AI Behaves More Like a Brain Than a Database (Hout & Robbins, The Conversation, 2026-10-07)
Michael Hout (New Mexico State) and Arryn Robbins (University of Richmond), both psychologists, argue that AI's defining traits come from its roots in cognitive science, not computer science alone.
The history they tell
At the 1956 Dartmouth workshop, the founders of "artificial intelligence" bet on writing the right rules. Frank Rosenblatt, a psychologist who was not at Dartmouth, built the Perceptron (1958) instead: a machine that learns from examples, loosely modeled on brain architecture. Donald Hebb's work on how neural connections strengthen with use fed into it. In the 1980s, Rumelhart, Hinton and Williams worked out how to train multilayer networks, the root of today's deep learning, which could generalize to new examples rather than only answer what it was told. Moore's-law scaling, GPUs and the transformer architecture made the engineering practical decades later, but the authors' point is that the governing idea, learning from examples instead of following rules, came out of people trying to model minds.
Why that explains "brain, not database"
A lookup table gives the same answer to the same question every time, and only to questions it holds. A system built like memory instead reconstructs an answer probabilistically, which lets it generalize to new cases at the cost of confabulating when it has nothing to retrieve. The authors tie this directly to Elizabeth Loftus's work on false memory: human memory fills gaps with plausible detail, and false memories are easy to implant experimentally. They read an AI hallucination as the same failure mode in a different substrate, not as a bug alien to how minds already work. The same frame explains two more AI behaviors this spoke already tracks adjacent territory for: answers that shift between identical questions (a child asked twice what it wants for dinner), and answers that move with how a question is framed.
What they draw from it
Three consequences, stated plainly: people treat a confident guess as a retrieved fact and skip verification, sometimes for decisions with money attached; schools are split between banning AI and teaching with it; and AI interpretability research, looking inside the model rather than only at its outputs, is a return to questions cognitive science has worked on for decades, after a behaviorist input-output era in AI testing. They note explicitly that AI doesn't necessarily think like people, citing computer vision that catches subtleties humans miss while misreading a defaced road sign.
Where it sits in this spoke, and what it doesn't close
This is an op-ed by domain specialists, not a primary history or a study: Hebb, Rosenblatt and the 1980s connectionists are named and dated but not quoted from their own papers, and Loftus's false-memory finding is cited by name only. Open edge 9 ("memory... Loftus on false memory, needs a T1 review or the original eyewitness studies") stays open; this source supplies the argument that connects memory research to AI behavior, not the primary memory literature itself. What's new here is the connective claim — AI's architecture, not just its vocabulary, descends from cognitive-science theories of mind — which the spoke did not have a citable source for before.
Tier
T3 — a newspaper-style explainer by credentialed academics (an associate dean and an assistant professor of psychology), for a general audience, with no citations to the primary papers it summarizes (Rosenblatt 1958, Hebb 1949, Rumelhart/Hinton/Williams 1986, Loftus's eyewitness studies). Historically accurate on the dates and names checkable independently; the explanatory claims (AI behavior as confabulation) are the authors' own synthesis, not a finding either author ran.