Should companies hand a decision to whoever's more accurate — a person or an AI — or is that the wrong question entirely?
How should organizations assign decisions between humans and AI based on accuracy?
This looks at how organizations should decide which decisions go to people and which go to AI, and whether 'who is more accurate?' is even the right question to ask.
This looks at how organizations should decide which decisions go to people and which go to AI, and whether accuracy is the right test for that split. The corpus has a surprising answer: picking whoever scores higher is often the wrong rule. The most useful question is how humans and AI get things wrong differently, and how the work is set up around them.
Start with the counterintuitive result. When humans and AI make predictions on the same task, combining their answers beats either one alone, as long as their mistakes don't overlap too much. Neither has to be especially accurate on their own Can humans and AI improve decisions by predicting together?. That changes who you should worry about replacing. In a human-AI team, the person most at risk isn't the least accurate one. It's the one whose errors look most like the AI's, because they add the least new information to the group Does error diversity matter more than accuracy for human replacement?. An organization that sorts people by accuracy alone may let go of exactly the people whose different blind spots were making the system better.
That's why some researchers treat this as an organization design problem rather than a matter of matching tasks to the best performer. Puranam treats a human-AI pair like a small multi-agent organization. Different kinds of tasks call for different ways of dividing the work. How well the pair performs over time also depends on how feedback flows back and who learns from it How should humans and AI systems divide decision-making tasks?. Accuracy today is a snapshot. The structure decides whether either side gets better.
There is also a middle ground between "human decides" and "AI decides." One approach has the AI stop issuing verdicts and instead point out which parts of a case deserve attention. This avoids the anchoring that happens when people see a machine's answer first Can AI guidance reduce anchoring bias better than AI decisions?. A related idea from philosophy says AI outputs should count as one piece of evidence to weigh, not an answer that ends the discussion. People should trust the AI less when a case falls outside what it knows, or when new evidence comes up Should AI outputs replace or supplement human judgment?. Both suggest that how you hand over a decision can matter as much as who gets it.
Finally, accuracy scores leave out what experts actually contribute. Part of expertise is deciding which differences matter in a situation, and AI's pattern-matching doesn't do that kind of selecting Can AI distinguish which differences actually matter?. Expert judgment also anticipates what an audience will accept as valid Can AI replicate the communicative work experts do?. And as decision authority moves toward AI, the corpus says monitoring has to come first, because AI reasoning gets harder to inspect over time Must AI monitoring precede incentive alignment as agents evolve?. The same concern appears in designs that match how much authority an agent takes to what users expect How should AI agents balance autonomy with user oversight?. The takeaway: assign decisions by how errors combine, how feedback flows, and what can still be watched, not by a leaderboard.
Sources 9 notes
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.
Choudhary and colleagues show that the human most at risk of replacement in an AI ensemble is not the least accurate but the one whose prediction errors most closely align with the AI's. Ensemble performance depends on error dispersion, not individual skill.
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.
Learning to Guide eliminates anchoring bias and unassisted hard cases by having machines supply interpretive guidance rather than autonomous decisions, keeping responsibility with humans while improving their judgment through enhanced perception.
Research argues AI should supplement rather than replace human reasoning, with deference withdrawn when domain mismatch, bias, conflicting authority, or new evidence emerges. This prevents opacity-driven failures that full preemption would mask.
Show all 9 sources
Experts observe by choosing which differences matter (qualitative judgment); AI finds patterns and probabilities (quantitative). AI generates text from prompts without observing context, audience needs, or knowledge states—producing fabrication that mimics observation's form without its epistemic process.
Expertise requires anticipating audience acceptability and social validity, not just retrieving information. AI lacks the mechanism to perform this communicative work, making its fluent output epistemically misleading despite its confident form.
Humberd and Latham argue that as AI becomes more autonomous, monitoring mechanisms must be established first as prerequisites to incentive alignment, because defining desirable AI behaviors is harder than for humans and AI decision-making becomes increasingly incomprehensible over time.
Research proposes autonomy alignment—matching decision authority to user expectations—as distinct from output alignment. It operates at model level (escalation policy), harness level (enforcement mechanism), and interaction level (how delegation surfaces to users).
Papers this line draws on 8
The research behind the notes this line reads — ranked by how closely each paper relates.
- Human-AI Ensembles: When Can They Work?
- Exploring the Impact of Artificial Intelligence: Prediction versus Judgment
- Complementarity in Human-AI Collaboration: Concept, Sources, and Evidence
- To Trust or to Think: Cognitive Forcing Functions Can Reduce Overreliance on AI in AI-assisted Decision-making
- GenAI as a Power Persuader: How Professionals Get Persuasion Bombed When They Attempt to Validate LLMs
- Can AI Do Strategy?
- Roles of Artificial Intelligence in Collaboration with Humans: Automation, Augmentation, and the Future of Work
- Human-AI collaborative decision-making as an organization design problem