GenAI and Split-Brain Research

During the holiday, I had the opportunity to read up a bit. I enjoyed some thrillers but also tackled a bit of work-related material.

One standout was The Elephant in the Brain by Kevin Simler and Robin Hanson. The book explores an intriguing idea: we don’t always know the “whys” behind our own behavior. But we certainly pretend to.

Split-Brain Experiments

A fascinating demonstration of this phenomenon is the split-brain patient experiments conducted by Roger Sperry and Michael Gazzaniga. These studies involved patients who had undergone surgery severing the connection between their brain hemispheres. In effect, these individuals had two separate consciousnesses within a single skull.

Here’s an example of how their experiments played out:

In a video a boy is asked: who is your favourite girlfriend, with the word girlfriend flashed only to the right hemisphere. As predicted, the boy can’t respond verbally. He shrugs and shakes his head, indicating that he doesn’t see any word, as had been the case with W.J.. But then he giggles. It’s one of those tell-tale teen giggles — a soundtrack to a blush. His right hemisphere has seen the message, but the verbal left-hemisphere remains unaware. Then, using his left hand, the boy slowly selects three Scrabble tiles from the assortment in front of him. He lines them up to spell L-I-Z: the name, we can safely assume, of the cute girl in his class. “That told us that he was capable of language comprehension in the right hemisphere,” Gazzaniga later told me. “He was one of the first confirmation cases that you could get bilateral language — he could answer queries using language from either side.”

This example highlights how our brains fill in gaps in understanding by inventing plausible, but not necessarily accurate, explanations.

Now for GenAI

This phenomenon reminded me of working with Generative AI tools, like ChatGPT. Often, when you ask these models to reason through a question or explain a response, you encounter the same experience: A seemingly confident but not always accurate explanation.

Just as the split-brain patient’s left hemisphere constructed a story to explain behavior it didn’t understand, GenAI often generates plausible-sounding responses without truly “knowing” why. So how much of what we consider “reasoning” is simply a storytelling mechanism?

Could studying the brain’s interpretive tendencies—like those seen in split-brain research—help us better understand and improve how GenAI models operate? Now I am in no way an expert but I wonder if cognitive neuroscience and GenAI inform each other in new, exciting ways?