Science
Experimental evidence on the productivity effects of generative artificial intelligence
Key Takeaway
In a preregistered experiment with 453 professionals, ChatGPT access reduced task time by 40% and increased assessed output quality by 18%. Lower-skill participants benefited most on the short writing tasks studied.
AI-generated HIE assessmentStudy Design
preregistered randomized experiment
title and abstract
What the Study Found
In a preregistered online experiment, 453 college-educated professionals completed occupation-specific, incentivized writing tasks, with half given access to ChatGPT. AI access reduced completion time by 40% and increased output quality by 18%; lower-skill participants benefited most. The study measures short, self-contained writing tasks and immediate outcomes, which limits conclusions about long-term workplace effects.
Original paper recordEvidence Strength
Evidence strengths
- Random assignment
- Preregistered experiment
- Incentivized occupation-specific tasks
Limitations
- Short and self-contained tasks
- College-educated sample
- Primarily immediate outcomes
What This Does Not Prove
This record does not establish effects beyond the population, tasks, duration, and methods described by the paper. It should not be read as a universal causal claim.
HIE Interpretation
Preregistered randomized evidence that generative AI changes speed, quality, and inequality in professional writing tasks.
AI-generated HIE assessment · AI assessment confidence 93Related Topics
- Human-AI Interaction
- Artificial Intelligence
- Learning
Connected Claims
- Supports
Human-AI collaboration does not reliably outperform the better standalone human or AI; results depend strongly on the task and interaction design.
Randomized access to ChatGPT improved speed and quality on bounded professional writing tasks, especially for lower-skill participants.
- Scientific evidence strength
- 91
- AI relationship confidence
- 93
- Source / review level
- Abstract reviewed
- Human review
- No human review
- Provenance
- codex-research-curation · initial-curation-v1
Original Source
DOI 10.1126/science.adh2586 ↗