Key Takeaways
- AI’s origins lie in cognitive science, not just computer science; early researchers sought to model the human mind.
- The first neural‑network model, the Perceptron (1958), was built to learn from examples, mirroring how brain connections strengthen with use.
- Core principles—learning from data, brain‑inspired architecture, and probabilistic generalization—explain why modern AI behaves like human memory rather than a deterministic database.
- AI “hallucinations” and variable answers stem from its reconstructive, probabilistic nature, not from errors in a lookup table.
- Understanding this interdisciplinary heritage helps users, educators, and researchers calibrate trust, improve verification practices, and advance both AI and cognitive science.
The Misconception of AI as a Database
When people imagine how ChatGPT or other large language models produce answers, they often picture a computer pulling a file from a hard drive or executing a fixed set of rules. This view treats AI as a deterministic database that simply retrieves stored facts. As the article notes, “Picturing AI responses as akin to your computer pulling up a file is a sensible line of thinking. But it’s incorrect.” Such a mental model leads to confusion when the model outputs falsehoods or gives different replies to the same query, because a true lookup system would never behave that way.
Origins at Dartmouth and the Rule‑Based Vision
The term “artificial intelligence” was first introduced in 1956 at a Dartmouth College workshop, where participants believed that intelligence could be engineered by encoding the right set of logical rules for a machine to follow. The article quotes this early optimism: “At the time, people thought that a machine could be made intelligent if you could write down the right set of rules for it to follow.” This rule‑based perspective dominated early AI research but overlooked the alternative route taken by psychologists who wanted to understand how the mind itself works.
Frank Rosenblatt’s Perceptron and Brain‑Inspired Learning
Psychologist Frank Rosenblatt diverged from the rule‑based crowd in 1958 by building the Perceptron, a machine that learned from examples rather than obeying pre‑written instructions. His insight was to create an artificial neural network loosely modeled on the brain’s architecture. The piece describes this advance: “Rosenblatt hadn’t been at the Dartmouth workshop, but he was thinking about how to make a machine intelligent in a fundamentally different, more human way. His key insight was to build an artificial neural network: a computer algorithm modeled loosely on the architecture of the brain.” The Perceptron’s many connections were likened to “a simple version of the neural connections in a brain,” establishing the idea that intelligence could emerge from distributed, adaptive units.
Hebb’s Rule and the Foundations of Neural Plasticity
The Perceptron built on psychologist Donald Hebb’s pioneering work on how synapses change with activity. Hebb proposed that “when an axon of cell A is near enough to excite a 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, as one of the cells firing B, is increased.” This principle—often summed up as “cells that fire together, wire together”—provided the biological basis for learning in artificial networks. The article highlights that “This approach built on the work of psychologists like Donald Hebb who were studying brains and behavior. Specifically, Hebb explored how neural connections strengthen when used, a fundamental insight that later seeded machine learning.”
The 1980s Breakthrough: Multilayer Networks and Deep Learning
In the 1980s, a coalition of cognitive and computer scientists—David Rumelhart, Geoffrey Hinton, and Ronald Williams—figured out how to train neural networks with multiple hidden layers, enabling them to capture more complex patterns. This development laid the groundwork for what we now call deep learning. The text notes: “Cognitive and computer scientists together … figured out how to train artificial neural networks with multiple layers, an idea that later became known as ‘deep learning.’ These multilayered neural networks were capable of more complex tasks and generalizing what they ‘learned’ to apply to new examples.” The ability to generalize, rather than merely memorize, is a direct inheritance from the brain’s capacity to extract abstractions from experience.
Scaling Up: Hardware, Transformers, and Modern AI
While the algorithmic ideas originated in cognitive science, the explosive performance of today’s AI relied on parallel advances in computing hardware and architecture. Graphics processing units (GPUs) provided the massive parallelism needed to train large networks, and the transformer architecture introduced mechanisms for handling long-range dependencies in language. The article explains: “In the following decades, computer scientists and engineers enabled computing capabilities to be scaled up, developing graphics chips, the transformer architecture and more. Without these systems, modern AI would still struggle to recognize letters and words, rather than be able to hold realistic conversations in natural language.” Thus, modern AI is the product of a symbiosis between brain‑inspired algorithms and engineered scalability.
Why AI Behaves Like Human Memory: Reconstruction and Hallucinations
Because AI learns from examples and generalizes probabilistically, its internal processes resemble the reconstructive nature of human memory rather than a static lookup table. Psychologist Elizabeth Loftus’s research showed that human memory often fills gaps with plausible details, sometimes creating false memories. The article draws this parallel: “Human memory doesn’t ‘look up’ answers. It is reconstructive and imperfect, filling in gaps with plausible details. Thanks to pioneering work by psychologist Elizabeth Loftus, researchers know that false memories are unfortunately common and can be implanted with relative ease.” Consequently, AI can “hallucinate” statements that sound confident but are not grounded in any training datum—mirroring how people confabulate when recalling events.
Probabilistic Outputs and Context‑Sensitivity
Unlike a calculator that returns the same result for identical inputs, AI’s responses vary with subtle changes in phrasing or context, reflecting its probabilistic nature. The article observes: “Thinking of AI as like a primitive brain also helps explain additional puzzling behaviors, such as getting different answers to the same question. Ask your young child what they want for dinner twice and you may get different requests. Framing matters when interacting with an AI system too, as these systems can be nudged into giving you answers you prefer. They operate in a probabilistic fashion, much like human memory, not in a deterministic fashion, like a calculator.” This variability is not a bug but a feature of systems designed to generalize across novel situations, much like humans adapt their answers based on mood, recent experiences, or subtle cues.
Implications for Trust, Education, and Research
Treating AI as a factual database leads users to overtrust its outputs, skipping verification steps they would apply to a human source. The piece warns: “Users overtrust the wrong things. With a significant proportion of ChatGPT users indicating that they use ChatGPT like a search engine, it seems that people are treating outputs as retrieved fact rather than fluent, confident guesses. This misunderstanding can cause users to skip the verification step they’d likely apply to a person and potentially make decisions, including financial ones, based on confabulation.” In academic settings, faculty are split between banning AI tools and integrating them as collaborators, creating confusion. Moreover, AI’s opacity has prompted a resurgence of interpretability research that mirrors behaviorist psychology’s focus on observable inputs and outputs, while also borrowing neuroscientific techniques to peer inside models.
The Two‑Way Street: Cognitive Science Benefits from AI
Just as AI draws on cognitive science’s theories of learning and memory, cognitive scientists now use AI models as testbeds for hypotheses about the mind. By probing how networks represent language, vision, or decision‑making, researchers can refine theories of human cognition. The article concludes: “The exchange runs both ways. Just as computer science draws on cognitive science’s questions and methods, cognitive science draws on the accessibility of AI models themselves, using them to test and refine theories of the mind. Computer science and cognitive science split off from one project decades ago: understanding how intelligence works. The more these fields keep trading questions and tools, the better placed humanity will be.” This reciprocal relationship underscores that appreciating AI’s cognitive‑science roots is essential not only for building better machines but also for deepening our understanding of human intelligence itself.
https://theconversation.com/ai-behaves-more-like-a-brain-than-a-database-cognitive-sciences-role-in-its-origin-story-helps-explain-why-287838