When AI gets better at predicting what will succeed, what's the smartest thing left for a human to watch for?

This explores why better AI prediction changes which kind of human judgment is worth the most, moving it from 'here's a chance worth taking' to 'here's a danger you're missing.' The clearest account in the collection is a formal economic model by Agrawal, Gans and Goldfarb Does better prediction change which human judgment matters most?. Their argument depends on what counts as the default. When predictions are poor, the safe choice is the sensible starting point, because you can't tell when a bold move will pay off. A human adds the most value by spotting a hidden opportunity that makes the risk worth taking. When predictions get good, the risky but rewarding action becomes the default, since the machine can usually tell when it will work. A human then adds the most value by spotting a hidden cost the prediction can't see, which is a reason to pull back to the safe option. Put simply, the human stops being the accelerator and becomes the brake.

This matters because good average accuracy can hide failures in exactly the cases that count. Separate work on decision-making shows that a model can predict well overall and still get things systematically wrong in the situations that matter most for the decision Why do accurate predictions lead to poor decisions?. Those are the blind spots that 'hidden cost' judgment has to catch. Taken together, the two notes suggest a better predictor doesn't remove the need for judgment. It moves that need into rarer, harder-to-see corners where the model is confidently wrong.

The awkward part is that this braking kind of judgment is the easiest kind to skip. The same authors show that managers with limited time and attention will rationally hand decisions to machines even when their own judgment would be better Do managers delegate to machines even when human judgment is better?. Judgment takes deliberate effort, so people ration it. When the machine's default is already the bold option, rationing judgment means the bold option goes ahead unchecked. The opportunity-spotting role used to cost little to skip, because skipping it just meant staying safe. Skipping the cost-spotting role means risk gets through.

A labor-market view in the collection reaches a similar conclusion. When first-pass thinking is cheap, the scarce asset is accountable judgment: people who check outputs, own the consequences, and learn from practice What makes accountable judgment scarce when AI cognition is cheap?. That note adds a condition. This role only survives if institutions protect it, by keeping people's right to question the machine and their chance to build the experience that makes their objections worth listening to.

The collection's direct material here is thin. One formal model carries the core mechanism, and the other notes are related support rather than independent evidence for the shift itself. Still, the combined picture holds up: better prediction makes bold action the norm, so the valuable human becomes the one who knows when to say no, and that is the role organizations are most tempted to drop.