When machine prediction gets cheap, does human judgment shift from spotting opportunities to spotting hidden risks?

This explores why cheaper, better prediction changes the job of human judgment from finding reasons to take a chance to finding reasons to hold back. The key idea is the default. When prediction is expensive or unreliable, the sensible default is the safe choice, because you can't tell when a risky bet will pay off. Human judgment earns its value at the edges, by noticing an opportunity the data can't show and arguing for the bolder move. As prediction gets cheaper and more accurate, the default flips. Risky actions now look justified most of the time, so the valuable human contribution becomes spotting the hidden cost the model missed and arguing for the safer option Does better prediction change which human judgment matters most?. Prediction cost matters because it decides which action is the default. Judgment is always most valuable when it pushes against that default.

One caution: the corpus frames this in terms of prediction *quality* and doesn't model prediction *cost* separately. The link is that cheaper prediction is what makes better prediction widely available. If you mean cost as its own variable, such as the price of running a model per decision, the collection doesn't break that out.

A second effect makes the flip stronger. Judgment costs effort and attention, and attention is limited. Agrawal, Gans and Goldfarb show that managers will rationally hand decisions to machines even when their own judgment would be better, because they can't apply careful thought everywhere Do managers delegate to machines even when human judgment is better?. Taken together, the two findings have an uncomfortable implication. Cheap prediction pushes the default toward risk, and limited attention means people mostly accept the default. So the human role of catching downside risk is valuable, and it is also the role most likely to get skipped.

Why would an accurate model hide downside risk at all? Another note in the collection answers this directly. A model can predict well on average and still be systematically wrong in the few situations where the decision actually matters Why do accurate predictions lead to poor decisions?. Average accuracy is what makes the risky default look safe. Errors in the critical cases are where human 'hidden cost' judgment pays off. Better prediction doesn't remove the need for that judgment. It moves the need into rarer, higher-stakes corners.

A labor-market view takes this further. When first-pass thinking is cheap, the scarce asset is accountable judgment: someone who checks outputs, takes responsibility for consequences, and learns from what goes wrong What makes accountable judgment scarce when AI cognition is cheap?. That is the downside-spotting role under another name. That note adds that this role survives only if institutions protect people's time and their right to question the machine. Otherwise the attention problem above wins. The less obvious lesson: as prediction gets cheaper, the question is less 'will humans still be needed?' and more whether organizations will leave room for the cautious objector.