Loan decisions with AI got better over 10 rounds, experiment finds

The study varied two factors: how accurate the AI's predictions were and how well incentives were aligned with organizational goals. Both affected outcomes on their own and in combination. Incentive alignment mattered most when the AI's predictive power was low.

The authors read this as a sign that aligned incentives may cushion organizations when AI performance degrades. They also report that cognitive capabilities, trust in AI, algorithm integration and decision quality shaped how individuals engaged with organizational learning processes.

The researchers draw on the absorptive capacity view, treating AI systems as organizational knowledge structures and incentive policies as mechanisms that transfer knowledge between people and algorithms. They examine how collaboration changes over time rather than at a single point.

In the experiment, financially incentivized participants assessed loan applications with AI help. The setup let the team vary AI predictive power and incentive alignment and observe their separate and combined effects on organizational outcomes.

The interaction result is that incentive alignment mattered most when the AI's predictive power was low. The authors suggest aligned incentives may buffer organizations against the negative effects of AI degradation, and argue human-machine collaboration could become a new source of organizational learning.