Contested · Abstract reviewed
Human-AI collaboration does not reliably outperform the better standalone human or AI; results depend strongly on the task and interaction design.
Plain-language explanation
A broad synthesis finds no universal synergy, while field and laboratory studies show meaningful gains in bounded tasks. The defensible conclusion is conditional, not categorical.
Supporting evidence
- Generative AI at Work
AI assistance improved productivity in a specific customer-support setting, with heterogeneous effects by worker experience.
- Evidence strength
- 90
- AI assessment confidence
- 91
- Method
- Staggered real-world deployment study.
- Source scope
- title and abstract
- AI provenance
- codex-research-curation · initial-curation-v1
- Human review status
- No human review
- Experimental evidence on the productivity effects of generative artificial intelligence
Randomized access to ChatGPT improved speed and quality on bounded professional writing tasks, especially for lower-skill participants.
- Evidence strength
- 91
- AI assessment confidence
- 93
- Method
- Preregistered randomized online experiment.
- Source scope
- title and abstract
- AI provenance
- codex-research-curation · initial-curation-v1
- Human review status
- No human review
Qualifying evidence
- When combinations of humans and AI are useful: A systematic review and meta-analysis
Across 106 experiments, combinations were worse than the better standalone agent on average, with task type explaining important variation.
- Evidence strength
- 95
- AI assessment confidence
- 95
- Method
- Preregistered systematic review and meta-analysis.
- Source scope
- title and abstract
- AI provenance
- codex-research-curation · initial-curation-v1
- Human review status
- No human review
- Does the Whole Exceed its Parts? The Effect of AI Explanations on Complementary Team Performance
Explanations increased acceptance of AI advice but did not increase complementary team performance.
- Evidence strength
- 87
- AI assessment confidence
- 88
- Method
- Mixed-method user studies across three datasets.
- Source scope
- title and abstract
- AI provenance
- codex-research-curation · initial-curation-v1
- Human review status
- No human review
Contradicting evidence
No public evidence in this category.
Contextual evidence
No public evidence in this category.