cognitive architecture
In plain wordsA blueprint for a complete thinking system — how memory, attention, learning and decision-making fit together — used both to build general AI agents and to model how the human mind works.
A specification of the fixed structures and processes of a general intelligent system — its memories, representations, learning mechanisms and the cycle by which perception, deliberation and action interact — intended to support a wide range of tasks once task knowledge is added. Cognitive architectures serve two goals, often together: modeling human cognition so that simulations match experimental data such as reaction times and errors, and engineering general-purpose agents. Prominent examples are Soar, built around production rulesproduction system, n.: A program built from many 'if this, then do that' rules plus a memory of current facts; it repeatedly finds a rule whose 'if' part matches and carries out its 'then' part., a goal hierarchy, subgoals created at impasses and chunking (learning new rules from problem solving); ACT-R, which combines symbolic productions with subsymbolic activation equations and maps its modules to brain regions; and hybrids such as CLARION and LIDA. Since 2023, frameworks such as CoALA have used cognitive-architecture concepts — working, episodic and procedural memory, decision cycles — to describe and design language-modellarge language model, n.: A computer program trained on enormous amounts of text to predict the words that come next. At vast scale, that simple skill lets it chat, answer questions, translate, summarize and write code. agentsAI agent, n.: An AI system that does not just answer but acts: given a goal, it decides on steps, uses tools such as a browser or code editor, checks the results and keeps going until the job is done..
Origin
Soar was created by John Laird, Allen Newell and Paul Rosenbloom from 1983, and John Anderson's ACT theories, from ACT* (1983) to ACT-R (1993), developed in parallel at Carnegie Mellon. Newell's 1987 William James Lectures, published as Unified Theories of Cognition (1990), made the case for such architectures.
Usage note
Unlike a neural-network architecture, which specifies a learnable function, a cognitive architecture specifies the organization of a whole mind or agent.
Related
- production systemproduction system, n.: A program built from many 'if this, then do that' rules plus a memory of current facts; it repeatedly finds a rule whose 'if' part matches and carries out its 'then' part.
- physical symbol system hypothesisphysical symbol system hypothesis, n.: The 1976 claim by two AI pioneers that a machine which stores and rearranges symbols according to rules has everything needed for general intelligence, and that anything generally intelligent must be such a machine.
- Allen NewellAllen Newell, prop. n.: An American computer scientist who, with Herbert Simon, wrote early AI programs that solved problems the way people do, and who sought a single theory of how the mind works.
- AI agentAI agent, n.: An AI system that does not just answer but acts: given a goal, it decides on steps, uses tools such as a browser or code editor, checks the results and keeps going until the job is done.
- blackboard systemblackboard system, n.: A design for AI programs that lets many independent specialist modules collaborate only through a shared workspace: each watches it, posts a piece of the answer when it sees a chance, and builds on what others posted.
- symbolic AIsymbolic AI, n.: An approach to AI that writes knowledge down as human-readable symbols and rules, such as 'every bird has wings', and reasons by manipulating them step by step instead of learning patterns from data.
Sources
- Laird, J. E., Newell, A. & Rosenbloom, P. S. (1987). SOAR: An Architecture for General Intelligence. Artificial Intelligence 33(1), 1–64.
- Newell, A. (1990). Unified Theories of Cognition. Harvard University Press.
- Sumers, T. R., Yao, S., Narasimhan, K. & Griffiths, T. L. (2023). Cognitive Architectures for Language Agents. arXiv preprint (published in TMLR, 2024).
Mentioned in
agent memoryagent memory, n.: The ways an AI assistant or agent keeps information — your preferences, past conversations, lessons from earlier tasks — and brings it back later, beyond what fits in a single conversation., Allen NewellAllen Newell, prop. n.: An American computer scientist who, with Herbert Simon, wrote early AI programs that solved problems the way people do, and who sought a single theory of how the mind works., architecturearchitecture, n.: The overall design of an AI model, meaning what kinds of parts it has and how they are connected, as opposed to the specific numbers it learns during training., chunkingchunking, n.: Cutting long documents into smaller pieces before storing them for AI search, so the system can find and hand the model just the relevant part instead of a whole manual., cognitive sciencecognitive science, n.: The scientific study of how minds work — how we reason, use language, remember and see — combining psychology, neuroscience, linguistics, philosophy, anthropology and artificial intelligence., global workspace theoryglobal workspace theory, n.: A leading scientific theory holding that we become conscious of something when the brain broadcasts it widely — as if putting it on a stage where every part of the mind can use it., intelligent agentintelligent agent, n.: In AI theory, a decision-maker, such as a robot, a self-driving car, a chess program or a helpful piece of software, that notices what is going on around it and responds in ways that serve its goals., physical symbol system hypothesisphysical symbol system hypothesis, n.: The 1976 claim by two AI pioneers that a machine which stores and rearranges symbols according to rules has everything needed for general intelligence, and that anything generally intelligent must be such a machine., The Society of MindThe Society of Mind, prop. n.: AI pioneer Marvin Minsky's 1986 book arguing that intelligence comes from huge numbers of tiny, simple processes working together, no one of which is smart by itself., working memoryworking memory, n.: The small amount of information held in mind right now while doing a task. For an AI model, that is roughly whatever currently sits in its context window., belief–desire–intention modelbelief–desire–intention model, n.: A way of designing software agents around three human-like ideas: what the agent believes about the world, what it wants, and which plans it has committed to carrying out., Ben GoertzelBen Goertzel, prop. n.: An American AI researcher who made 'artificial general intelligence' a widely used name and built a research community around the goal of human-like, broadly capable AI., blackboard systemblackboard system, n.: A design for AI programs that lets many independent specialist modules collaborate only through a shared workspace: each watches it, posts a piece of the answer when it sees a chance, and builds on what others posted., production systemproduction system, n.: A program built from many 'if this, then do that' rules plus a memory of current facts; it repeatedly finds a rule whose 'if' part matches and carries out its 'then' part., subsumption architecturesubsumption architecture, n.: A way to build robot brains from stacked, reflex-like behaviors — a bottom layer that keeps clear of obstacles, others that roam or explore — where higher layers can take over from lower ones, with no central map or plan.