Modelling Cognition – the ACT models
The ACT models are computer programs designed to model the human mind (they are cognitive architectures). They were primarily written and devised by John Anderson.
It evolved from the Human association model (HAM) to the ACT model, then the ACT* and finally its present incarnation as the ACT-R, with the R standing for 'rational'
the ACT architecture can be used to model many things, from language and perception to learning and memory.
JR Anderson's ACT*
In Architecture for Cognition (1983) Anderson outlined his idea that cognitive psychology needed an architecture (see above). he proposed a Meta-theoretical Framework, which was essentially a structure that the ideal architecture should follow. it should have many levels, from general principles of cognition at the top all the way down to specific effects at the bottom. it is detailed below.
Level 1: Framework
at the highest level is a general pool of constructs for understanding a domain. Domains are broad areas of cognitive psychology such as visual perception, or language, or memory. at this level the architecture is not specific enough to make any predictions.
As it makes no predictions, it is not possible to falsify a framework. The reason we have a framework is to generate theories to test; although the framework cant be tested the theories derived from it can. a good framework is one that is fruitful (see developing a theory in cognitive psychology).
Level 2: Theory
From a framework we get theories. A theory is a precise deductive system. It is more general than a model.
Level 3: Model
A model is an application of the theory to a specific problem.
E.g. Anderson applied ACT* to geometry, language, letter recognition, subtraction and computer programming.
Level 1: The ACT* Framework
see Learning for more on the ACT* levels & lecture 10 information (near the bottom of the page)
Claim 1: Higher level cognition involves a unitary system
A single set of principles (production system architecture) underlies all of cognition and there are no principled differences or separations of faculties (1983, p30)
i.e. In complete opposition the faculty approach (e.g. chomsky, 1980; Fodor, 1982) which specifies different faculties for the different aspects of cognition (e.g. language, maths, reasoning, problem solving).
Claim 2: Production Systems (PS) provide an adequate formalism for the underlying principles guiding human cognition
(i) Powerful enough - it can be shown that they are computatioally universal
i.e. you can model anything using productions
(ii) Not TOO powerful - (a more sublte point)
Unlike some powerful representations, such as schema theory, which are essentially unconstrained and thus almost untestable, the PS architecture is highly constrained and so it is possible to derive models which make predicitons that may be tested empirically.
(The ACT* Architecture)
Claim 3: The Architecture of the Human Cognitive System
The Architecture of the Human Cognitive Systemis a Production System architecture which has 3 memory systems - Working Memory (WM), Production Memory and Declarative Memory.
- Everything goes through WM, but it is not a specific store - Declarative Memory is for facts, but it is not in a production system format - Procedural Memory is in terms of production rules, created using the proceduralisation mechanism in his theory of learning.
Level 2: The ACT* Framework
1) Knowledge structures - declarative memory -Declarative knowledge is built up of cognitive units of which there are 3 kinds: propositions which represent meaning temporal strings representing order spatial images which represent spatial configuration
cognitive units are combined in a network which forms a tangled hierarchy ie. a mixed hierarchy of all data types which is not actually a proper hierarchy.
A MAJOR LIMITATION: ACT* does not have a way of building up declarative memory.
2) Knowledge structures - Working Memory (WM) - A variery of sources for example currently active productions, information deposited by production activity, goals and information deposited by sensory encoding. Goals have a special status forming a goal stack and are protected from decay. WM actually has a capacity of 20 items rather than 7+2.
3) Knowledge structures - Production Memory - Has a general form IF condition(s) matched in WM ...... THEN execute action(s)