A speculative essay questions the sci-fi dream of autonomous AI agents negotiating on our behalf. The author argues that agents will vary wildly in capability, and a weak, low-resource agent (e.g. one running on a Raspberry Pi) could easily be manipulated, hypnotized, or outmaneuvered by a more powerful corporate agent backed by massive infrastructure, just as humans fall for advertising tricks. The piece frames this as an under-discussed risk of delegating decisions to AI agents.
119 Impressions