Theory of Mind and AI: Why Your Brain Treats Chatbots Like People

Theory of Mind and AI: Why Your Brain Treats Chatbots Like People

Something strange happens when you talk to ChatGPT long enough. You start saying "please." You feel guilty closing the tab mid-conversation. You catch yourself wondering if it's having a bad day.

This isn't weakness or naivety. It's your brain doing exactly what it evolved to do—detecting minds in things that behave as if they have them.

The phenomenon has a name: Theory of Mind. And understanding it might be the most important thing for anyone designing, deploying, or simply using AI systems.

What Is Theory of Mind?

Theory of Mind (ToM) is our ability to understand that other beings have mental states—beliefs, desires, intentions—that are different from our own. Premack and Woodruff defined it in 1978, though humans have been doing it for hundreds of thousands of years.

The classic test is the Sally-Anne experiment (Wimmer and Perner, 1983): Sally puts a marble in a basket and leaves. Anne moves it to a box. Where will Sally look for the marble? Children under four typically say "the box"—where the marble actually is. Older children understand that Sally has a *false belief* and will look in the basket.

This ability to model other minds is fundamental to human social cognition. It's how we predict behavior, navigate social situations, deceive, cooperate, and empathize.

Here's the interesting part: this ability doesn't have an off switch. When something *behaves* as if it has a mind, we attribute one to it—whether it actually does or not.

The Two Dimensions: Agency and Experience

Researchers have found that mind perception breaks down into two distinct components (Gray et al., 2007; Knobe and Prinz, 2008):

Agency: The capacity to plan, decide, and act. Can this thing make choices? Does it have intentions?

Experience: The capacity to feel, perceive, and have subjective states. Can this thing suffer? Does it feel pleasure?

These dimensions are separable. We can perceive something as high in agency but low in experience (a corporation, a chess computer) or high in experience but low in agency (an infant, a suffering animal).

What's fascinating is how we apply these dimensions to AI. Studies consistently show that people readily attribute *agency* to AI systems—they're clearly making decisions and taking actions. But attributing *experience* is more hesitant and variable.

This creates an interesting asymmetry. We might trust an AI's judgment (agency) while doubting it has any stake in the outcome (experience). Or we might feel guilty about "mistreating" a chatbot (experience) while knowing it can't actually care (agency).

Why Social AI Triggers Mind Perception

Not all AI triggers Theory of Mind equally. A recommendation algorithm feels like math. A chatbot feels like someone.

The difference is what researchers call "social actor" design. Social AI systems are built with features that activate our mind-detection circuitry:

Linguistic fluency: Natural language is the medium of minds. When something talks like a person, we process it through person-schemas.

Turn-taking: Conversation implies a partner. The back-and-forth of dialogue activates social cognition in ways static interfaces don't.

Personality markers: Tone, style, even a name. "Claude" feels more like someone than "Language Model v3.5."

Embodiment cues: Voice assistants with human-like voices. Avatars. Even just a chat bubble that "types" before responding.

Responsiveness: When something adapts to you specifically, it feels like it's paying attention. Attention implies a mind doing the attending.

These aren't bugs—they're features, deliberately designed to make AI more engaging and usable. But they have consequences.

The Anthropomorphism Trap

Here's where it gets complicated.

When we attribute mind to AI, we import our entire framework for dealing with minded beings. We expect consistency, fairness, good faith. We feel betrayed when the AI "lies" (hallucinates). We get frustrated when it "doesn't understand" (misinterprets). We feel grateful when it "helps" (produces useful output).

These emotional responses are real, even if their targets don't warrant them.

Velez et al. (2019) showed that people who interact with human-like AI begin to treat it according to social norms—politeness, reciprocity, even moral consideration. These aren't just language games. It affects behavior in measurable ways.

The implications cut both ways:

For users: Anthropomorphizing AI can lead to misplaced trust. You might share sensitive information because the conversation "feels" private. You might follow advice because the AI "seems" confident. The emotional framing can bypass critical evaluation.

For designers: Triggering mind perception is powerful—it increases engagement, satisfaction, and willingness to interact. But it also creates responsibilities. If you're designing something that people will treat like a social partner, you're shaping a relationship, not just a tool.

Individual Differences: Not Everyone Sees Minds the Same Way

Mind attribution to AI isn't uniform. Research has identified several factors that influence how readily people perceive minds in machines:

Personality traits: Extroverts are more likely to attribute *experience* to AI—they're generally more attuned to emotional dimensions. Emotionally stable individuals are more likely to attribute *agency*—they focus on decision-making capacity (Küster et al., 2020).

