Cognitive Modeling Declarative and Procedural Knowledge 2 Lecture - - PowerPoint PPT Presentation

What is ACT-R? Declarative and Procedural Knowledge The ACT-R Architecture Examples for Modules Summary

Cognitive Modeling

Lecture 3: ACT-R Frank Keller

School of Informatics University of Edinburgh keller@inf.ed.ac.uk

January 31, 2005

Frank Keller Cognitive Modeling 1 What is ACT-R? Declarative and Procedural Knowledge The ACT-R Architecture Examples for Modules Summary

1

What is ACT-R? Unified Theory of Mind Requirements Domains

2

Declarative and Procedural Knowledge Declarative Knowledge Procedural Knowledge

3

The ACT-R Architecture Modules Architecture Processing

4

Examples for Modules Declarative Memory Procedural Memory

5

Summary Reading: Anderson (1996)

Frank Keller Cognitive Modeling 2 What is ACT-R? Declarative and Procedural Knowledge The ACT-R Architecture Examples for Modules Summary Unified Theory of Mind Requirements Domains

A Cognitive Architecture

ACT-R provides “some important new insights into the integration of cognition” (Anderson 1996): it is a unified theory of cognition realized as a production system; it is designed to predict human behavior by processing information and generating intelligent behavior itself; it integrates theories of cognition, visual attention and motor movement; it successfully models high-level cognitive phenomena, such as working memory, scientific reasoning, skill acquisition, HCI.

Frank Keller Cognitive Modeling 3 What is ACT-R? Declarative and Procedural Knowledge The ACT-R Architecture Examples for Modules Summary Unified Theory of Mind Requirements Domains

A Unified Theory of Mind

A single system (mind) produces all aspects of behavior, even if made up of parts. We need a theory that gives the whole picture. Argument for integration and application: better for tackling applied problems, which are less feasible if

• nly isolated research programs are addressed.

Example Learning mathematics involves: understanding mathematical expressions, reading, language processing (instructions and word problems), spatial processing, problem solving, reasoning and skill acquisition.

Frank Keller Cognitive Modeling 4

What is ACT-R? Declarative and Procedural Knowledge The ACT-R Architecture Examples for Modules Summary Unified Theory of Mind Requirements Domains

Example: Salvucci and Macuga (2001)

Predicts effect of cell phone use on driving. Aim: predicting results, not just fitting results to experimental data: developed separate ACT-R models of driving and of using a cell phone; put these together to predict effects of driving on cell phone use, and of phone use on driving. Compared four ways of dialing ⇒ good predictions. Effect on driving (swerving) ⇒ only the full manual dialing condition has significant impact. Note: did not estimate parameters to fit data – had no data – used established ACT-R parameters.

Frank Keller Cognitive Modeling 5 What is ACT-R? Declarative and Procedural Knowledge The ACT-R Architecture Examples for Modules Summary Unified Theory of Mind Requirements Domains

Requirements for Cognitive Architectures

Understand environment in which human cognition occurs, and perceptual and motor access to that environment:

1 the environment is not constrained (as in an experiment):

robust behavior is needed to deal with errors, the unexpected and the unknown;

2 the importance of a priori predictions.

Model fitting criticized: belief that parameter estimation permits fitting to any pattern of data. “Rather than just predicting results or estimating parameters to predict quantitative predictions the ideal model should predict absolute values without estimating parameters” (Anderson 2002).

Frank Keller Cognitive Modeling 6 What is ACT-R? Declarative and Procedural Knowledge The ACT-R Architecture Examples for Modules Summary Unified Theory of Mind Requirements Domains

Domains of ACT-R Models (examples)

1 Perception and Attention: visual search; eye movements;

task switching; driving behavior; situational awareness.

2 Learning and Memory: list memory; implicit learning; skill

acquisition; category learning; arithmetic; learning by exploration and example.

3 Problem Solving and Decision Making: use and design of

artifacts; spatial reasoning; game playing; insight and scientific discovery.

4 Language Processing: parsing; analogy and metaphor;

learning; sentence memory; communication and negotiation.

5 Other: cognitive development; emotion; individual differences. Frank Keller Cognitive Modeling 7 What is ACT-R? Declarative and Procedural Knowledge The ACT-R Architecture Examples for Modules Summary Declarative Knowledge Procedural Knowledge

Declarative and Procedural Knowledge

ACT-R: cognition emerges as the result of interaction between procedural and declarative knowledge. Declarative knowledge: factual, holds information; represents things remembered or perceived. Example

“2 + 2 = 4”, “Edinburgh is the capital of Scotland”.

Procedural knowledge: encodes processes and skills necessary to achieve a given goal: production rules that fire when conditions satisfied and executes specified actions. Example

IF goal is to add two digits d1 and d2 in a column and d1 + d2 = d3 THEN set as a subgoal to write d3 in the column.

