Publication
PNAS
Stage
Journal article
What we read
Summary of the paper
Authors
Julian Berger, Jason W. Burton, Ralph Hertwig, Thomas Kosch, Ralf H. J. M. Kurvers, Benito Kurzenberger, Christopher Lazik, Linda Onnasch, Tobias Rieger, Anna I. Thoma, Dirk U. Wulff, Stefan M. Herzog
Universities and research institutions
Universitat Pompeu Fabra, University of Southern Denmark, Max Planck Institute for Human Development, Barcelona School of Economics, University of Copenhagen, Humboldt-Universität zu Berlin, Technische Universität Berlin, Vienna University of Economics and Business

Institution metadata: OpenAlex record ↗

What they did and found

Researchers reanalyzed 74 studies on people working with AI tools. They found that giving people feedback after each decision often leads to better results, especially when the AI also provides simple explanations. Without feedback, benefits are unclear or negative.

Why it matters

Using AI to aid decisions can improve outcomes with feedback and explanations, but the benefit is not guaranteed and depends on task design.

Learning improves when people see the AI's choice, a brief explanation, and feedback on results; this helps them know when to trust AI.

Practically, design systems that show results after each decision and give simple explanations. But not all studies show clear benefits, so test in your setting first.

Original sources · 1
  1. Human learning is an understudied but promising lever for boosting human–AI synergy ↗PNAS · 2026-09-21

Check the original paper for its authors, methods, version and access terms.

What remains uncertain

Findings are correlational; not all studies showed gains, and stronger causal evidence requires experiments that actively vary learning opportunities.