The Newell Test for a Theory of Mind

The Newell Test for a Theory of Mind

• Anderson, John R., & Lebiere, Christian (forthcoming), “The Newell Test for a Theory of Mind”, Behavioral & Brain Sciences

• Newell, Allen (1980), “Physical Symbol Systems”, Cognitive Science 4: 135-183.

• Newell, Allen (1990), Unified Theories of Cognition (Cambridge, MA: Harvard University Press).

The Newell Test

• Based on lists of criteria that cognitive theories should satisfy.

• Anderson & Lebiere:

• Classical connectionism:
• Newell-test strengths:
• Addresses empirical phenomena
• Especially in language, cognitive development
• NT weaknesses:
• Fails to acknowledge symbolic level of thought
• ACT-R (Anderson’s own theory):
• NT strengths:
• Includes symbolic & subsymbolic components
• Both are tightly integrated
• NT weaknesses:
• Lacks intensive analyses of some NT criteria

Newell’s Criteria: Summary

• 12 functional constraints on cognitive architecture

• First 9:
• Things needed “to implement human intellectual capacity”
• Last 3:
• “Constraints on how these functions are to be achieved”

• There are others, not on the list:

• E.g., a system could satisfy all 12, but “not correspond to the human mind”
• So: could add others (but won’t):
• Behavioral adequacy:
• cognitive theory should correspond to details of human cognition
• Be capable of practical application (e.g., to education, therapy)

• There is a grading scheme!

Newell’s 12 Criteria

• Flexible behavior

• Real-time performance

• Adaptive behavior

• Vast KB necessary

• Dynamic behavior for

• Knowledge integration cognition

• Natural language

• Consciousness

• Learning

• Development

• Evolution constraints

• Brain

1. Flexible Behavior

• I.e., computational universality

• Grading:

• If theory is well-specified, then s/b easy to determine whether computationally universal
• Doesn’t require that:
• people find everything equally easy
• human performance be error free

2. Real-Time Performance

• Theory should explain how humans can satisfy criterion 1 in real time

• Grading:

• If theory has well-specified constraints on speed of its processes, then s/b easy to determine whether it can achieve real time for specific cases of cognition

3. Adaptive Behavior

• Human computations are relevant to their needs

• How do basic processes of the architecture serve useful functions?
• How is the whole system put together?
• Does its overall computation meet human needs?

• Grading:

• If theory is completely specified, then s/b able to determine if its behavior would be functional in real world

4. Vast Knowledge Base

• Distinguishes human cognition from expert systems

• Problems:
• Not all K is equally reliable or relevant
• Relevance can change with time
• Storage/retrieval problems

• Grading:

• Determine how performance changes with scale of KB
• If theory is well specified, then s/b able to formally analyze
• Size may affect performance
• Cf. learning names of 20 students vs. 200 students

5. Dynamic Behavior

• Precondition for survival:

• World can change in unexpected & uncontrollable ways
• Human actions can have unexpected effects
• Requires theories of perception, action, & cognition
• BUT: NOT(reaction is all there is to cognition)
• But also: NOT(cognition can ignore external world)

• Grading:

• Test system in an uncontrolled environment

6. Knowledge Integration

• I.e., achieving the capability to “combine symbols intellectually” in order to make inferences about the external world

• Grading:

• Can the theory deal with inference, induction, metaphor, analogy

7. Natural Language

• Complete theory of mind must be able to communicate in NL

• Newell: language depends on symbol manipulation (! )

• Grading:

• Read a passage & answer questions about it.
• Requires parsing, comprehension, inference, use of prior knowledge
• Conversational ability is nice, but not necessary

8. Consciousness

• Important for a full theory of human cognition

• [what about a full theory of cognition, simpliciter?]

• Grading:

• Ability to produce subliminal perception, implicit learning & memory, metacognitive processes
• & show how they are functional aspects of human cognition

9. Learning

• “Theory of cognition must account for humans’ ability to acquire their competencies”

• Grading:

• Semantic memory, episodic memory, skills, priming, conditioning
• Based on Squire’s (1992) theory of memory in the hippocampus

Development (constraint #1)

• Human cognition is constrained to unfold as the organism grows & responds to experience.

• Grading:

• Can the theory account for specific cases of developmental progression?

11. Evolution (constraint #2)

• Evolutionary constraint:

• Human cognitive abilities evolved

• Comparative constraint:

• Differences between human and other-mammalian cognition
• E.g., cognitive plasticity, language
• What’s unique to human cognition?

• Grading:

• How does the theory relate to these constraints?

12. Brain (constraint #3)

• Degree to which data about brain functioning constrains the cognitive theory.

• Grading:

• Enumeration:
• Mapping of parts of the cognitive architecture onto brain structures
• Proof:
• Brain computations match cog-arch computations

Grading

• BEST: criterion on which theory has done the best

• Better: 4 criteria on which the theory has done better

• mixed: 2 criteria on which theory has most mixed record

• worse: 4 criteria on which the theory has done worse

• worst: criterion on which theory has done the worst

Scoreboard

Comparative Ranking