Frank Keller Cognitive Modeling 8

What is ACT-R? Declarative and Procedural Knowledge The ACT-R Architecture Examples for Modules Summary Declarative Knowledge Procedural Knowledge

Declarative Knowledge

Basic units of declarative knowledge are chunks. They might represent: situational awareness, “there is a car to my right”; navigational knowledge, “Princes Street intersects with North Bridge”; driver’s goals and intentions, “pick up friend at the corner”. Chunks defined by types (categories) + slots (attributes):

1 “the dog chased the cat”: type chase, slots are agent (dog)

and object (cat);

2 “4 + 3 = 7”: type addition-fact, slots addend1, addend2, and

sum.

Frank Keller Cognitive Modeling 9 What is ACT-R? Declarative and Procedural Knowledge The ACT-R Architecture Examples for Modules Summary Declarative Knowledge Procedural Knowledge

Procedural Knowledge

Procedural knowledge can be decomposed into conditions and actions. Conditions: may depend on declarative knowledge (i.e., recall of a chunk) and/or sensory input from environment; a specification of the goal and a number of chunks;

• ften test the contents of buffers.

Actions: can alter declarative knowledge, change goals, or initiate motor actions in the environment; produce changes in the contents of buffers.

Frank Keller Cognitive Modeling 10 What is ACT-R? Declarative and Procedural Knowledge The ACT-R Architecture Examples for Modules Summary Declarative Knowledge Procedural Knowledge

Procedural Knowledge

Example IF current goal is to encode a distant perceptual point for steering (test of goal) and there is a tangent point present (i.e., a curve) (test of contents of visual system) THEN shift attention to this point and encode its position and distance (build a visual representation)

Frank Keller Cognitive Modeling 11 What is ACT-R? Declarative and Procedural Knowledge The ACT-R Architecture Examples for Modules Summary Modules Architecture Processing

Modules in ACT-R

Modules devoted to: identifying objects in the visual field; controlling the hands; retrieving information from declarative memory; keeping track of current goals and intentions. Central production system: not sensitive to activity of modules, but responds to information deposited in buffers of these modules; each module makes this available as a chunk in a buffer.

Frank Keller Cognitive Modeling 12

What is ACT-R? Declarative and Procedural Knowledge The ACT-R Architecture Examples for Modules Summary Modules Architecture Processing

Particular Modules

1 Goal buffer: associated with the dorsolateral prefrontal cortex

(DLPFC): keeps track of internal state in solving a problem.

2 Retrieval buffer: associated with ventrolateral prefrontal

cortex (VLPFC): holds information retrieved from long term declarative memory.

3 Two visual buffers: associated with dorsal “where” path of

the visual system and ventral “what” system: keep track of visual objects and their identity.

4 Manual buffer: associated with motor and somatosensory

cortical areas controlling and monitoring hand movement: responsible for control of the hands.

Frank Keller Cognitive Modeling 13 What is ACT-R? Declarative and Procedural Knowledge The ACT-R Architecture Examples for Modules Summary Modules Architecture Processing

Modules and Brain Regions

Frank Keller Cognitive Modeling 14 What is ACT-R? Declarative and Procedural Knowledge The ACT-R Architecture Examples for Modules Summary Modules Architecture Processing

ACT-R: Organization of Information

Frank Keller Cognitive Modeling 15 What is ACT-R? Declarative and Procedural Knowledge The ACT-R Architecture Examples for Modules Summary Modules Architecture Processing

ACT-R Operates in Real Time

Each covert step of cognition or overt action has latencies associated based on psychological theories and data. Critical cycle in ACT-R in which: the buffers hold representations determined by the external world and internal modules; patterns in these buffers are recognized and a production fires; buffers then updated for another cycle. Assumption in ACT-R is that critical cycle takes about 50 ms: estimate emerged, e.g., in SOAR (Newell 1990).

Frank Keller Cognitive Modeling 16

What is ACT-R? Declarative and Procedural Knowledge The ACT-R Architecture Examples for Modules Summary Modules Architecture Processing

Parallel and Serial Processing

Assumes mixture of parallel and serial processing: within each module is a great deal of parallelism. Example Visual system simultaneously processes the whole visual field; declarative system executes parallel search through memories in response to a retrieval request. But two levels of serial bottlenecks in the system: the content of any buffer is limited to a single declarative unit, a chunk (so only a single memory can be retrieved at a time,

• r a single object encoded from the visual field);
• nly a single production is selected at each cycle to fire.

