Why does a human plus an AI often make worse calls together than the sharper one would alone?
Why do most human-AI decision teams underperform the better individual alone?
This explores why pairing a person with an AI on a decision so often produces worse results than just letting whichever one is better (human or AI) decide alone, and what the corpus says about when that pattern breaks.
This explores why human-AI teams on decision tasks so often do worse than the stronger partner working alone, and what conditions might reverse that. The short answer from the corpus: the problem isn't that either side is weak. Combining them adds its own losses, and the conditions that would make the pair better than either one rarely hold in practice. A meta-analysis of over 100 studies found that human-AI combinations score below the best individual on average, with decision tasks doing worst and content-creation tasks the main exception where gains show up When do human-AI teams actually outperform either alone?. The authors put the difference down to how the task is structured, not to how capable either partner is.
One useful way to see the problem is to separate two sources of possible synergy. The first is simply picking the better answer on each case. The second is actually exchanging information or reasoning so the joint answer beats both Can humans and AI teams achieve performance neither reaches alone?. Both depend on each side knowing something the other doesn't. Even then, the pair has to recognize *when* to trust whom, and people are poor at that. A 535-person study makes this concrete. When the LLM got more accurate on a given item, people using it captured only about half of that gain, which left them below what the AI alone would have scored Why does assisted accuracy capture only half the LLM gain?. In other words, the human acts as a leaky filter. They override correct AI answers and accept wrong ones often enough to drag the team down.
The pattern is that combining helps only when the two sides fail differently. A purely statistical approach shows that averaging human and AI predictions beats either alone whenever their errors are diverse enough, even if neither is very accurate Can humans and AI improve decisions by predicting together?. That points to an uncomfortable implication: a mechanical blend can outperform a human who *thinks* they're judiciously weighing the AI's advice. Organization-design work makes the same point from another angle. How you divide tasks and set up feedback changes what the pair learns and how well it performs, so a team with no deliberate design can underperform by default How should humans and AI systems divide decision-making tasks?. Attention matters too. Decision-makers with limited time hand decisions to machines even when their own judgment would be better Do managers delegate to machines even when human judgment is better?. So in practice, 'human oversight' can quietly turn into rubber-stamping.
The less obvious lesson comes from AI-only teams, which show the same problem. Teams of per-user agents consistently did worse than a single coordinator agent serving everyone Do teams of personal agents outperform a single coordinator?. Diverse multi-agent ideation teams without real expertise did worse than one competent agent, because diversity without knowledge creates coordination overhead instead of insight Does cognitive diversity alone improve multi-agent ideation quality?. One fix from that world is to measure each member's contribution and drop the ones adding nothing Can multi-agent teams automatically remove their weakest members?. That suggests underperformance is a general cost of teamwork, not something specific to humans and AI. Teams have to earn their coordination costs.
Where combinations do pay off, the task is usually open-ended creation rather than picking a right answer. At Procter & Gamble, individuals working with AI matched the output of two-person teams without AI Can generative AI replace the benefits of having a human teammate?. Research teaming agendas argue that clearly defined roles that can shift as needed, plus shared awareness of the situation, are the missing ingredients What blocks effective human-AI teams from performing well?. If you're designing a human-AI decision process, the corpus suggests starting with two questions. Do the two sides actually make different mistakes? And does anyone know when to defer to whom?
Sources 11 notes
A meta-analysis of over 100 studies finds human-AI combinations score lower than the best of human or AI alone (g = −0.23), except in content-creation tasks where gains emerge. Decision tasks show losses; the difference lies in task structure, not ability.
A framework distinguishes two routes to team synergy: inherent potential from picking the better individual answer per task, and collaborative potential from exchanging information or reasoning approaches. Information and capability asymmetry drive both, but genuine complementarity remains rare because the conditions enabling it are rarely satisfied.
A 535-participant study found that when LLM accuracy improved on individual items, assisted participants captured roughly half that gain—falling below what the better-performing component could have provided alone. This shows complementarity creates potential but does not guarantee synergy.
Choudhary et al. show that aggregating human and AI predictions on identical tasks outperforms either alone whenever their errors differ sufficiently, even if neither achieves high accuracy individually. This requires diversity in how each errs, not specialization.
Puranam frames human-AI pairs as multi-agent organizations requiring design choices about task division and learning configuration. He distinguishes three task types and shows learning mechanisms produce different performance outcomes based on feedback structure and decision architecture.
Show all 11 sources
Agrawal, Gans and Goldfarb show that decision-makers with limited attention will delegate to machines even when human judgment would decide better. The mechanism is attention scarcity: judgment requires deliberate cognitive effort, forcing managers to choose where to apply it, making machine delegation optimal in aggregate.
Across five frontier models and 77 scenarios in four shared-resource environments, a single coordinator agent serving all users consistently delivered better group outcomes than teams of per-user agents. Teams collapsed entirely in two environments without communication channels and showed substantial performance gaps even with coordination mechanisms.
Multi-agent teams substantially outperform solo ideation, but only when members possess genuine senior knowledge. Diverse teams without expertise underperform even a single competent agent, because cognitive stimulation without expertise triggers process losses instead of insight.
DyLAN's three-step importance scoring mechanism (propagation, aggregation, selection) quantifies individual agent contributions and automatically removes uninformative agents during inference, optimizing team composition without task-specific tuning.
In a randomized field experiment with 776 P&G professionals, individuals using AI produced solutions as strong as two-person teams without AI. AI also reduced functional silos by prompting more balanced solutions across professional backgrounds.
Effective human-AI teaming requires extending Team Situation Awareness theory with role specification and role fluidity, grounded in trust-building. The research proposes four focus areas—formulation, coordination, maintenance, and training—to operationalize this framework.
Papers this line draws on 8
The research behind the notes this line reads — ranked by how closely each paper relates.
- Complementarity in Human-AI Collaboration: Concept, Sources, and Evidence
- Human-AI Ensembles: When Can They Work?
- Unraveling Human-AI Teaming: A Review and Outlook
- Navigating the Jagged Technological Frontier: Field Experimental Evidence of the Effects of Artificial Intelligence on Knowledge Worker Productivity and Quality
- When combinations of humans and AI are useful: A systematic review and meta-analysis
- Roles of Artificial Intelligence in Collaboration with Humans: Automation, Augmentation, and the Future of Work
- Can AI Do Strategy?
- Quantifying Human-AI Synergy