Human Associative Memory (HAM) Model Published by John R. Anderson and Gordon H. Bower
In 1973, John R. Anderson and Gordon H. Bower, both at Stanford University, published 'Human Associative Memory', introducing the HAM model, a propositional network architecture representing semantic memory as binary trees, providing a computationally explicit theory of human memory that influenced subsequent cognitive architectures.
Background
By the early 1970s, artificial intelligence researchers and cognitive psychologists were both trying to work out how knowledge is stored and retrieved. The two fields had different instincts about this. AI researchers tended to build systems that worked, without worrying much about whether they worked the way human minds do. Psychologists had decades of experimental data on human memory but no precise formal language to describe what was happening inside people’s heads.
The memory models that existed were mostly verbal. A researcher might describe how people link concepts together, or how context affects recall, but the description would stop well short of anything a computer could execute. That gap mattered, because an informal theory is hard to test: you can always adjust your explanation after the fact to fit new data.
Several research groups in the late 1960s and early 1970s were beginning to build semantic networks, graph-like structures where concepts sit at the nodes and labelled connections carry the meaning between them. Quillian’s Teachable Language Comprehender, published in 1968, was one early example. These networks could represent knowledge, but they often lacked a clear account of how the structure mapped to what people actually do when they remember something.
What happened
In 1973, John R. Anderson and Gordon H. Bower, both working at Stanford University, published Human Associative Memory through V. H. Winston & Sons. The book introduced HAM, a model that attempted to represent human semantic memory with enough precision to be written as a running computer program.
The architecture used propositional binary trees. Each proposition, a meaningful statement about the world, was broken into a fixed branching structure of nodes connected by labelled links. Every node carried a unique identifier. The links were not generic; each one had a specific functional label that determined what kind of relationship it expressed, such as the distinction between a subject and its context, or a predicate and its object. Word indexing across the network was handled in LISP, which let the system search through nodes by tracing the labelled connections. The formalism was strict enough that the model’s predictions about memory behaviour could be compared against actual experimental results and, in principle, shown to be wrong.
That last point was the real ambition. Anderson and Bower were not just building a tool; they were trying to write psychology down in a form that could be falsified. The book ran to several hundred pages of theory, worked examples and experimental evidence. A review published in the American Journal of Psychology in 1974 treated it as a serious contribution to both cognitive psychology and the formal study of memory. The model’s direct line of descent runs through Anderson’s later ACT architecture and then ACT-R, which became one of the most widely used frameworks in cognitive modelling over the following decades.
Why it mattered
HAM was among the earliest attempts to formalise human semantic memory as a computable structure, representing knowledge as labelled binary trees of propositions implemented in LISP. Its influence extended directly into Anderson's later ACT and ACT-R architectures, which became foundational frameworks for cognitive modelling and informed the design of knowledge-representation systems in AI. The model demonstrated that psychological theories of memory could be expressed with enough precision to be simulated and falsified computationally.
People
John R. Anderson, Gordon H Bower
Organisations
Stanford University, V H Winston and Sons
Sources
- Human Associative Memory.V. H. Winston & Sons (reprinted by Psychology Press/Taylor & Francis).Primary source
- Review of Human Associative Memory.American Journal of Psychology (University of Illinois Press).Secondary
- The Place of Cognitive Architectures in a Rational Analysis: The Example of ACT-R.Cognitive Science (Elsevier), contextualises HAM's role in the lineage leading to ACT and ACT-R.Secondary
Cite this page
AI Achievements. (1973). Human Associative Memory (HAM) Model Published by John R. Anderson and Gordon H. Bower. Retrieved 2026-08-22, from https://achievements.ai/milestone/ham-model-by-anderson-and-bower
@misc{achievements_ham_model_by_anderson_and_bower,
title = {Human Associative Memory (HAM) Model Published by John R. Anderson and Gordon H. Bower},
author = {{AI Achievements}},
year = {1973},
url = {https://achievements.ai/milestone/ham-model-by-anderson-and-bower}
}