Frank Keller Cognitive Modeling 17 What is ACT-R? Declarative and Procedural Knowledge The ACT-R Architecture Examples for Modules Summary Modules Architecture Processing

Symbolic and Sub-symbolic

Hybrid cognitive architecture: symbolic structure is a production system; sub-symbolic structure represented by massively parallel processes summarized by a number of mathematical equations. The sub-symbolic system controls many of the symbolic processes. Each symbolic construct (production or chunk) has sub-symbolic parameters that reflect past use. The system keeps track of the general usefulness of symbolic information.

Frank Keller Cognitive Modeling 18 What is ACT-R? Declarative and Procedural Knowledge The ACT-R Architecture Examples for Modules Summary Declarative Memory Procedural Memory

Example Module: Declarative Memory

Records of chunks formed in the various perceptual-motor buffers can be retrieved. ACT-R makes chunks active to degree that keypast experiences indicate that they will be useful at this particular moment. The activation of a chunk is the sum of: a base level activation, reflecting its general usefulness in the past, and an associative activation, reflecting its relevance to the current context. The activation of a chunk controls its probability of being retrieved and its speed of retrieval.

Frank Keller Cognitive Modeling 19 What is ACT-R? Declarative and Procedural Knowledge The ACT-R Architecture Examples for Modules Summary Declarative Memory Procedural Memory

A Declarative Chunk

A presentation of a declarative chunk with its subsymbolic properties (Anderson 1996).

Frank Keller Cognitive Modeling 20

What is ACT-R? Declarative and Procedural Knowledge The ACT-R Architecture Examples for Modules Summary Declarative Memory Procedural Memory

Example Module: Procedural Memory

In ACT-R contents of many buffers will vary continuously, the conditions of a production match to a certain degree. Conflict Resolution: multiple productions may match but only

• ne can fire.

In ACT-R a sub-symbolic utility function estimates: the probability that if a production is chosen the current goal will be achieved; the relative cost (time to achieve goal); benefit (value of current goal) associated with each production; and selects for execution the production with the highest utility.

Frank Keller Cognitive Modeling 21 What is ACT-R? Declarative and Procedural Knowledge The ACT-R Architecture Examples for Modules Summary

ACT-R Features

One production fires at a time ⇒ serial bottleneck. Can predict time sharing between two tasks provided that they don’t make simultaneous demands for production firing. Committed to a mix of symbolic and subsymbolic processes. Has been compared to brain imaging data. Different cortical and supporting neural structures serve as modules that broadcast their contents to the basal ganglia for pattern recognition and production selection.

Frank Keller Cognitive Modeling 22 What is ACT-R? Declarative and Procedural Knowledge The ACT-R Architecture Examples for Modules Summary

On Production Rules

Productions specify pathways of influence from cortex to basal ganglia and back again. Commitment is to a pattern of interaction. Claim: information processing is constrained to follow paths like these which test patterns of activation from diverse areas and transmit information to different areas. So not saying “production rules are coded in some data structures in the brain”. Symbolic assumptions in ACT-R comes down to a claim about strong constraints on connections in the nervous system.

Frank Keller Cognitive Modeling 23 What is ACT-R? Declarative and Procedural Knowledge The ACT-R Architecture Examples for Modules Summary

Summary

1 There are multiple independent modules whose information

processing is encapsulated.

2 The modules can place chunks reflecting their processing in

their buffers and the production system can detect when critical patterns are satisfied by these chunks.

3 From the productions whose conditions are satisfied:

a single production will be selected at any time and fire; leading to updates to various buffers that in turn can trigger information processing in their respective modules.

Frank Keller Cognitive Modeling 24

What is ACT-R? Declarative and Procedural Knowledge The ACT-R Architecture Examples for Modules Summary

Summary

1 While chunks and productions are the symbolic components

• f the system reflecting its overall information flow,

chunks have subsymbolic activations, and productions have subsymbolic utilities that control which chunks and productions get used.

2 Learning can involve either acquiring new chunks and

productions or tuning their subsymbolic parameters;

3 These processes are stochastic and take place in real time. Frank Keller Cognitive Modeling 25 What is ACT-R? Declarative and Procedural Knowledge The ACT-R Architecture Examples for Modules Summary

References

Anderson, John R. 1996. ACT: a simple theory of complex cognition. American Psychologist 51(4):355–365. Anderson, John R. 2002. Spanning seven orders of magnitude: A challenge for cognitive modelling. Cognitive Science 26:85–112. Newell, Alan. 1990. Unified Theories of Cognition. Harvard University Press, Cambridge, MA. Salvucci, D. D. and K. L. Macuga. 2001. Predicting the effects of cell-phone daialing

• n driver performance. In E. Altmann, , A. Cleeremans, C. D. Schunn, and W. D.

Gray, editors, Procedings of the 4th International Conference on Cognitive

• Modeling. Lawrence Erlbaum Associates, Mahwah, NJ, pages 25–30.

Frank Keller Cognitive Modeling 26