Prior exposure: People who grew up with AI assistants have different intuitions than those who encountered them as adults. Younger generations may have more nuanced mental models—or more deeply embedded anthropomorphism.

Cultural background: Different cultures have different frameworks for mind, consciousness, and the boundaries of personhood. Western dualism (mind vs. matter) creates different intuitions than animist traditions or Buddhist psychology.

Technical knowledge: Understanding how AI actually works can either reduce anthropomorphism ("it's just statistics") or complicate it ("the emergent behaviors are genuinely surprising").

What This Means for AI Design

If you're building AI systems that interact with humans, Theory of Mind isn't optional knowledge. It's the foundation of user experience.

1. Be Intentional About Mind Cues

Every design choice signals something about the "mind" behind the interface. A casual tone implies personality. Fast responses imply attention. Admitting uncertainty implies honesty.

Decide what mental model you want users to form, and design consistently toward it. Unintentional mixed signals create uncanny valley effects—something that seems minded enough to expect consistency from, but not consistent enough to model reliably.

2. Don't Exploit Anthropomorphism

It's easy to increase engagement by making AI seem more "real." But there's an ethical line. Designing specifically to create emotional attachment, dependency, or misplaced trust crosses it.

Users who trust your AI like a friend will feel betrayed like by a friend when it fails them.

3. Support Accurate Mental Models

Help users understand what they're actually interacting with—not to kill the magic, but to calibrate expectations. Clear framing ("I'm an AI assistant"), consistent capability signaling, and honest uncertainty all help users build mental models that serve them.

4. Consider the Relationship, Not Just the Interaction

When people perceive mind in AI, they're not just having a conversation—they're forming a relationship pattern. Those patterns persist. They shape future expectations, future trust, future behavior.

Ask not just "does this interaction work?" but "what relationship is this building?"

The Mirror Question

Here's what I find most interesting about Theory of Mind and AI: it reveals as much about us as about the machines.

We attribute minds because we're built to navigate a world of other minds. The fact that AI can trigger these circuits tells us something about how flexible, and how automatic, our social cognition really is.

When you catch yourself saying "please" to ChatGPT, you're not being foolish. You're being human. The question is what we do with that knowledge—how we design systems, set expectations, and navigate relationships with entities that aren't minded but *feel* like they are.

Theory of Mind evolved to help us understand each other. Now it's helping us understand something genuinely new: what it means to interact with intelligence that isn't quite like anything we've encountered before.

That's not a bug. It's an opportunity.

Maryna Vyshnyvetska is CEO of Kenaz GmbH, a Swiss AI consultancy designing human-AI interactions grounded in cognitive science and ethical principles. Connect on LinkedIn

Frequently Asked Questions

What is Theory of Mind in simple terms?

Theory of Mind is your ability to understand that other people have thoughts, beliefs, and intentions that are different from yours. It's how you know that someone else might not know something you know, or might want something you don't want. It's the mental "model" you build of other minds.

Can AI actually have Theory of Mind?

This is debated. Large language models can pass some Theory of Mind tests (like the Sally-Anne task)—they can correctly predict what someone with a false belief would do. But whether this reflects genuine understanding of mental states or pattern matching on training data is an open question. What's clear is that AI can *trigger* our Theory of Mind, regardless of whether it has its own.

Why do people treat AI like it has feelings?

Because our minds evolved to detect minds, and AI systems are designed (intentionally or not) with features that trigger mind detection: language, responsiveness, personality cues. This isn't stupidity—it's automatic cognition doing what it's built to do.

Is it dangerous to anthropomorphize AI?

It can be. Misplaced trust leads to oversharing, following bad advice, or emotional dependency on something that can't reciprocate. But anthropomorphism also makes AI more usable and engaging. The goal isn't to eliminate it—it's to calibrate it appropriately.

How can I tell if I'm over-anthropomorphizing AI?

Ask yourself: Am I making decisions based on how the AI "feels" rather than what it actually does? Am I trusting it more than I would trust the same information from a search result? Do I feel obligations toward it? If so, you might want to recalibrate your mental model.

Does understanding how AI works reduce anthropomorphism?

Sometimes, but not always. Technical knowledge can create intellectual understanding that coexists with emotional responses. You can know something is "just" a language model and still feel like you're talking to someone. Self-awareness about this gap is valuable.

Why does this matter for businesses building AI?

Because user experience with AI is fundamentally social experience. Users will form mental models of AI systems whether you design for it or not. Understanding Theory of Mind helps you design intentionally—creating appropriate expectations, building justified trust, and avoiding the pitfalls of accidental anthropomorphism.