Problem 5 Flashcards
Rules can be used tp represent many different kinds of knowledge.
- General info about the world
- Info about how to do things in the world
- Linguistic regularities
Cognitive architecture
Refers to a quantitative theory that applies to a broad array of behaviors + tasks
–> integrates theories of
a) memory
b) perception
c) action
Cognitive modeling
Refers to a
- theory about the structure of the human mind
- computational instantiation of such theory used in the fields of AI + computational cognitive science
What is the goal of cognitive architecture ?
Summarizing the various results of cognitive psychology in a comprehensive computer model
Cognitive model
Refers to an approximation to human cognitive processes for the purposes of comprehension + prediction
BUT: can be developed within OR w/o cognitive architecture
What is the goal of cognitive models ?
Focusing on a single cognitive phenomenon/process
- How 2 or more processes interact
- making behavioral predictions for a specific task/tool
Cogntivie science
Is concerned with understanding the processes that the brain uses to accomplish complex tasks
Name 4 major characteristics of cognitive models.
- Described in formal mathematical/computer language
- Derived from basic principles of cognition
- Builds a bridge btw behavior + neural underpinnings
- Describes human info processing at a more abstract level of analysis
Why do cognitive models have an advantage over conceptual models ?
Cognitive models can be used to derive new predictions fro new relationships that go beyond the original data
–> can be used to predict how with practice
a) the variance of the response time changes
b) accuracy changes
What is the difference between conceptual + cognitive models ?
Conceptual models aren’t based on any cognitive principles
Why are cognitive medals beneficial over neural models ?
Cognitive models provide an abstract level of analysis making it computationally easy to derive precise predictions from complex tasks
–> neural models level of analysis makes it too difficult to address complex cognitive tasks
Name the steps involved in cognitive modeling.
- Taking conceptual theoretical framework + reformulate assumptions into more mathematical language
- Making additional detailed (ad hoc) assumptions if theory is weak
- Estimate the parameters from some of the observed data
- Compare the predictions of competing models with respect to their ability to explain the empirical results
- Start all over + reformulate framework in light go the feedback obtained from new experimental results
Symbolic modeling
Refers to one approach to cognitive modeling
Representation
Consists of symbols
–> each representing a particular component of the environmental situation
How is a problem solved ?
By manipulating the symbols according to the given inference rules
–> this way one finds the best combination that answers to the question
Rules
Refer to “if-then” structures
–> similar to conditionals but have different representation + computational properties
Name the most vital properties of rules.
- Simple structures
- -> if + then - Can have multiple conditions/ actions
- Can represent
a) general info
b) info how to do things
c) linguistic regularities
Rule based processing
Uses rules and can be
a) serial
- -> one being applied at a time
b) parallel
- -> many being applied simultaneously
Name the 3 most notable Rule based cognitive systems.
- General problem solver (GPS)
- -> simulates human solutions to various kinds of problems - Adaptive control of thought (ACT)
- SOAR
- -> both aim to develop fixed computational building blocks, necessary for general intelligent agents
Are rule based system superior to logic based systems ?
- Increased computational power but lose out on some representational power
- Most psychological applications
- -> explains conditioning, learning - Neurological plausibility
- -> rules are stored in a set of buffers in the PFC - Practical applicability
BUT: do not have to be regarded as universally true
Production rules
Consist of two parts
- Sensory precondition
- -> if-statement - Action
- -> then-statement
THUS: if the productions precondition matches the current state of the world, then the production is triggered
Production system
Contains
- Database
- -> WM, maintains data about current states - Rule interpreter
- -> provides a mechanism for prioritizing productions when more than one is triggered
ACT-R
Adaptive control of thought - Rational
Refers to a cognitive architecture that suggests that complex cognition arises from an interaction of procedural + declarative knowledge
–> furthermore it
- aims to find the basic and irreducible cognitive + perceptual operations that enable the human mind
- describes human cognition as a set of independent modules that interact through a production system
Production system
Consists of production rules whose conditions are matched against the modules
–> if conditions are met, it can fire and the action specified can be carried out
There are 5 different modules, each implementing a different cognitive process.
Name them.
- Declarative module
- -> info storage in + retrieval from declarative M - Intentional module
- -> stores a persons goals - Imaginal module
- -> holds info necessary to perform a current task - Visual module
- -> perception - Manual module
- -> motor actions
BUT: within each module, info can only be processed in a serial manner, while different can operate in parallel
How does the production rule coordinate the modules ?
By acting on the info available in the buffers
–> thus it can never access all info but just those in the buffers
What is the ACT-R composed of ?
- Symbolic system
- -> e.g. production rules, modules, buffers - Sub-symbolic system
- -> equations which determine access to the info stored in the modules + buffers
Declarative knowledge
Involves facts
–> is represented by chunks + created by simple encoding of objects in the environment
Procedural knowledge
Involves knowledge about how we do things
–> is represented by production rules related by simple encodings of transformation in the environment
Process models
Predict
- what decision a person will make
- how info that was used to make the decision will be processed
Compensatory models
Claim that people evaluate alternatives by using knowledge about their attributes as cues
–> for each alternative, one counts the + vs - cues, making the most + one the correct
Compensatory decision making
Refers to evaluating each alternative in terms of relevant attributes
–> selecting the highest weighted score
