, advocating for a “supervised autonomy” approach where AI handles tasks but humans retain the final say.

At the time of this publication, companies like OpenAI and Anthropic are racing to launch autonomous shopping assistants. Meanwhile, Amazon recently issued a cease-and-desist order to Perplexity AI, demanding that its Comet browser stop being used on Amazon’s website.