The Summer of ’56: When AI and Cognitive Science Were Born
Let’s be clear about one thing: the first "artificial intelligence" wasn’t sparked in a tech startup or a Silicon Valley garage. It happened—like so many brilliant, chaotic ideas—in the summer of 1956 at Dartmouth College.
June 18 marked the opening of the Dartmouth Summer Research Project on Artificial Intelligence, a workshop that ran until August 17. That document—the official proposal—contains the most ambitious promise ever attached to a machine: "every aspect of learning or any other feature of intelligence can in principle be so precisely described that a machine can be made to simulate it."
John McCarthy coined the phrase "artificial intelligence" there, and names like Marvin Minsky, Nathaniel Rochester, Alan Newell, Herb Simon, George Miller, and Duncan Luce were there—not as specialists in one field, but as polymaths comfortable moving between psychology, computer science, and mathematics.
But here’s what history books often skip: the same summer also gave birth to something else—cognitive science.
In September, just weeks after Dartmouth wrapped up, another workshop convened at MIT. This one focused on information theory—coding, signal detection, language processing. Nathaniel Rochester presented on Hebbian cell assemblies; George Miller spoke about memory limits—yes, the "magical number seven" work that would later inform everything from UI design to AI tokenization. Alan Newell and Herb Simon shared their early problem-solving program.
George Miller himself called this second workshop "the moment of conception" of cognitive science. Not as a rival to AI, but as its intellectual twin—two sides of the same coin: one building intelligent machines, the other explaining intelligent minds.
The Birth of a Scientific Marriage
You don’t have to squint too hard to see the family resemblance between AI and psychology. Every term—intelligence, learning, neural networks, memory, training, recall, reinforcement—came from psychology first.
The Dartmouth workshop didn’t just name a field—it borrowed the language of human cognition and said, "What if machines could do this too?" The result wasn’t just technical—it was philosophical. If a machine could "recall" information, did it also forget? If it "learned," what was the psychological equivalent of its training process?
Donald Hebb’s 1949 book The Organization of Behavior—a psychology text, not a computer science manual—laid the groundwork with his famous postulate: "When an axon of cell A is near enough to excite cell B and repeatedly or persistently takes part in firing it, some growth process or metabolic change takes place in one or both cells such that A’s efficiency...is increased." This simple idea—neurons that fire together wire together—is the bedrock of both Hebbian learning in neuroscience and the weight adjustments in artificial neural networks.
Frank Rosenblatt, a psychologist-coder hybrid, built the Perceptron in 1958, the first neural network model. Later, David Rumelhart and Jay McClelland’s PDP (Parallel Distributed Processing) models formalized backpropagation—the algorithm that powers deep learning today. None of this would exist without psychologists asking: How do humans learn? and then building computational models to answer it.
In other words: modern LLMs—the entire foundation of generative AI—are a direct descendant of cognitive psychology research from the mid-20th century. Deep learning isn’t just like human learning; it’s literally a computational approximation of psychological processes we first documented in behavior labs and fMRI scanners.
Why Your LLM Hallucinates Like a Human
It’s tempting to see AI errors—hallucinations, factual inconsistencies—as bugs. But if you step back and consider the psychological roots of these systems? They look suspiciously like features.
After all, our own memory doesn’t just retrieve stored facts; it reconstructs them. Every recall is a mini-reimagination, prone to bias, suggestion, and confident error. The same is true for large language models: they don’t store facts in a database; they predict sequences based on statistical patterns. The result? They can generate highly coherent, utterly false narratives—just like humans do when they’re overconfident or under-informed.
This isn’t a failure of AI—it’s evidence that the architecture works as intended. We designed machines to simulate human intelligence, not abstract rationality. And since human intelligence is noisy, context-dependent, and sometimes wildly wrong in convincing ways...so too are our best models.
Even robotics follows this pattern. You don’t need legs to navigate a factory floor, and you certainly don’t need eyes—but when robots look human, people trust them more. That’s not engineering necessity; it’s social cognition, a psychological response to anthropomorphic cues. The Dartmouth proposal aimed for machines that simulate intelligence; what we got—unintentionally, perhaps—was machines that trigger our innate tendency to anthropomorphize.
Max Louwerse puts it bluntly in Understanding Artificial Minds through Human Minds (2025): "AI isn’t just mimicking intelligence. It’s mimicking how humans think they think."
The Cognitive Revolution: From Behaviorism to Neural Nets
Before 1956, psychology was mostly about what you could observe: stimuli and responses. B.F. Skinner’s behaviorism dominated the field—the mind was a black box, best ignored in favor of measurable actions.
The Dartmouth and MIT workshops didn’t just launch a new technology. They reignited psychology’s interest in internal processes: attention, working memory, mental models, and computation. The result? The cognitive revolution—a slow but profound shift from behavior to cognition, and from reaction to representation.
This wasn’t just theoretical. George Miller’s 7±2 work on short-term memory directly influenced early human-computer interaction and interface design. Allen Newell and Herb Simon’s General Problem Solver led to both cognitive architectures and early AI planning systems. Even something as mundane as error messages in software (“404 Not Found” vs. “A system error occurred”) draws on decades of human-computer interaction research rooted in cognitive psychology.
The irony? The tools we built to study the mind ended up changing how we think—and how machines simulate that thinking. Today’s LLMs, despite their scale and complexity, are trying to do what Miller tried in the 1950s: model how information flows through cognitive systems. They just do it on a laptop instead of in the brain.
The boundary between cognitive science and AI has blurred so completely that modern labs don’t even bother drawing lines. Institutions like MIT’s Center for Brains, Minds and Machines and Tilburg University’s Line of Cognitive AI explicitly treat cognition and intelligence as two sides of the same coin—something psychology helped establish in 1956.
A Birthday Worth Celebrating—Twice
So this July, when you raise a glass to celebrate AI’s 70th birthday, pause for a second and consider who else got invited.
Cognitive science, too, turns 70 this summer. And psychology—well, psychology got a serious upgrade thanks to the same workshop that gave us AI.
We’ve gone from machines with no memory, to machines that simulate it; from behaviorist black boxes, to neural networks that mirror our own learning pathways; from simple problem solvers, to language models that generate human-like prose—and sometimes human-like errors.
The original Dartmouth goal was bold: machines that simulate any aspect of intelligence. Today, LLMs achieve this not through symbolic logic or rigid rules, but through statistical approximation—mimicking how humans process language, often repeating our mistakes along the way.
And here’s the real gift: by studying how machines get things wrong, we’re learning more about how we do too. That’s the marriage McCarthy and Miller started in 1956—two fields that, for all their technical differences, still speak the same psychological language.
So go ahead: light a candle for AI. But also one for cognitive science, and another for psychology—the disciplines that didn’t just help build artificial intelligence; they became its first students.