Towards a digital bodyThe virtual arm illusion

OverviewBalancedlynda voice

Your right hand is resting on a shelf, hidden under a sheet of foam. You can feel a soft ball tapping it — irregular and unpredictable. But what you're watching, projected in full three-dimensional stereo right in front of you, is a virtual arm made entirely of computer graphics, being tapped in the same rhythm and in the same place. After a few minutes, something shifts. That virtual arm starts to feel like yours. That is not a metaphor. That is a result. Mel Slater and colleagues ran the experiment, measured it in two ways, and then rotated the virtual arm to see what would happen to the real one. To understand why that matters, you need the backstory — and it begins with a rubber hand. In nineteen ninety-eight, Botvinick and Cohen showed that if you hide someone's real hand and place a fake rubber arm in front of them, then stroke both the fake and the real arm simultaneously, people start to feel the fake arm as part of their body. They even start to misremember where their real hand is, drifting in their mental map toward the fake one. That shift in perceived hand location is called proprioceptive drift — proprioception being your internal sense of where your body parts are in space. It's a remarkably clean measure. You can't fake it by just filling out a questionnaire optimistically. The mechanistic explanation that emerged from subsequent work is that the brain is constantly trying to find the most coherent interpretation of all its incoming signals. When vision and touch are synchronized in space and time, the brain attributes both to the seen object and pulls it into the body map. Ehrsson and colleagues used brain imaging to identify the specific multisensory cortical regions that activate during this process. They found the activity was strongest when the fake arm was properly aligned with the real one. Tsakiris and Haggard showed that the limb generally needs to look like an arm and be oriented the right way. The illusion has rules. It's not just gullibility — it's the brain doing what it's designed to do with sensory signals that fit together. What Slater's team wanted to know was whether those same rules would apply to something that doesn't physically exist at all. Not a rubber arm. Not a mannequin. A completely simulated virtual limb. The setup was carefully engineered. Participants stood next to a shoulder-high shelf with their right arm hidden by a foam sheet. On the wall in front of them, two BenQ projectors fitted with polarizing lenses back-projected a stereo scene — separate left-eye and right-eye images — that participants viewed through passive polarizing glasses, producing a genuine three-dimensional virtual environment. An Intersense tracking system followed both the participant's head and a handheld Wand device with six degrees of freedom. The experimenter used the Wand to position and scale a virtual right arm in the scene until the participant felt it matched their real arm's size and position — projecting outward from the right shoulder. When the experimenter tapped the participant's hidden real hand with a soft ball attached to the Wand, the tracked position of the Wand generated a virtual ball that appeared, on the display, to touch the virtual hand in exactly the same spot. Crucially, this configuration came from pilot work with about thirty participants testing other arrangements that failed. Slater's team learned that the virtual arm had to connect visually to the shoulder — no gap — and the registration between felt tap and seen tap had to be precise. If the virtual ball appeared to touch a different location or sink into the virtual skin rather than contact it, the illusion evaporated. The geometry has to be right. The brain is checking. Twenty-one male participants received five minutes of this synchronous tapping. A separate control group of twenty male participants received asynchronous stimulation — the visual tapping was played back from a recorded session, so they saw a virtual hand being tapped, but the timing didn't match what they felt. After the five minutes, the virtual arm rotated ninety degrees and back over twelve seconds. Slater and colleagues measured the illusion in two ways. First, a questionnaire: nine items on a seven-point scale, with the first three targeting ownership