Theory of Mind and Gaze

People treat other people's eyes as though invisible beams of force emanate from them, gently pushing on objects in the world. They do this unconsciously, consistently, and even when they explicitly reject the idea that vision works by eye-beams. This finding, from a series of experiments by Arvid Guterstam and colleagues at Princeton, is one of the strangest and most revealing results in social cognition research.¹

The Eye-Beam Illusion

The experimental setup is elegant. Participants judge the angle at which a paper tube would topple over. A photograph of a face appears to one side, looking at the tube. When the face has open eyes, participants consistently judge that the tube can be tilted more steeply toward the face before falling — as though the gaze were exerting a gentle supporting force, like a light puff of air. The implied force is about a hundredth of a newton. When the eyes are blindfolded, the asymmetry disappears. When the face is turned away from the tube, it disappears. When participants are told the person is looking past the tube at the far wall, it disappears. When the tube is described as heavy concrete instead of light paper, it disappears.¹

That last control is the clincher. If the effect were a low-level visual tilt illusion caused by the asymmetric presence of a face, it should persist regardless of the tube's described weight. The fact that it vanishes for heavy objects is exactly what you'd predict if participants are implicitly modelling a weak force from the eyes — a force that could nudge paper but not concrete.

Nobody in any of the experiments guessed the purpose of the study. Only about 5% of participants reported any explicit belief in the extramission theory of vision (that something flows out of the eyes). The other 95% correctly described vision as caused by light entering the eye — and showed the eye-beam effect anyway. The implicit model persists regardless of explicit knowledge.¹

The Deep Roots of Extramission

The belief that eyes emit something has appeared across virtually every human culture. Ancient Greek philosophers proposed extramission theories. The evil eye persists in Mediterranean, Middle Eastern, and South Asian traditions. Piaget found that most children hold a naive extramission belief. Earlier studies claimed 50-60% of American college students did too, though Guterstam's sample puts the figure much lower, around 5% for explicit endorsement.¹

The universality of the belief, and particularly its universality in children, suggests it's not a prescientific error but the surface expression of a deeper cognitive mechanism. The brain constructs a simplified but functional model of other people's vision — a model that represents gaze as an active process emanating from the agent, capable of physically affecting the world. This model is wrong about the physics but useful for social cognition. It helps you track where other people are looking, what they're attending to, and what they might do next.

This connects to a broader point about how the brain represents other minds. We don't build accurate physics simulations of other people's optical systems. We build simplified, schematic models that capture the functionally relevant features — direction, intentionality, agency — and discard the rest. The eye-beam model is a case of useful fiction, much like the desktop interface that Hoffman proposes for perception more generally. The brain doesn't need to know how vision actually works in order to predict gaze direction and its behavioural consequences. A force-beam model does the job.

Gaze and Social Agency

The findings have a sharp edge for thinking about social cognition and theory of mind. What we call "reading" other people — understanding their attention, intentions, desires — may rely less on sophisticated mentalising and more on fast, automatic, physics-like models that represent mental states as quasi-physical forces. Gaze as a push. Attention as a beam. Desire as a pull.

This raises the question of what else we implicitly model about other minds using physical metaphors that don't survive scientific scrutiny. Do we unconsciously treat willpower as a finite substance (explaining why willpower depletion felt so intuitively right before it failed to replicate)? Do we model beliefs as objects that can be "held" or "dropped" — and does that metaphorical scaffolding shape our actual social cognition?

The eye-beam result suggests that social perception, like visual perception, is a construction — not a transparent window onto other minds but a functional model built from heuristics and shortcuts. The construction is invisible to us. We don't feel ourselves attributing force to gaze any more than we feel ourselves constructing the visual scene from retinal fragments. But the construction shapes our behaviour in measurable ways, right down to the angle at which we think a tube will fall.

The fact that even babies track gaze direction and prefer to look at faces with open eyes suggests the gaze-as-force model is either innate or develops so early and universally that the distinction barely matters. We are, from the very beginning, minds that model other minds — and the models we build are physical, simplified, and wrong in exactly the ways that make them useful.

The Empathy Trap

Paul Bloom's case against empathy extends this line of thinking in an uncomfortable direction. If social perception is a construction rather than a transparent window onto other minds, then empathy — the celebrated ability to "feel what others feel" — might be a less reliable moral compass than it appears.²

Bloom's sharpest example is the character of O'Brien in Orwell's 1984. O'Brien's capacity to discern Winston Smith's responses is exquisitely refined — he can divine Winston's greatest dread (rats) without Winston ever articulating it. This is empathy in its purest form: the accurate modeling of another mind's internal states. And O'Brien uses it to destroy Winston. A low score on the empathy index is commonly associated with psychopathy, but many psychopaths are supremely able to model others' feelings, which is exactly what makes them effective torturers.²

The deeper problem is that empathy is "innumerate and biased." We empathize more with people who resemble us or whom we find attractive. A photograph of one suffering child triggers a global response; the statistical knowledge that thousands of equally affected children exist does not. Bloom argues this makes empathy actively dangerous as a guide to policy — it narrows focus, distorts priorities, and can be manipulated by anyone who knows how to present a compelling individual case.

What Bloom advocates instead is rational compassion — sympathy without identification. A surgeon who felt her patient's pain couldn't operate. A parent who over-identifies with a child's distress can't enforce the short-term suffering (the cold swimming pool, the dentist's chair) that serves the child's long-term interests. The distinction between empathy and sympathy matters: sympathy is an imaginative attempt to sense another's otherness without claiming to own their experience, while empathy purports to collapse the distance between self and other — a collapse that may be neither possible nor desirable.²

This connects back to the eye-beam illusion. We model other minds using simplified physical metaphors — gaze as force, attention as beam, desire as pull. Empathy may be another such metaphor: the feeling that we've bridged the gap between self and other, when what we've actually done is run a simulation inside our own prediction engine, constrained by our own biases and experiences. The simulation can be useful. But mistaking it for direct access to another mind is a construction error of the same kind that makes us think eyes push tubes.