Adaptive Control of Thought

from Wikipedia, the free encyclopedia

Adaptive Control of Thought (ACT), originally Active Control of Thought, is a cognitive psychological theory to explain the cognitive performance (including thinking , language and memory ) of humans. It was first introduced in 1983 by John R. Anderson and has been continuously developed since then. On the one hand, she wants to explain how people structure their knowledge and why they are capable of intelligent behavior . On the other hand, it wants to provide the cognitive sciences and artificial intelligence research with a model that enables computer simulations of human intellectual performance.

ACT is a so-called cognitive architecture . Cognitive architectures represent a kind of "tool kit" with which certain phenomena of human cognition can be modeled (e.g. reading a sentence including extracting meaning). Such cognitive architectures represent a contrast to so-called “modeling from scratch” (starting from scratch, without a model), as certain - experimentally based - requirements are already implemented . This can be, for example, latency times when recognizing a word or the like.

ACT belongs to the class of "production systems", so it is a model that is based on the creation, storage and execution of so-called productions . These productions are rules of the form “If condition A is fulfilled, then carry out action B” (so-called material implication ).

ACT-R

ACT comes in many versions. Usually the suffix after "ACT" denotes a special extension in a certain direction of research. The "R" stands for "Rational", ie " reasonable ". It points out that this conceptual model is based on normative cost-benefit analyzes , mostly calculated with Bayes' theorem .

Basic building blocks

Memory types

ACT comprises three memory modules . Pure factual knowledge, comparable to human declarative memory , is stored in the declarative memory . Its elementary unit are the Working Memory Elements , or WMEs for short, or Wimees . You only need one storage space (see below) because they form a meaningful unit. For example, TÜV would be a Wimee , VÜT usually not. In other models, these knowledge elements are called “chunks” .

ACT has its own long-term memory for storing production rules , the production memory (similar to, but not to be confused with, the human procedural memory ). Its smallest unit is the productions . They are formed by declarative knowledge to executable procedures compiled is. This process models the learning ability of ACT.

The third memory module is working memory, which is modeled very closely to human working memory . Like that, it has a limited number of memory locations ( slots ) that are filled with the current Wimees . These wimees can come from the outside world (via the input buffer, see below), or they can be the result of a production (see example).

Interfaces

Another important component of ACT are the input and output modules, so-called "buffers". These represent the interfaces to the outside world, i.e. correspond to the sensor and motor functions . For example, Wimees are created in the “visual buffer” which the sensor system (the “eye” of ACT, so to speak) has just extracted from the (virtual) environment. With the help of suitable rules, these wimees can be transferred to the declarative memory.

The target module defines which target should be pursued in the production system. In the production memory there are rules that determine which action is carried out when a selected goal is to be achieved, and which contents must be present in the working memory so that the action can be carried out successfully. This " pattern matching " leads to the selection of a production rule and determines the action of the output module.

example

My acoustic input buffer created the Wimee “the alarm clock is ringing” and transferred it to the working memory . The production memory is then queried as to whether this Wimee is the condition part of a production , for example if the alarm goes off, carry out the action "press snooze" . If this is the case, the action is carried out by the output module, its result “snooze button pressed” becomes a Wimee , and the production memory is queried again whether this is the condition part of another production .

Theory of Spreading Activation

In ACT, the so-called "Theory of Spreading Activation" (see activation spread ) is implemented, which was largely developed by John R. Anderson and represents the basis for the so-called " priming effects".

literature

Web links