Most AI literacy training focuses on understanding how AI works, recognizing its limitations, and crafting effective prompts. But new empirical research suggests there’s a crucial dimension we’ve been overlooking: the social-cognitive skills needed to collaborate effectively with AI.
The key? A surprisingly human ability called Theory of Mind.
What Is Theory of Mind?
Theory of Mind (ToM) is our ability to understand that others have their own thoughts, beliefs, knowledge, and intentions; different from our own. It’s what allows you to explain a concept differently to a child versus an adult, or to realize when someone doesn’t necessarily have the context which they need to understand what you’re saying and to adapt your method of explaining or questioning accordingly.
ToM is fundamentally a social skill – it’s about understanding and working with others effectively. We use it constantly in human interactions, often without thinking about it. But recent research shows that this same social-cognitive skill is critical when working with AI. This parallels a well-established finding in education and career research: academic performance and intelligence alone don’t predict success in business and life. Multiple studies have shown that social skills, like communication, teamwork, and the ability to build relationships, are equally important, and sometimes more predictive of career success than grades or test scores. Research has demonstrated that students with strong social networks during school earn more in adulthood, and that personality traits like conscientiousness (which can be developed) matter as much as IQ for job performance.
What Makes This Research Different?
The idea that mental models matter for human-AI interaction has been gaining momentum since around 2019, with researchers exploring how people’s understanding of AI affects collaboration. What’s novel here is the empirical proof at scale that Theory of Mind specifically, not just general mental models, drives successful collaboration.
A comprehensive study of 667 people working with AI on complex problems revealed three key findings:
1 1. Being good at working with AI is a different skill than being good at working alone.
The researchers could measure and separate these abilities; they’re not the same thing. Someone might excel at solo problem-solving but struggle to get good results from AI, while another person might be average on their own but exceptional when collaborating with AI.
2. People with stronger ToM performed significantly better when working with AI compared to people with weaker ToM; but this advantage disappeared when they worked alone.
Think of it this way: someone might be a brilliant solo analyst who can solve complex problems independently, but if they lack the social-cognitive skills to explain their needs, provide context, and adapt their communication, they’ll struggle to get good results from AI. Meanwhile, someone who might be average at solo work but excels at collaboration, understanding what information others need, adjusting their explanations, reading between the lines, will get dramatically better results from AI. This proves that ToM isn’t about being “smarter” in general; it’s specifically a collaboration skill that enables effective human-AI teamwork.
3. ToM isn’t just a fixed trait you either have or don’t have.
The quality of AI responses varied based on how much perspective-taking a person used in each individual interaction. When someone took more time to consider what the AI needed to know, the AI gave better answers, even if that same person had given less context in their previous question.
The bottom line: treating AI like a collaborative partner, rather than a simple tool,makes all the difference.
The Equalizing Power of ToM
Perhaps most importantly, the research reveals that AI combined with ToM can be a great equalizer. Lower-skilled users who apply ToM effectively can experience dramatic performance gains (up to 34% improvement in
some studies), allowing them to perform at levels approaching naturally talented individuals. Meanwhile, high-skilled users still improve, but gain less percentage-wise. Someone brilliant at solo work may even underachieve with
AI if they lack the collaborative skills to communicate effectively with it. This mirrors a well-established finding from career research: academic brilliance doesn’t guarantee business success without social skills. The same principle applies here—individual ability alone doesn’t guarantee AI collaboration success without Theory of Mind. AI acts as a cognitive amplifier that’s particularly powerful when combined with strong collaborative skills, regardless of your baseline ability.
Why This Changes How We Teach AI Literacy
Researchers have long recognized that mental models matter for AI interaction. What’s changed is that we now have empirical evidence showing which specific cognitive skill matters most and that it operates independently from general intelligence or problem-solving ability.
This has concrete implications for AI literacy education:
1. Mental Model Building
- • Understanding what the AI “knows” and doesn’t know
- • Recognizing when you need to provide additional context
- • Adjusting your approach based on the AI’s responses
2. Strategic Delegation
- • Knowing when and how to hand off tasks to AI
- • Deciding whether to accept, challenge, or refine AI suggestions
- • Creating a division of cognitive labour (like effective human teams do)
3. Adaptive Dialogue
- • Treating interaction as a conversation, not just commands
- • Providing clarification when the AI seems confused
- • Building on previous exchanges rather than starting fresh each time
Practical Takeaways for Everyone
You don’t need to be a technical expert to improve your AI collaboration. Here’s what the research suggests:
Be Explicit About Context Don’t assume the AI shares your background knowledge. Instead of “answer this question,” try “I’m working on a physics problem about momentum. I understand the basic
concept but get confused when multiple objects are involved. Here’s the specific question…”
Challenge and Refine High-ToM users point out mistakes, ask for justifications, and provide feedback. If an AI response doesn’t quite fit your needs, don’t just accept it – engage: “That’s helpful, but I was actually asking about X, not Y. Can you focus on that aspect?”
Use “Chain of Thought” Prompting Ask the AI to think step-by-step, especially for complex problems. This mirrors how you’d work through a problem with a human colleague: “Let’s break this down. First, what are the key factors we need to consider?”
Establish Rapport and Goals Simple phrases like “Help me understand…” or “I’m trying to accomplish…” signal your intentions and create a more collaborative frame. State your knowledge level (“I’m a beginner in this area”) and your goals (“I need a comprehensive explanation, not just the answer”).
Treat It as Dynamic Your effectiveness varies based on your cognitive state. When struggling, reset – take a moment to reconsider what the AI needs to know and how you’re framing your request.
The Bigger Picture
This research reveals something profound: effective AI collaboration is less like using a calculator and more like managing a sophisticated colleague. The AI won’t always understand your unstated assumptions, may need reminding about previous context, and benefits from clear communication about your goals.
But here’s the truly exciting part: these skills can level the playing field. While high performers will always have advantages, AI combined with strong Theory of Mind skills allows average performers to punch well above their weight class. The research shows that the performance gap between skilled and less-skilled workers narrows dramatically when both use AI effectively; AI tends to reduce output inequality and helps create more equitable outcomes.
The good news? These are learnable skills. By consciously applying Theory of Mind, by thinking about what your AI partner needs to know and adapting your interaction accordingly, you can dramatically improve your collaborative outcomes.
As AI systems become more integrated into our work and daily lives, the ability to collaborate effectively with them may become as important as any technical skill. And that ability starts with recognizing that even
though AI isn’t human, working with it successfully requires very human cognitive and social skills.
The future of AI literacy isn’t just about understanding technology: it’s about understanding how to think with technology. And that means bringing our most sophisticated social intelligence to the table, even when our partner is artificial.
References
Anonymous authors. (under review). Quantifying human-AI
synergy. Paper under double-blind review.
Matta, D. (2024). AI and Theory of Mind. Unpublished
manuscript, May 1, 2024.
Article based on research using a novel Bayesian
framework to quantify human-AI synergy across 667 participants, providing the
first large-scale empirical evidence that Theory of Mind capabilities
significantly and specifically predict collaborative performance with AI
systems.