Class Notes: Chapter 5

Chapter 5: Section 5.1

Solving the integration challenge:

The integration found in studies of reasoning, and the analysis of the BOLD signal, are cases of local integration. They apply to a very restricted range of phenomena
What can be said in general about a methodology for integrating various disciplines in cognitive science?

Describe the logical positivists’ model of intertheoretic reduction. Why is the model likely to be unsuccessful in cognitive science?

Reduction in the sciences:
Phenomena at one level are explained by reference to relationships at a more fundamental level
e.g., Biology ? Chemistry ? Physics
Recall the organization of the neurosciences

In the cognitive sciences the prospect seems unlikely
Will biology explain psychology?
Reduction in psychology: There are opposing attitudes:
“Psychology is something to do until the biologist arrives” !?
Reductionism = materialism = evil !? Removes what is essentially human?

Daniel Dennett: Distinguish between serious reduction and greedy reduction
It's often the latter that scares people
Serious reduction builds cranes – from the ground up, one spar at a time. Greedy reductionism uses sky hooks.
Serious reduction uses established theories at one level to explain phenomena at other levels.
Behaviorism is an example of greedy reduction. Explain all behavior in terms of conditioning
The underlying principles do not exist.

Describe Cummins’s model of functional decomposition.

Decomposition of a cognitive capacity into more basic capacities, with a model that specifies the interconnections
Example from memory: Three processes
Registration (encoding), Storage, Retrieval
These are functionally independent processes

Why do cognitive scientists believe that short-term and long-term memory are different systems?

Double dissociation: Interventions that affect one do not affect the other (or affect it in different ways)
The reverse holds also.
LTM may be further decomposed.
This is a kind of reduction, but not inter-theoretic

Describe Baddeley’s functional decomposition of the short-term memory system.

Another example of functional decomposition, addressing shortcomings in the older models of memory
Research elaborates on the components and further refines them
But there is (as yet) no effort to reduce theories of STM to other theories

Why does cognitive science tend to lack laws?

There are a number of so-called laws: Stevens’ law, Weber’s law, Fechner’s law
But while cognitive science observes regularities (relationships), these do not lead to explanation at a more fundamental level

Is Cummins right that functional decomposition is the main methodology of psychology? Should it be the main methodology?

It certainly keeps psychologists busy
It enables prediction and, often, successful manipulation
Example: Using theories of memory to improve performance
Is there any viable alternative?

Chapter 5: Section 5.2

Why does Marr's (or Stanovich's) approach represent a possible solution to the integration challenge?

Three levels at which we can deal with cognition. The levels are distinct but not independent of each other
The three levels may represent work by separate disciplines
Lower levels address issues raised at higher levels

Explain the modular/non-modular distinction. Why is it important in thinking about levels of analysis?

We met this distinction in the discussion of evolutionary approaches.
Modular systems are specific to one domain
Non-modular systems are domain-general
Designed to solve a particular problem of adaptation

What is the "frame problem"? Why does it present a problem for the levels of analysis approach?

Non-modular systems are intended to apply in any situation.
Dennett’s parable of the robot illustrates the problem of defining algorithms for such systems.
The algorithm must address a potentially infinite set of circumstances.
What are the prospects for success at the algorithmic level?

Domain-general rules: IF A THEN B (for all possible A and B)
How can a system consider changes brought about by an action without explicitly considering everything the action does not bring about
There is again a potentially infinite set of things that might not happen.
This is one reason why evolutionary psychologists prefer a modular approach
Other implications are seen in the next chapter

Is it convincing that Marr’s approach works only for modular systems, and hence cannot represent a global solution to the integration challenge?

Suppose that all systems are modular …
Consider complex decisions as an example of tasks where a non-modular system might be used
Note the difficulty of the operating at the algorithmic level

The Problem with Domain General Algorithms

A medical problem: Should I elect for surgery to fix a damaged knee?

There exists a domain-general procedure:
Consider all possible outcomes following surgery and following non-surgery
Evaluate those outcomes
Combine the evaluations for each choice
Select the best option

This is a non-terminating algorithm!

Modular alternatives

Always follow my doctor’s advice
Never opt for knee surgery unless one is immobile

The point has been used by some to argue against using domain-general procedures

But, does this necessarily rule out the three-level approach as a framework for integrating theories in cognitive science?
Some theories are clearly aimed at implementation. E.g., neurological approaches
Some theories are clearly aimed at the computational/intentional level - e.g., evolutionary
Some theories are clearly algorithmic – the framing issue suggests they may need to be modular

Modular versus Domain-General

Modular systems:
They use a finite set of rules that apply to a fixed set of conditions
Easier to see how they can evolve (The robot’s rebellion hypothesis is not yet well defined, thus not testable)
But, how do we solve problems not previously encountered?

Domain-General Systems:
Problem-solving mechanisms that can be used at any time
But, a finite set of rules will be subject to the frame problem
A totally general system may not be definable using a Turing machine

So What?

Why worry about Turing machines?
We know a lot about Turing machines. Their properties are well understood
They fit well into Marr’s tri-level system
Science may prefer an approach that is tractable, even if it is less than ideal

Chapter 5: Section 5.3

Describe the mental architectures approach to the integration challenge. What are its three important questions?

Three questions:

In what format does a system carry information?
How does the system transform information?
How is the mind organized to function as an information processor?

Taking seriously the idea that cognitive systems are information processing systems.

Note the difference between faculties, which are domain general, and modules.
Example: Perception is a general faculty. Face recognition is a module

We’ll explore two sets of answers to the questions: Physical symbol systems, Connectionist networks

One More Approach to Integration: Integration through Game Theory

Herbert Gintis: A strong proponent of interdisciplinary integration:
“How can the basic conceptual frameworks . . ., as outlined by their respective Great Masters and as taught to Ph. D. candidates, have almost nothing in common?”
Proposes a common framework using game theory

Game theory is the mathematics of rational choice in games such as prisoner’s dilemma and centipede game
To this Gintis adds an evolutionary component: “The choreographer” - a set of social norms
Norms are learned by individuals genetically predisposed to recognize and obey them.
People tend to obey the norms even when costly to do so
Hence, for example, cooperation in the PD and trust in the C.G.

Integration of the Cognitive Sciences through Game Theory?

“I propose . . . a conceptual integration . . . that is analytically and empirically defensible and could be implemented now were it not for the virtually impassible feudal organization of the behavior disciplines . . ., the structure of research funding agencies . . . and interdisciplinary ethics that value comfort and tradition over the struggle for truth”.
Will it work? It remains to be seen.
But Gintis implies that the integration challenge is harder to meet than Bermúdez believes,for political reasons