The Neuron That Wanted to Be God

7,000 words to tell an 80-year-old story

I.

When scientists Warren McCullough and Walter Pitts invented the first mathematical model of the neuron in 1943, they couldn’t have imagined that eighty years later, humanity would be pouring hundreds of billions of dollars into their idea. But here we are, and they are not, so they cannot tell us just how patently absurd the current artificial intelligence frenzy has become.

If you take a look at the modern pipeline of how AI models are created — from the first line of code some unknown developer writes to the moment Sam Altman takes the stage and claims “With this new model, I did feel the AGI” — there are various points at which one can seriously argue that “ok, this should have been done some other way.” (Maybe it’s not something to be proud of that the next generation of AI models needs a quadrillion tokens of data — that is, 1,000,000,000,000,000 as many — to increment their performance half a percentage point in some unknown benchmark test.)

However, it’s genuinely surprising that you’d have to go all the way back to the inception of the field to find the first controversial decision that never received sufficient care or consideration — not even after more powerful computers and more money and more data tokens were devoted to the cause.

You could wonder why AI labs are trying to scale large language models (LLMs) to the size of the human brain (in the ballpark of a…