What should AI contribute to a decision?

What should AI contribute to a decision?

AI earns a place in decision-making when it improves the decision surface, not when it merely produces an answer. Its strongest contribution is to make more evidence, uncertainty and alternatives visible before a person or governing body commits resources.

Three roles are especially valuable. As a compressor, AI turns dispersed information into a structured brief with traceable sources. As a challenger, it tests the prevailing view, searches for disconfirming evidence and exposes assumptions. As a simulator, it explores scenarios and sensitivities faster than a team can do manually. None of these roles requires transferring final authority.

Performance is sharply task-dependent. In a field experiment with 758 consultants, AI improved speed by more than 25%, performance by over 30% and completion by more than 12% on tasks inside its capability frontier. On a task outside that frontier, users of AI were 19 percentage points less likely to reach the correct answer. The lesson is operational: classify the decision before selecting the human�AI configuration.

Separate observation, inference, recommendation and authority. The system may extract facts, estimate a base rate and generate options; the accountable decision-maker still chooses objectives, weighs non-quantifiable trade-offs and owns exceptions. For high-consequence choices, require source provenance, explicit assumptions, missing information, alternative explanations and a route to escalation.

Do not measure success by recommendation acceptance. Track calibration, reversals, override quality, cycle time, realised outcomes and the cost of errors against a credible baseline. A useful system makes judgement more rigorous�even when the best decision is to reject its recommendation.