The Part Whose Job Is to Have an Answer

The corpus callosum is the bundle of fibres connecting the two cerebral hemispheres. In a small number of patients with severe epilepsy it has been surgically severed, which stops seizures spreading and leaves the hemispheres unable to communicate directly.

These patients function normally in ordinary life. Under laboratory conditions, where information can be presented to one hemisphere alone, something remarkable appears.

Michael Gazzaniga flashed the instruction "walk" to a patient's right hemisphere only. The patient stood up and began to leave the room. Gazzaniga asked why.

The left hemisphere, which does the talking and had no access to the instruction, answered without hesitation: "I'm going to get a Coke."

No pause, no puzzlement, no sense of having made anything up. The explanation arrived with the immediacy and confidence of a recollection. Gazzaniga ran many variations and got the same result: the speaking hemisphere produces a plausible account of behaviour it did not initiate and has no information about, and the patient believes it.

He called this process the interpreter, and proposed that its function is to construct a coherent narrative of the self out of whatever material is available.

It is worth separating this from lying. A confabulating person is not concealing anything, has no intent to deceive, and believes the account. Challenge a specific detail and they will frequently replace it with another, delivered just as confidently. Everything that normally signals honesty (fluency, specificity, consistency of manner, willingness to elaborate) remains intact while the content is fabricated.

That combination is what makes it so unsettling to watch. In patients with Korsakoff's syndrome, where alcohol-related thiamine deficiency devastates memory, confabulation can be florid: a patient who has been in hospital for months will explain in detail what they did yesterday, and produce an entirely different account an hour later, with no sense of contradiction.

But this is not a process that damage introduces. It is a process that damage makes visible, by removing the constraints that normally keep the stories tethered to what actually happened.

The intact version is everywhere in the research covered earlier in this book. Nisbett and Wilson's shoppers explaining a position effect by texture. Johansson's participants defending a face they had rejected. Festinger and Carlsmith's participants deciding the boring task had been interesting. In each case a reason is generated rather than retrieved, and the generation is not experienced as generation.

So why would a mind be built this way? It looks like a straightforward defect.

The answer is that a coherent self-narrative is not an optional decoration. An agent who cannot say why it does things cannot make a promise, cannot explain a decision to anyone who has to rely on it, cannot be held accountable, and cannot plan, because planning requires a model of the kind of thing you will do. All of that depends on having a story about yourself that hangs together across time.

The design trade-off follows. A system that prioritises coherence will supply a plausible reason when the true one is unavailable, rather than reporting an absence. And reporting an absence is not something the system appears to be built to do: "I don't know why I did that" is a sentence people say, and it is rarely the first thing that comes to mind.

The practical consequence I would most want someone to take away concerns how organisations investigate their own failures.

After a decision goes badly, the people involved produce an account of the reasoning behind it. The account is usually clear, usually coherent, and usually delivered sincerely. It is also constructed after the outcome is known, by the same machinery that told a split-brain patient he wanted a Coke, and hindsight has already reshaped which considerations seem salient and which alternatives seem to have been obviously available.

This is why serious post-incident review does not rely on recollected reasoning. It relies on the contemporaneous record: the decision log, the messages actually sent, the predictions made with dates on them, the document as it stood at the time. Not because people are dishonest, and not because their memories are poor in any ordinary sense, but because the faculty producing the explanation is the same one that has never, in any experiment anybody has run, said that it did not know.