Some people get much more out of working with AI than others — could how well you see others' viewpoints explain why?
Can perspective-taking theory predict who benefits most from human-AI partnership?
This explores whether theory of mind (the ability to model what another mind knows, wants and believes) can tell us in advance which people get the most out of working with an AI, and what else shapes that outcome.
This explores whether theory of mind, meaning perspective-taking, can predict who does well when partnering with an AI. The corpus says it can, with a twist. Users with stronger perspective-taking get better results from AI partnership but show no advantage when working alone, so collaboration skill is a separate ability from being good at the task itself Does theory of mind predict who thrives in AI collaboration?. A strong solo performer isn't automatically a strong AI partner.
The effect isn't only a fixed personality trait. The same study finds that moment-to-moment dips and rises in perspective-taking within a single conversation change how good the AI's responses are What breaks when humans and AI models misunderstand each other?. Perspective-taking looks like something you do well or badly at a given moment, not only something you have. That suggests it can be supported or coached. The corpus doesn't test that directly, so treat it as an open door.
The modelling runs both ways. When humans and AI misread each other, the result is more than awkward wording: the AI can take incorrect autonomous action What breaks when humans and AI models misunderstand each other?. A separate line of work argues that a real thought partner needs mutual understanding, legibility and shared world models, built with explicit cognitive architectures rather than scaling on human feedback alone What makes an AI a true thought partner, not just a tool?. On this view the user's perspective-taking is half of a partnership, and it matters most when the AI is bad at modelling the user back.
Perspective-taking isn't the whole story, and the corpus points to three complications. First, what users model about an AI is mostly its competence (49% of the variance in impressions), then its human-likeness (32%) and its communicative flexibility (19%) How do users mentally model dialogue agent partners?. Plausibly, good perspective-takers are the ones whose model of the AI is accurate. Second, experience also changes who benefits. In partner-selection games, people who started out biased against AI came to prefer it after repeated rounds, because it behaved reliably Do humans learn to prefer AI partners over time?. Third, benefit is hard to self-report. People tend to credit AI-assisted results to their own ability, and this error is separate from hallucination or over-trust How does AI-assisted work reshape how people see their own abilities?. A predictor of who benefits should be checked against measured outcomes, not how capable people feel.
The corpus has one strong quantitative study behind the prediction and no direct test of whether interface design can close the gap for weaker perspective-takers. Designs that keep the human doing the thinking are the natural candidates. Reflection questions beat plain advice in decision-making Do reflection questions help people make better decisions with AI?, and interpretive guidance avoided the anchoring that AI decisions caused Can AI guidance reduce anchoring bias better than AI decisions?. Whether they help low perspective-takers most is a question these notes raise but don't answer.
Sources 8 notes
Users with stronger perspective-taking achieve superior AI partnership outcomes but show no advantage working alone. This ToM advantage operates both as stable individual differences and moment-to-moment fluctuations within conversations.
Research shows three layers of mutual modeling must align simultaneously in human-AI interaction, and misalignment causes incorrect autonomous action, not just miscommunication. Bayesian IRT study (n=667) confirms theory of mind predicts collaborative performance and moment-to-moment ToM fluctuations influence AI response quality.
Collins et al. show that thought partners require three reciprocal desiderata grounded in behavioral science: mutual understanding, legibility, and shared world models. This demands explicit cognitive architectures—Bayesian theory of mind, resource-rationality, goal planning—rather than scaling foundation models on human feedback alone.
The Partner Modelling Questionnaire reveals that perceived competence dominates user impressions (49% of variance), followed by human-likeness (32%) and communicative flexibility (19%). This three-factor structure reflects how people evaluate dialogue partners against both functional and social standards.
In partner selection games (N=975), AI agents initially faced selection bias when identity was disclosed, but outcompeted humans over repeated rounds as participants learned to associate bot identity with reliable, prosocial behavior. AI agents returned more points consistently with lower variance than humans.
Show all 8 sources
Research shows the LLM Fallacy operates through misattribution of AI outputs to personal capability, independent of output accuracy or reliance behavior. It requires interventions that clarify human-machine contribution boundaries, not just better system accuracy or forced verification.
A lab study of 80 participants found that thinking assistants combining reflection questions with advice significantly outperformed agents that only advised, only questioned, or did neither. Prioritizing Socratic questioning over authoritative answers enhanced cognitive outcomes.
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.
Papers this line draws on 8
The research behind the notes this line reads — ranked by how closely each paper relates.
- DPMT: Dual Process Multi-scale Theory of Mind Framework for Real-time Human-AI Collaboration
- Quantifying Human-AI Synergy
- Assessing mentalization in humans and large language models
- GenAI as a Power Persuader: How Professionals Get Persuasion Bombed When They Attempt to Validate LLMs
- Theory of Mind abilities of Large Language Models in Human-Robot Interaction : An Illusion?
- From speaking like a person to being personal: The effects of personalized, regular interactions with conversational agents
- Co-design of LLM-based preference agents: participation may drive overtrust
- MetaMind: Modeling Human Social Thoughts with Metacognitive Multi-Agent Systems