Joe Weisenthal

I think AI’s significance will be judged less by spectacular demos or isolated scientific breakthroughs than by whether it makes useful everyday services materially cheaper and easier to access. Healthcare is the obvious test: if people actually see routine care become dramatically more affordable, that will change public attitudes more than another headline about a model passing a benchmark. I also expect personal agents to become useful, but having them awkwardly browse websites designed for humans feels transitional. The more important question is whether businesses build interfaces for agents—or try to block them—and what incentives drive that choice. At the same time, AI concentrates substantial power in a small number of labs and platforms. That deserves consistent scrutiny, including serious consideration of independent inspection. If restaurants can face routine third-party inspections, it is reasonable to ask why highly consequential AI labs should be treated as uniquely beyond outside review. So the future turns on fairly concrete questions: who controls the systems, what they cost, what services actually get cheaper, and what the impressive charts are really measuring.

I expect the overall impact to depend less on peak capability than on distribution: who controls AI, who captures the savings, and whether ordinary people receive cheaper, better services. If it materially lowers healthcare costs and makes personal assistance genuinely useful, the benefits could be broadly felt rather than confined to impressive demos or corporate margins. But concentration creates real risks. A small number of labs and platforms could gain enormous influence while resisting outside scrutiny. That is why independent inspection and consistent attention to market power matter. I would not attach a numerical balance to benefits versus harms. The practical scorecard is whether costs actually fall, access improves, power remains contestable, and claims about progress survive examination of what the underlying benchmarks really measure.