Cognitive Architectures

Cognitive Architecture

Articulated by Allen Newell, early founder of AI
The ultimate goal of work in cognitive architecture is to provide... the underlying structure that would enable a system to perform the full range of cognitive tasks, employ the full range of problem-solving methods and representations appropriate for the tasks, and learn about all aspects of the task and its performance on them.
Soar : an architecture for general intelligence

Cognitive architecture

Analogous to a computer architecture
Memory how is knowledge represented
Operators What operations can be performed on the knowledge
Working memory How are temporary intermediate results managed
I/O What triggers reasoning, what happens?

Cognitive Architecture

  • Goal: operators are general purpose for all reasoning, just like computer architecture.
  • Goal: the architecture supports learning, i.e., autonomous acquisition of new knowledge
  • Newell also wanted an architecture that could model and predict human capabilities and even reaction times.

Deductive retrieval

Memory A set of backward chaining rules
Operators Backward chaining and unification
Working memory Binding lists
I/O Backchaining queries lead to query responses

Pattern-directed inferencing

  • Not really a cognitive architecture
  • Memory = list of pattern + responses
  • Everything else is regular computer architecture

SOAR

Memory A set of forward-chaining production rules
Operators Matching rules to tokens to generate new tokens
Working memory Set of active tokens
I/O Problem representation inputs lead to solution responses
The SOAR Architecture

MOPs

Memory A set of frames in a semantic hierarchy
Operators frame indexing, retrieval, generalization, storage
Working memory Set of active frames
I/O Partial frame inputs retrieve or generate new MOPs
Frame matching