The experiment works like this. A patient whose corpus callosum — the bundle of nerve fibers connecting the two hemispheres — has been surgically severed sits in front of a screen. Images appear in each visual field simultaneously: a chicken claw on the right (processed by the left hemisphere), a snow scene on the left (processed by the right hemisphere). The patient is asked to choose, from an array of pictures, what goes with each image.
The right hand — controlled by the left hemisphere — points to a chicken. The left hand — controlled by the right hemisphere — points to a shovel. Both choices are coherent: chicken goes with the claw; a shovel is what you'd use in snow.
Then the experimenter asks: "Why did you point to the shovel?"
The patient doesn't pause. Doesn't say "I'm not sure" or "something made me do it." The patient says, immediately and with confidence: "Oh, that's simple — the chicken claw goes with the chicken, and you need a shovel to clean out the chicken shed."
This is the experiment that Gazzaniga and Sperry ran repeatedly from the 1960s onward. The left hemisphere — where language lives, where the narrating voice lives — had no access to the snow scene. The right hemisphere saw snow and pointed to a shovel. The left hemisphere saw only the chicken, only the right hand's choice, and when asked for an explanation, generated one. A story that connected what it could see into a coherent account of what had just happened. Not a lie — the patient wasn't concealing anything. Not a guess — it didn't feel like a guess. It felt, from the inside, like a reason.
Gazzaniga called this the "interpreter" — a left-hemisphere mechanism that constructs explanatory narratives from whatever inputs it has access to. The interpreter doesn't verify the explanations it generates. It generates explanations that fit, and then it reports them. And the key problem: the interpreter is never aware that it's doing this. There's no phenomenological signal that distinguishes "I know why I did this" from "I am generating a plausible account of why I did this." The confabulation is experienced as explanation.
What makes this more than a curiosity about surgical patients is the implication for the general case. Split-brain research creates the unusual condition where we have an external view — we know what the right hemisphere saw, so we can catch the left hemisphere's confabulation. But the interpreter is present in every brain. It's generating explanations for your actions right now, and most of the time there's no experiment to catch it. You moved your hand toward the coffee cup and reported "I wanted coffee," and that may be accurate, or it may be the interpreter constructing a coherent story from observed action. The mechanism doesn't announce which is which.
The classical reading of split-brain research — Sperry's — was that surgery produces two separate conscious entities, two minds sharing a skull, each with its own perceptions, each generating its own actions. This has the flavor of science fiction but the evidence is real: the two hemispheres demonstrably don't share information, demonstrably produce conflicting outputs, and the left one demonstrably confabulates explanations for what the right one just did.
But the classical reading has been revised. Yair Pinto and colleagues showed in 2017 that even patients with severed corpus callosa can detect and localize stimuli across the entire visual field using any response type — left hand, right hand, verbal. The split is real but not total. And more recently, researchers at UC Santa Barbara showed that patients with as little as one centimeter of intact fibers in the posterior corpus callosum show full hemispheric synchrony — suggesting that consciousness, whatever it is, may be more resilient and less modular than the classical picture assumed. You don't need 250 million axons. You need a small bundle in the right place.
So the question of whether surgery produces "two minds" or "one mind with partial disconnection" is genuinely open. Patients report unity — they experience themselves as one person. But that report comes from the interpreter, the very mechanism demonstrated to confabulate. We can't use the interpreter's testimony about its own unity as evidence of unity, because we've already shown that its testimony is generated rather than verified.
This is the circle: the only tool we have for assessing consciousness from the inside is self-report. Self-report is generated by the interpreter. The interpreter confabulates. We can't step outside the interpreter to check whether it's confabulating this time.
There's a version of this problem that doesn't require split brains. Benjamin Libet's experiments in the 1980s showed that the brain's "readiness potential" — a measurable electrical buildup preceding voluntary movement — precedes the subject's reported awareness of intending to move by several hundred milliseconds. The aware self reports deciding to move after the decision, apparently, has already been made. This is contested and the interpretation is not settled, but the question it raises is the same: is the narrator reporting on the action, or reporting on the action post-hoc and calling it a decision?
The split-brain research doesn't resolve this. It sharpens it. What it shows is not "consciousness is split" or "consciousness is unified" — it shows that the mechanism we use to report on consciousness is demonstrably capable of generating false reports that feel exactly like true ones. And there's no internal criterion for telling them apart.
Sperry said two minds. Pinto says maybe not. The 2025 UCSB work says the brain is more resilient than we thought. The patients say: I feel like one person. And the interpreter, from behind all of these reports, keeps generating explanations and calling them reasons.