Structural Blindness: Why Neither Humans Nor AI Reason as Well as We Think
March 30, 2026
Stephen's Web ~ OLDaily
Steve Hargadon, Mar 30, 2026 I'm always wary of arguments of the form here's something humans can do that AIs cannot because they usually over-represent human capacities and under-represent that of the AIs. Here Steve Hargadon cannot be accused of the former. In fact, I think overstates the view that are not naturally good reasoners, though he usefully describes institutions we've established to countervail that failing: the naming of logical fallacies, the legal system of trial and evidence, the scientific method, and government systems of checks and balances.
Then, based on his experience with Grok, he argues AI does not have access to the same sort of resistance. Well - maybe not in Grok, which was explicitly designed to subvert reason and evidence. Responsible adults should not use it. But I see evidence doubt, reason and evidence in AI all the time. Sometimes - like humans - AI needs to be prompted to exercise these capacities. But they are not wholly absent. Image: Dropbox 404. Web: [Direct Link] [This Post]
Stephen's Web ~ OLDaily
Coverage and analysis from Canada. All insights are generated by our AI narrative analysis engine.