Survival of the Stickiest

Yesterday, the Guardian published a sort of obituary for the »lizard brain«.¹ The idea, popularised by neuroscientist Paul MacLean in 1959, was that the human brain consists of three evolutionary layers: a primitive reptilian brain, a mammalian emotional brain, and a uniquely human rational brain. Neuroscience begs to differ. Yet »my lizard brain took over« remains a perfectly intelligible thing to say.

Maybe it won’t be enough for a new explanation to be more accurate than the old one. Maybe it needs to also become more transmissible.

Why do some ideas survive after the evidence has killed them?

Science has an unusual relationship with being wrong. Philosopher of science Karl Popper argued that a scientific theory should expose itself to falsification. It must leave open the possibility that evidence will show it to be wrong.² Being wrong and overturned isn’t a failure of science. It’s how science advances. It needs ideas that can die.

Another philosopher of science, Thomas Kuhn – following theoretical physicist Max Planck – observed that scientific revolutions are often less about persuading opponents than waiting for them to die and a new generation to grow up with a different picture of the world.³ Inside science, there’s a mechanism for retirement. Outside science, there isn’t.

Scientific ideas don’t necessarily die, or disappear, when scientists disprove them. The lizard brain escaped neuroscience to become a metaphor: a compact story about the animal underneath the rational self. Its scientific content could be discarded without destroying its usefulness. »Your lizard brain took over« is easier to understand than a more correct explanation involving elements such as competing predictions, neural networks, energy budgets, and evolutionary development.

A third philosopher of science, Imre Lakatos, described scientific research programmes as having a »hard core« surrounded by a »protective belt« of auxiliary hypotheses.⁴ The lizard brain has undergone almost the reverse transformation: while the hard scientific core collapsed, the useful belt survived. The evolutionary biologist Richard Dawkins offers another way of looking at it. In his book The Selfish Gene, he coined the word »meme« for a unit of cultural transmission, analogous to the gene.⁵ Ideas compete for reproduction, and their ability to spread and stick in people’s minds doesn’t necessarily have much to do with whether they’re actually true. The lizard brain is a remarkably successful meme.

In other words, the lizard brain lost the scientific competition but won the cultural one. And that distinction could become a lot more important now that we have built machines whose particular talent is cultural memory. Large language models (LLMs) are trained on enormous bodies of text containing scientifically sound ideas, but also obsolete theories and the explanations that replaced them. What LLMs don’t have is science’s slow institutional mechanism for retirement. A scientific community can explicitly or silently decide that a theory no longer survives the evidence. An LLM encounters millions of traces of an idea that has survived culturally, and can reproduce the most familiar explanation when somebody asks for one.

This isn’t just a problem of AI »hallucinating« facts. Something stranger is happening here. The machine might reproduce an idea precisely because it has been culturally successful, even when it has ceased to be scientifically useful. And the old idea, the meme version of the theory, has an advantage. It has already been simplified, packaged, turned into metaphor, made easy to say, and repeated. The newer idea has to compete for attention with all of that accumulated memetic sediment.

This points to a fundamental and potentially necessary change in scientific theory building and communication. Maybe it won’t be enough for a new explanation to be more accurate than the old one. Maybe it needs to also become more transmissible. Stickier.

Predictive neuroscience may eventually find its own equivalent of »lizard brain« – a phrase or an image or a story simple enough to carry a much more complicated truth. If it does, it will spread for the same reason that the old metaphor did: not because it’s scientifically superior, but because it’s easier to think with.

We find ourselves at the beginning of an arms race between the next good metaphor and the last one. For the first time in history, something other than human memory is helping both sides reproduce – and it isn’t yet clear which one it’s helping faster.

References