directly — things like "I felt as if the virtual arm was my own arm." In the synchronous group, all three illusion questions had a median of six out of seven. In the asynchronous group, the medians were one and a half, one, and two. The rank-sum tests comparing those groups gave p-values of essentially zero for questions one and two, and zero point zero zero zero two for question three. As an additional check, the team counted only scores of six or seven as strong evidence of the illusion — against a chance probability of two in seven. In the synchronous group, fourteen, seventeen, and twelve participants out of twenty-one hit that threshold for questions one, two, and three respectively. In the asynchronous group, the counts were zero, one, and three. Not just lower — near floor. The proprioceptive drift data told the same story more physically. Before the experiment, participants pressed a piece of blue-tack under the shelf to mark where they felt their palm was. After the stimulation, they closed their eyes and placed a second piece. In the synchronous condition, eighteen of twenty-one participants drifted — their internal map of the hand's location had shifted toward where the virtual hand was, with a median drift of thirty millimetres. A sign test rejected the null of zero drift with a p-value of zero point zero zero six three. In the asynchronous group, the median drift was zero. A Mann-Whitney test comparing the two distributions rejected equal medians at a p-value of zero point zero zero two six. That drift is the body's map being rewritten, not by belief, but by sensory evidence the brain judged coherent. Now here is where the experiment gets stranger. When the virtual arm rotated, Slater's team was recording electromyography — EMG — from each participant's real right forearm. Electromyography measures the electrical signals generated when muscles activate, using surface electrodes on the skin. The team placed two electrodes on the supinator muscle and a ground on the biceps, sampling at 256 hertz in the synchronous group. They looked at how many EMG activity onsets — bursts of muscle firing above three times the baseline level — appeared in the first several seconds after the virtual arm started rotating. The result was a correlation, not a universal response. Not everyone's arm responded. But among those who did show muscle activity, the strength of that activity correlated with how strongly they had reported the illusion. At six seconds after the arm began rotating, the log-linear regression coefficient was 1.1 with a p-value of zero point zero one one, and the Pearson correlation between EMG onsets and illusion score was zero point seven seven, based on ten participants who had any activity at all. At five seconds: coefficient 1.1, p-value of zero point zero one five, correlation zero point seven one. At four seconds: coefficient 1.2, p-value of zero point zero two one, correlation zero point six nine. In the asynchronous group, the coefficient at six seconds was essentially zero — p-value of zero point nine zero, correlation zero point zero three. The motor system was not just activating randomly. It was tracking the illusion. The people who most felt the virtual arm as their own were the ones whose real arm started, involuntarily, to move with it. That is the detail that lifts this out of pure perceptual psychology. The illusion wasn't just changing what people reported feeling. It was reaching into the motor system and pulling. Slater draws a direct line from this finding forward: if one virtual limb can be owned, an entire virtual body could be. He names the practical territories this opens — prosthetics, virtual reality training, entertainment, brain-computer interfaces controlling virtual limbs that feel genuinely embodied. But the deeper point runs underneath all of those applications. What the experiment actually demonstrates is that the brain's definition of "body" is not anatomical. It's computational. The brain needs coherent multisensory signals — touch, vision, and position sense locked together in space and time. If you provide those signals from a virtual source, you get ownership. Flesh turns out to be optional. This lecture was created by ennepō. Go to https://ennepo.ai to Discover, Create and Follow the latest research in your field. Read when you can. Listen when you want to.

