Adaptive Control of Thought
Hello Everyone, so today I will be giving you some
brief information about network theory of memory has been developed by John
Robert Anderson. Called the Adaptive Control of
Thought (ACT) model of memory, it has evolved the
almost 40 years of its existence, and various versions (ACT-*, ACT-R) exist.
Based on analogies to computers, ACT has given rise
to several computer simulations of cognitive processing of different task. ACT
model distinguish among three kinds of memory systems: Working Memory,
Declarative Memory and Procedural Memory.
Declarative Memory
Declarative Memory (also known as descriptive
knowledge) is knowledge that we are conscious of and can verbalize. It is
information we directly encode from the environment and doesn’t require much
synthesization. It emphasizes what one needs to do to solve an issue rather
than how to solve it. In essence, it is content that can be recited or
memorized.
Declarative memory stores information in networks that contains nodes. There are different types of nodes, including those corresponding to spatial images or to abstract propositions. As with others network models, ACT models allow both for activation of any node and for spreading activation to connected nodes. Example of declarative memory include facts, world history, or rules for solving mathematical equations.
Procedural Memory
Procedural Memory (also known as imperative knowledge) is knowledge you use while performing a task, but may not be able to verbalize. It is information encoded from synthesizing and observing transformations of the environment (behaviors). Procedural memory is about how we do something.
This memory store represents information in production rules. Production rules specify a goal to achieve, one or more conditions that must be true for the rule to apply, and one or more actions that result from applying the rule. Examples of procedural memory include behaviors we do habitually, such as riding a bike or driving a car.
Another example, a typical college student could
use this production rule: “If the goal is to study actively and attentively (goal)
and the noise level in the dormitory is high (condition) and the campus library
is open (condition), then gather your study materials (action) and take them to
the library (action) and work there (action).” Okay, that example was a bit
contrived. But psychologists, computer scientists, and others have used
production rules to build computer programs that simulate human problem solving.
J. R. Anderson’s (1983) proposal not meant merely
to address the question of knowledge representation. Instead, his aim was to
create a theory of cognitive architecture, a theory of how human cognition
actually operates in practice. He proposed a system that includes both memory
storage and particular processing structures. Interestingly, the broad goal led
him to develop proposals about knowledge representation that fit well with
those of researchers whose aims were more focused.
In the ACT models, Working Memory is
actually that part of declarative memory that a very activated at any particular
moment. The production rules also become activated when the nodes in the
declarative memory that correspond to the conditions of the relevant production
rules are activated. When production rules are executed, they can create new
nodes within declarative memory. Thus, ACT models have been described as very “activation-based”
models of human cognition (Luger,1994).
History
Canadian psychologist John Anderson has an
extensive background researching and developing his Adaptive Control of Thought
model. His underlying assumption is that knowledge can be reduced into a theory
or model, and therefore that a system can be created to perform human cognitive
tasks. This assumption rests on the belief, articulated by Anderson in his 1990
book, that “we can understand a lot about human cognition without considering
in detail what is inside the human head. Rather, we can look in detail at what
is outside the human head and try to determine what would be optimal behavior
given the structure of the environment and the goals of the human”.
With the goal in mind of creating a model to depict
human knowledge, Anderson developed the Human Associative Memory (HAM) model
alongside cognitive psychologist Gordon Bower in 1973. This model computed the mathematical
theories of human cognition prevalent in the 1950s and 1960s. However, the
model only accounted for human memory and did not accomplish Anderson’s mission
of showing that all higher cognitive processes (memory, language, problem
solving, imagery, deduction and induction) have the same underlying system.
Replacing the HAM model, Anderson developed the ACT model in 1976, which was
able to account for these higher cognitive processes.
In 1990, Anderson developed another version of ACT
which he named ACT*. His hypothesis was that ACT*, alike ACT, showed that the
mind is unitary (all of thoughts and mental faculties can be explained by the
same underlying system) and that our experiences are stored in different
facilities (linguistic, geometric, etc.) depending on their subject matter.
Information comes through a ‘buffer’, known as working memory, which determines
whether the information should be stored as declarative and later retrieved, or
as procedural knowledge and executed in the moment to match current activity.
The model looks as follows:
The original ACT model only accounted for one kind of ‘cognitive unit’ that the mind would process, also known as ‘chunks’. These units were words and statements like ‘hate’ or ‘my mom is nice’. And Anderson speculated that working memory is able to process around 5 chunks at once before becoming overwhelmed.
However, in the ACT* model, information could come in the form of spatial images and temporal strings as well as abstract propositions. What counted as a cognitive unit was expanded. Temporal strings encode the order of a set of items; spatial images encode the spatial configuration of a cognitive unit; abstract propositions encode meaning. Another key element of ACT* was its suggestion that all information begins as declarative, and combines with method learning to produce procedural knowledge.
Moreover, ACT suggests that information was
serially processed, whereas ACT* suggests information could be processed
simultaneously (parallel processing). The last notable difference is that the
ACT* theory added a ‘sub-symbolic’ component: a feature that could determine
what meaning is activated when we run into a cognitive unit with many
possibilities. For example, if we encounter the sentence “The robber took money
from the bank,” the word ‘bank’ has two meanings: a financial institution, or
the land sloping down next to a body of water. The sub-symbolic component of
ACT* activates the financial institution meaning because it knows that meaning
is processed in relation to money or robbery.
A few years later, Anderson, alongside his
colleagues at Carnegie Mellon University, developed the most recent model of
ACT by combining the original with rational analysis. This model is known as
ACT-R and predicts behavior based on the idea that humans act in ‘optimal’
(see: rational) ways. The ACT-R is essentially a production system, like a
machine, that operates according to ACT*. The theory was moved to a computer
program, on which researchers can download the ACT-R code, input information
about a specific task, and analyze people’s predicted performances.
I hope you found this article helpful & learnt something new. I would be glad to know your thoughts on it in the comment section below. If you like this article, share it with your friends and colleagues!
THANK YOU
