The meeting of a neurophysiologist and a teenage autodidact during World War II laid the foundation for one of the most influential theories in the history of Artificial Intelligence. The McCulloch–Pitts model showed that brain activity could in principle be described by formal logical structures, even though the actual functioning of neurons was later shown to be much more complex. Still, this theory provided the first conceptual foundation for modern neural networks, long before the world began to ask questions about them.

During World War II, sometime in the early 1940s, the paths of two unusual figures crossed in Chicago. One was Warren McCulloch, a neurophysiologist and philosopher who studied logic at Yale and earned a medical degree from Columbia University. He had long been fascinated by the question of how the human brain thinks. The other was Walter Pitts, a young autodidact often regarded as an outsider throughout his life. As a teenager, Pitts taught himself logic and several languages, including Greek and Latin. According to later accounts, at the age of twelve he wrote a letter to Bertrand Russell pointing out an error in one of the arguments in Principia Mathematica, though this story is difficult to verify. Later, he ran away from home and ended up in Chicago, where he lived in university lecture halls. Though not an enrolled student, he regularly attended logic and mathematics classes and wrote papers for scholars such as Rudolf Carnap, who received them with surprise and admiration.

McCulloch was looking for a collaborator with whom he could develop an idea he had been contemplating for years: could the functioning of the nervous system be described using formal logical structures? Pitts’s interests and knowledge proved to be a perfect fit, and they began working together almost naturally. The unusual pair found a common language in mathematics and logic. After sharing a dinner, McCulloch invited Pitts to his home, where they worked side by side for many months, often late into the night. They were convinced that neurons were not merely biological units, but elements that carried out logical operations—yes-no decisions, like a simple binary system.

Their collaboration culminated in 1943 with a landmark study titled “A Logical Calculus of the Ideas Immanent in Nervous Activity.” In this paper, they introduced the model now known as the McCulloch–Pitts neuron. In this framework, neurons function as binary units: they “fire,” that is, activate, whenever a certain threshold of input is reached. The model also takes into account the strength and direction of synaptic connections.

Surprisingly, even this simplified system could perform basic logical operations: it could simulate AND, OR, and NOT functions. The most striking claim of the study was that such networks could in principle be Turing complete, meaning they are theoretically capable of carrying out any computational task that a universal machine can perform.

Although the McCulloch–Pitts model fell far short of real neurons in biological detail (for example, it ignored the timing of synapses, learning processes, and the nonlinear responses of neurons) its theoretical significance proved to be invaluable. The model demonstrated that universal computation could be achieved by appropriately connecting simple binary units. This idea provided direct inspiration for the perceptron models of the 1950s, as well as for the later development of artificial neural networks.

While AI research in the mid-20th century also took other directions, such as symbolic AI and rule-based systems, from the 1980s onward, neural networks returned to the forefront thanks to deep learning and backpropagation algorithms. Modern deep learning architectures, which today can recognize images, model language, or even make complex decisions, are fundamentally based on the same principle: the collective intelligence of simple, interconnected units.

The McCulloch–Pitts model was not only a groundbreaking theoretical innovation, but also the root of today’s AI. It is a root that continues to nourish new breakthroughs to this day.

The result was, in many respects, ahead of its time. Although the study did not immediately attract widespread public attention, several prominent figures in the scientific community, including John von Neumann and Norbert Wiener, recognized its significance. The paper laid the foundations for the first formal neural network model, offering a mathematically robust structure for describing brain function.

While the model was highly idealized, ignoring the dynamics of biological neurons, the process of learning, and synaptic plasticity, it nevertheless made one thing clear: by connecting enough simple, binary-operating units, any computational operation could be performed. Building on this foundational concept, later researchers developed the perceptron models of the 1950s and, from the 1980s onward, artificial neural networks driven by backpropagation algorithms. Today’s deep learning architectures, which interpret images, generate text, and make complex decisions, still fundamentally follow this principle. The McCulloch–Pitts model is not merely a historical curiosity, but a living legacy. It is the root from which the entire tree of modern artificial intelligence has grown.

In the meantime, McCulloch and Pitts became increasingly active figures in the scientific community. Both participated in the Macy Conferences, held between 1946 and 1953, which aimed to explore the connections between feedback, cybernetics, and cognitive processes from an interdisciplinary perspective. McCulloch was particularly interested in the neural basis of consciousness, memory, and psychosis, while Pitts consistently focused on problems of logic and network theory. He also made a significant contribution to the study “What the Frog’s Eye Tells the Frog’s Brain,” which, together with co-authors Jerome Lettvin and Humberto Maturana, introduced a novel approach to understanding the neural mechanisms of sensory processing.

Over time, Pitts gradually withdrew from the public eye. The deterioration of his relationship with Norbert Wiener, along with personal losses, led him to increasing isolation. Although he continued working at MIT, he published less and eventually destroyed several of his own manuscripts. He died in 1969, at the age of just forty-six, receiving far less recognition than he deserved. Throughout his life, McCulloch spoke of him with deep respect and regarded him as his intellectual son.

Beyond their personal stories, the legacy of McCulloch and Pitts has left a broader lesson about how science often advances. Without their meeting, friendship, and shared vision, we might not view the modeling of human thought the way we do today. The story of the McCulloch–Pitts model reminds us that scientific breakthroughs are often born not of deliberate planning, but of unexpected encounters, openness, and curiosity. The unusual and fertile merging of logic and neuroscience took place at a moment when the world was not yet ready for the neural network revolution. But the seed had already been planted quietly, in the shadow of a university lecture hall.

István ÜVEGES, PhD is a Computational Linguist researcher and developer at MONTANA Knowledge Management Ltd. and a researcher at the HUN-REN Centre for Social Sciences. His main interests include the social impacts of Artificial Intelligence (Machine Learning), the nature of Legal Language (legalese), the Plain Language Movement, and sentiment- and emotion analysis.