When you ask ChatGPT a question, it's tempting to imagine a vast digital library being searched, with the system retrieving a pre-written answer. That mental model, rooted in traditional computing, is understandable but fundamentally misleading. The reality is that modern AI, particularly large language models, operates more like a biological brain than a hard drive.
This distinction isn't just academic. It has practical implications for how we trust and use these tools, especially as they become embedded in everything from customer service to medical diagnostics across the Indo-Pacific region.
The cognitive science lineage
The term "artificial intelligence" was coined at a 1956 Dartmouth workshop, where the prevailing belief was that intelligence could be encoded as a set of explicit rules. But psychologist Frank Rosenblatt, who wasn't at that workshop, took a different path. In 1958, he built the Perceptron, a machine that learned from examples rather than following pre-programmed instructions. His key insight was to model the system on the brain's architecture, creating an artificial neural network.
This approach built on the work of psychologist Donald Hebb, who in the 1940s showed that neural connections strengthen with use—a principle that became the foundation of machine learning. In the 1980s, cognitive and computer scientists like David Rumelhart, Geoffrey Hinton, and Ronald Williams developed methods to train multi-layered neural networks, giving rise to what we now call deep learning. These networks could generalize from examples, a capability that rule-based systems lacked.
While engineers later scaled up these ideas with graphics chips and the transformer architecture, the fundamental breakthroughs came from studying the mind. AI was designed to learn, not to retrieve.
Why AI 'hallucinates'
This origin story explains a puzzling behavior: AI's tendency to confidently state falsehoods, often called "hallucinations." If you think of AI as a database, this seems like a glitch—how can a system look up an answer that doesn't exist? But human memory doesn't work that way. It's reconstructive, filling in gaps with plausible details. Psychologist Elizabeth Loftus's research on false memories shows how easily our brains can implant fabricated recollections.
AI, like human memory, is probabilistic. It generates responses based on patterns in its training data, not by retrieving fixed facts. This allows it to answer novel questions but also makes it prone to confabulation. The same question asked twice might yield different answers, just as a child might change their dinner preference. This flexibility is a feature, not a bug, but it requires users to adjust their expectations.
Implications for users
Many people use ChatGPT as a search engine, expecting factual accuracy. But treating AI outputs as retrieved facts is a misunderstanding. These systems are more like a fluent, confident guesser. This is particularly relevant as AI tools are adopted in sectors like education and healthcare across Asia, where over-reliance could have serious consequences.
Understanding AI's cognitive roots can help users build better mental models. Instead of expecting deterministic answers, we should approach AI as a tool that can assist with brainstorming, drafting, and pattern recognition, but not as a reliable source of truth. As rethinking AI investment becomes a priority for governments and businesses, setting realistic expectations is crucial.
In the broader context of South Korea's evolving tech strategy and China's military AI ambitions, the way we perceive AI's capabilities shapes policy and public trust. A more accurate understanding—that AI is a brain-like system, not a database—can lead to more responsible use and development.
Ultimately, AI's unpredictability is a trade-off for its adaptability. By recognizing its cognitive science heritage, we can better navigate its strengths and limitations, ensuring that we use it wisely in an increasingly interconnected world.
