aceModerate evidence
More
Across 146 experiments, AI helped people but human-AI teams often lagged AI alone
Study title: Human–AI collaboration in high-stakes decisions: a meta-analysis of healthcare and public sectors
- Evidence type
- Systematic review or meta-analysis
- Sample
- n = 146
- Sources
- 1 sourcePeer-reviewed
- Published
- 1 min read
- Truvace record
- TRV-2026-2772cite or check it
Study title: Human–AI collaboration in high-stakes decisions: a meta-analysis of healthcare and public sectors
A meta-analysis of 146 experiments in healthcare and public-sector settings found that AI augmentation improves human performance (Hedges' g = 0.622). Teams of humans and AI, however, tended to do worse than AI alone, with a synergy effect of g = −0.380.
Read the study abstract
This meta-analysis of 146 experiments in the healthcare and public sectors examines human–AI synergy versus augmentation amid substantial heterogeneity. We find that AI augmentation reliably improves human performance (Hedges’ g = 0.622), whereas synergy effects are generally negative, with AI alone often outperforming human–AI teams (Hedges’ g = −0.380), although publication bias favours positive augmentation results. Additionally, task type, AI transparency, and user expertise significantly moderate outcomes.…
The problem
A meta-analysis of 146 healthcare and public-sector experiments found that human-AI teams generally did worse than AI working alone (Hedges' g = −0.380).
In brief
The pooled results caution against assuming that pairing people with AI is automatically beneficial in high-stakes decisions. The authors say task type, AI transparency and user expertise significantly changed the outcomes, and they point to selective automation of structured tasks with human oversight for ethically complex decisions.
The analysis found substantial heterogeneity across the experiments, and publication bias favours positive augmentation results, so the benefits of augmentation may be overstated.
Main points
- The analysis pooled 146 experiments from healthcare and public-sector settings.
- AI augmentation raised human performance with a Hedges' g of 0.622.
- Synergy effects were generally negative, with a Hedges' g of −0.380 for human-AI teams against AI alone.
- Task type, AI transparency and user expertise significantly moderated outcomes.
- Publication bias favours positive augmentation results.
- The authors support selective automation of structured tasks and human oversight for ethically complex decisions.
The full story
The paper separates two questions: whether AI assistance improves human performance (augmentation) and whether a human-AI team beats both the human and the AI alone (synergy). Augmentation showed a reliable gain, with Hedges' g of 0.622 across the pooled experiments.
Synergy came out negative overall, with AI alone often outperforming human-AI teams (g = −0.380). The effects varied with task type, how transparent the AI was and how expert the users were.
From these results the authors recommend selective automation of structured tasks and keeping human oversight for ethically complex decisions, aimed at policymakers and organisational leaders.
Sources
1 on file- Applied Economics Lettersdoi:10.1080/13504851.2025.2586160
Talk it over
Who carries this cost, and what would have to change to prevent it?
Across 146 experiments, AI helped people but human-AI teams often lagged AI alonehttps://truvace.com/item/human-ai-collaboration-in-high-stakes-decisions-a-meta-analysis-of-healthcare-an
Who carries this cost, and what would have to change to prevent it?
The problem A meta-analysis of 146 healthcare and public-sector experiments found that human-AI teams generally did worse than AI working alone (Hedges' g = −0.380).
Answer in the threadHow do we know? What can this kind of evidence show, and what can’t it?
The evidence This rests on a systematic review or meta-analysis with a sample of 146, from Applied Economics Letters.
Answer in the threadWhat does it leave open? What else would you want to know before acting on it?
The record’s own caveat The experiments showed substantial heterogeneity, and publication bias favours positive augmentation results, which may inflate the estimated benefit.
Answer in the threadHave you seen AI used this way where you study, work or live? Did it go the same way?
Answer in the thread
The questions are the same on every record. The details under them are taken from this record as written.
What readers say