The McCulloch-Pitts Model

Today, the origin of artificial neural networks. The University of Houston presents this series about the machines that make our civilization run, and the people whose ingenuity created them. 

______________________

In 1935, the famous philosopher Bertrand Russell received a letter from the US pointing out errors in his monumental work, the Principia Mathematica. This three-volume tome attempted to lay the foundations for all of mathematics – for example, on page 379 of volume 1, you'll find a proof from which it follows that 1 + 1 = 2. Russell was impressed by the letter and invited its author to Cambridge University. There was just one problem: the correspondent, Walter Pitts, was twelve years old.

Walter Pitts was the son of a Detroit boilermaker. He fled home at when he was fifteen. He fled to the University of Chicago, where Bertrand Russel was visiting. After Russel returned to England, the teenaged Pitts stayed.  Like a real-life Will Hunting, he impressed the faculty at the University of Chicago with his skills. There he also met Warren McCulloch, a neurophysiologist fascinated with the brain, and determined to find out how it works. McCulloch believed the neural circuits in the brain follow algorithmic rules, and that the whole organ works like a computer. But McCulloch lacked the mathematical skills to translate his ideas into a coherent theory. Walter Pitts was exactly who he needed.

McCulloch and Pitts in 1949.  Image source https://www.semanticscholar.org/paper/On-the-legacy-of-W.S.-McCulloch-Moreno-D%C3%ADaz-Moreno-D%C3%ADaz/8056242a82ecc5e0064d4ff187fb07c5853fe8a6

After many late nights fueled by whiskey and their determination to solve the brain, McCulloch and Pitts published their groundbreaking paper in 1943. They proposed that groups of neurons - the 86 billion cells that make up much of your brain - could act like logical circuits. Neurons fire action potentials, all-or-nothing impulses generated when they receive a sufficiently strong input. Therefore McCulloch and Pitts assume that each neuron is in one of two states: firing or not firing, ON or OFF, just like a digital computer's ones and zeros. A neuron turns ON if it receives a strong input. It remains OFF its input is weak. By carefully wiring the output of one neuron to that of another, they showed that you can construct different logical computations. However, for complicated operations the required networks could become very elaborate.

McCulloch and Pitts thus made a bold proposal: Human thought can be reduced to logical operations implemented by neural switches. The famous mathematician John von Neumann built on their ideas to design self-replicating machines - machines that could hypothetically copy themselves, mine the Moon, or colonize the galaxy. The excitement culminated in the 1956 Dartmouth Summer Research Project: an ambitious proposal to study the idea that all aspects of intelligence can be simulated on a computer. Seventy years later, we're still working on it.

We now know that McCulloch and Pitts gave us a spectacularly wrong answer to how the brain works. But their answers spurred decades of research that inspired our current AI models. Thus, sometimes even wrong, but provocative answers can lead to surprising destinations.

This is Krešo Josić at the University of Houston, where we're interested in the way inventive minds work.

(Theme music)

Here is a more detailed, but not too technical account. 

Walter Pitts led a tragic life. He worked with Norbert Wiener at MIT, but never received a PhD. You can read more about him here and here.  Warren McCulloch was quite a character himself. Here is an interview with him in his later years, where, for some reason, he is shirtless.

This episode first aired on November 12, 2025.