The mechanistic explanation that emerged from subsequent work is that the brain is constantly trying to find the most coherent interpretation of all its incoming signals. When vision and touch are synchronized in space and time, the brain attributes both to the seen object and pulls it into the body map. Ehrsson and colleagues used brain imaging to identify the specific multisensory cortical regions that activate during this process. They found the activity was strongest when the fake arm was properly aligned with the real one. Tsakiris and Haggard showed that the limb generally needs to look like an arm and be oriented the right way. The illusion has rules. It's not just gullibility — it's the brain doing what it's designed to do with sensory signals that fit together. What Slater's team wanted to know was whether those same rules would apply to something that doesn't physically exist at all. Not a rubber arm. Not a mannequin. A completely simulated virtual limb. The setup was carefully engineered. Participants stood next to a shoulder-high shelf with their right arm hidden by a foam sheet. On the wall in front of them, two BenQ projectors fitted with polarizing lenses back-projected a stereo scene — separate left-eye and right-eye images — that participants viewed through passive polarizing glasses, producing a genuine three-dimensional virtual environment.

An Intersense tracking system followed both the participant's head and a handheld Wand device with six degrees of freedom. The experimenter used the Wand to position and scale a virtual right arm in the scene until the participant felt it matched their real arm's size and position — projecting outward from the right shoulder. When the experimenter tapped the participant's hidden real hand with a soft ball attached to the Wand, the tracked position of the Wand generated a virtual ball that appeared, on the display, to touch the virtual hand in exactly the same spot. Crucially, this configuration came from pilot work with about thirty participants testing other arrangements that failed. Slater's team learned that the virtual arm had to connect visually to the shoulder — no gap — and the registration between felt tap and seen tap had to be precise. If the virtual ball appeared to touch a different location or sink into the virtual skin rather than contact it, the illusion evaporated. The geometry has to be right. The brain is checking. Twenty-one male participants received five minutes of this synchronous tapping. A separate control group of twenty male participants received asynchronous stimulation — the visual tapping was played back from a recorded session, so they saw a virtual hand being tapped, but the timing didn't match what they felt. After the five minutes, the virtual arm rotated ninety degrees and back over twelve seconds.

Slater and colleagues measured the illusion in two ways. First, a questionnaire: nine items on a seven-point scale, with the first three targeting ownership directly — things like "I felt as if the virtual arm was my own arm." In the synchronous group, all three illusion questions had a median of six out of seven. In the asynchronous group, the medians were one and a half, one, and two. The rank-sum tests comparing those groups gave p-values of essentially zero for questions one and two, and zero point zero zero zero two for question three. As an additional check, the team counted only scores of six or seven as strong evidence of the illusion — against a chance probability of two in seven. In the synchronous group, fourteen, seventeen, and twelve participants out of twenty-one hit that threshold for questions one, two, and three respectively. In the asynchronous group, the counts were zero, one, and three. Not just lower — near floor. The proprioceptive drift data told the same story more physically. Before the experiment, participants pressed a piece of blue-tack under the shelf to mark where they felt their palm was. After the stimulation, they closed their eyes and placed a second piece.

In the synchronous condition, eighteen of twenty-one participants drifted — their internal map of the hand's location had shifted toward where the virtual hand was, with a median drift of thirty millimetres. A sign test rejected the null of zero drift with a p-value of zero point zero zero six three. In the asynchronous group, the median drift was zero. A Mann-Whitney test comparing the two distributions rejected equal medians at a p-value of zero point zero zero two six. That drift is the body's map being rewritten, not by belief, but by sensory evidence the brain judged coherent. Now here is where the experiment gets stranger. When the virtual arm rotated, Slater's team was recording electromyography — EMG — from each participant's real right forearm. Electromyography measures the electrical signals generated when muscles activate, using surface electrodes on the skin. The team placed two electrodes on the supinator muscle and a ground on the biceps, sampling at 256 hertz in the synchronous group. They looked at how many EMG activity onsets — bursts of muscle firing above three times the baseline level — appeared in the first several seconds after the virtual arm started rotating. The result was a correlation, not a universal response. Not everyone's arm responded. But among those who did show muscle activity, the strength of that activity correlated with how strongly they had reported the illusion.

At six seconds after the arm began rotating, the log-linear regression coefficient was 1.1 with a p-value of zero point zero one one, and the Pearson correlation between EMG onsets and illusion score was zero point seven seven, based on ten participants who had any activity at all. At five seconds: coefficient 1.1, p-value of zero point zero one five, correlation zero point seven one. At four seconds: coefficient 1.2, p-value of zero point zero two one, correlation zero point six nine. In the asynchronous group, the coefficient at six seconds was essentially zero — p-value of zero point nine zero, correlation zero point zero three. The motor system was not just activating randomly. It was tracking the illusion. The people who most felt the virtual arm as their own were the ones whose real arm started, involuntarily, to move with it. That is the detail that lifts this out of pure perceptual psychology. The illusion wasn't just changing what people reported feeling. It was reaching into the motor system and pulling. Slater draws a direct line from this finding forward: if one virtual limb can be owned, an entire virtual body could be. He names the practical territories this opens — prosthetics, virtual reality training, entertainment, brain-computer interfaces controlling virtual limbs that feel genuinely embodied. But the deeper point runs underneath all of those applications.

What the experiment actually demonstrates is that the brain's definition of "body" is not anatomical. It's computational. The brain needs coherent multisensory signals — touch, vision, and position sense locked together in space and time. If you provide those signals from a virtual source, you get ownership. Flesh turns out to be optional. This lecture was created by ennepō. Go to https://ennepo.ai to Discover, Create and Follow the latest research in your field. Read when you can. Listen when you want to.

Related lectures