What Happened
In 1935, a twelve-year-old Walter Pitts hid from bullies in a Detroit public library. He read Principia Mathematica, a two-thousand-page treatise on logic, cover to cover in three days.
He found errors. He wrote to Bertrand Russell. Russell replied: come to Cambridge.
Walter was twelve. He couldn’t. Three years later, he ran away to Chicago anyway. He lived homeless. He listened to lectures and he corrected books. He met Warren McCulloch and moved in with him. They drank whiskey and talked until 4 a.m.
They published a paper in 1943. A Logical Calculus of Ideas Immanent in Nervous Activity.
This paper is the grandfather of all neural networks. It’s the reason you can read this sentence on a screen. Alan Turing’s machine needed this. John von Neumann cited it exclusively. It defined how logic, memory, and thinking could be reduced to switching circuits.
Then Walter burned it.
Not the paper. The follow-up. The work on three-dimensional neural networks. His own notes. His own future. He set fire to years of research in a boarding house in Cambridge, Massachusetts.
When he was forty-five he died of cirrhosis due to heavy drinking. McCulloch died four months after that.
What We Should Have Learned
You might call Pitts a tragic genius or cautionary tale. I don’t buy it. I think he was the only person in the room who understood the assignment.
Pitts and McCulloch proved that machines could think. They also proved that human brains don’t work like machines. The two findings pull in opposite directions, and that tension has never been resolved. It runs through AI research from 1943 to now.
In 1949, Pitts worked with biologist Umberto Materson. They were trying to map how the brain processes visual information. They expected light to hit the retina, travel to the brain, and get converted into data.
They were wrong. The eye processes information before it ever reaches the brain. The hardware does the work before the software gets involved. This shattered their model. It proved that scaling up logic gates doesn’t create life, it just creates faster switches.
Fast forward to today and we’re building larger models, pushing more parameters, throwing more compute at the problem. We’re doing exactly what Pitts did in 1949. We’re assuming that if we scale the math, the mind will follow.
It won’t.
Pitts knew this. He saw that you can build a Turing machine, but you can’t use it to explain how a human actually thinks. The randomness of the brain doesn’t resolve into order through math alone. The hardware matters. The biology matters. The messiness matters.
When Norbert Wiener cut off his funding, likely engineered by his wife who hated McCulloch, Pitts lost his patron. He lost his direction. He started drinking. He stopped going outside. MIT offered him a PhD. He refused to sign. He burned the dissertation.
I think that was clarity, not a breakdown. I think he saw they were building a mirror instead of a mind.
What We Didn’t Learn
We are stuck in 1949. We have more compute and better tools, but the same misunderstanding. We think intelligence is a function of scale. We ignore the biological reality that Pitts spent his life trying to map.
Scaling alone won’t get us to artificial general intelligence. It never will and no amount of data center monstrosities will change that. Pitts proved that and we just didn’t listen.
The problem is that we’re still walking the same path he walked away from. We’re building faster switches and ignoring the wiring.