Chapter 1

There were the behaviorists, whose minds were made up ...
… or would have been, except they did not think they had minds.
There was Turing, who showed how a machine could think just like a person …
… and why you can't trust machines any more than you can trust people.
There were the linguists, who knew the difference between deep structure and surface structure …
… although they could never explain it in words that others could understand.
There were the information theorists, who discovered a magic number …
… and understood, long before cell phones were invented, why you should never drive and use a cell phone.

Behaviorism

For a behaviorist, all theories must be expressed in terms of observable behavior – i.e., stimuli and responses
Learning, then, consists of associations between stimuli and responses: S --> R
“Habit strength” is acquired as the consequence of reinforcement
The challenge for the critics: How can we show experimentally that “thoughts” exist?

Describe the experiment that led Tolman to argue for the existence of cognitive maps. Explain why the results of the experiment support this argument.

Learning a place is easier (faster) than learning a response
Hard to explain if learning consistsonly of S-R associations
But is there another way to explain these results, in S-R terms?

Explain the difference between linearly and hierarchically organized behavior. Give an example of a hierarchical behavior.

There are several levels at which one can describe the behavior
Higher levels control lower levels
Consider, e.g., attending class
Levels: The broad plan, down to muscle movements
Lower levels operate without our awareness

Why do hierarchically organized behaviors pose a problem for behaviorists?

Difficult to explain in terms of stimulus-response chains: We need to postulate higher levels of control
Notice how important are the concepts of “planning” and “goals”. This is how we explain the control by higher levels – mentalistic constructs

For discussion: Behaviorism is no longer the dominant paradigm in psychology, but what can we learn from it? Are parts of behaviorist thinking correct?

Consider what the behaviorists were rebelling against
Introspection was the central methodology
A fundamental principle of science: empiricism, which may be impossible with introspection

One can always add further elements to a theory to handle apparent failures
Behaviorists added the theoretical constructs of “implicit responses” and “implicit stimuli”: S - r - s - R
In the long run, there may be no empirical way to differentiate behavioral theories from the alternatives

A recurring theme in psychology: The difficulty of making empirical distinctions among competing theories

How then does one make the differentiation?
Pragmatic considerations perhaps: Ease of application. Practical value in solving problems
There are some situations where a behaviorist approach seems appropriate (e.g., some training situations)
Others where concepts such as goals and planning are necessary – the behaviorist analysis is just too complex

The theoretical approaches may operate at different levels
Several authors have tried to spell out in a systematic way the levels at which theories might operate
We’ll look at classifications suggested by David Marr (for vision) and Keith Stanovich (for reasoning)

Levels of Analysis

Marr: Vision

Stanovich: Reasoning

Computational
What purpose is served by the behavior?
Intentional
What are the person’s goals, knowledge, and beliefs?

Algorithmic How is the purpose achieved?
Algorithmic What cognitive processes account for these intentions?

Implementational
How is the algorithm implemented ? Biological What biological processes operate to make this happen?

Why did the rat learn to run the maze?
Intentional level: It felt hungry. It wanted food. It thought food was at the end of the maze
Computational level: Learning enables an animal to adapt to changes in its environment
Algorithmic level: It had developed a cognitive map of the maze. It planned to follow the map
Implementation level: The map was represented by a network of implicit s-r associations

What is an algorithm? (Give an example)

A rule for accomplishing a goal. Includes decision points if necessary
IF condition THEN action

What is a Turing machine?

Idealized computing machines that can be specified purely mathematically
Do not have to take any particular physical form (i.e. multiply realizable)
The machine table of a Turing machine is the program, an algorithm that specifies the appropriate output for a given input
An algorithm is closely related to the notion of proof in mathematics

Gödel's Incompleteness Theorem

Kurt Gödel proved that no system of mathematics is self-sufficient (it contains statements that are true but cannot be proved to be true)
To prove the truth of some statements, a separate algorithm must be employed
Turing’s results on effective calculability prove essentially the same thing
The result defines the limits on what any algorithmic procedure can accomplish

(See separate notes on Gödel's Incompleteness Theorem)
Some have seen this as “proof” that minds cannot be machines. But maybe the same limits apply to minds
Gödel used the result to argue for dualism and the existence of God
(Dualism: The mind exists independently of the brain, or material objects. Cognitive scientists are as a rule materialists)
Turing concluded that humans are merely computing machines, and there is no God

What is the Church-Turing thesis, and why is it relevant to cognitive science?

Church (following Turing) wanted to provide a precise mathematical definition for the intuitive notion of effective calculability
Effectively calculable functions turn out to be the functions that can be computed by a Turing machine
The thesis provides an empirical solution to the problem of studying mental processes.
C-T defines what minds and machines can do
Note how this takes care of the Behaviorist argument that we cannot study minds scientifically
It’s no different from studying machine scientifically

For discussion: Is computation what minds do, as many cognitive scientists believe?

Is the mind a machine? Can machines think?
Notice: In order to answer either of these questions, we need to wrestle with the problem of definitions
E.g., if you define “think” as something that only living organisms can do, the questions are answered, but in an uninteresting way.

Explain the distinction between the deep (or phrase) structure and the surface structure of a sentence.

Consider three sentences:
“These students are happy to help”
“These students are hard to help”
“It is hard to help these students”
Equivalent surface structure versus equivalent deep structure

In what way did the concept of a transformational grammar contribute to the development of cognitive science?

Two ideas we have seen already:
A hierarchical organization that defies a behaviorist’s explanation
An algorithm: Rules for transforming sentence structures
The rules get us from one level to the other

Explain what chunking is. Give an example. Why do we need to chunk information?

We can only hold 7 items in STM - but we can use chunking/recoding to hold far more than 7 bits of information
I.e., information is processed in terms of a particular coding or format
Think of how we remember phone numbers…or names

Explain the phenomenon of selective attention. Give your own example.

Dichotic listening task:
Subjects did best when they reported items from one ear then the other, rather than by perceived order
Limited attentional resources: Cell phones and driving!

Broadbent’s flowchart represents the flow of information through the system

Is Broadbent’s strategy of giving flowcharts a good model of explanation for cognitive science? Why or why not?

What does one look for in a successful model or theory? Testability, parsimony, and accuracy
Do flow charts meet these criteria?
Diagrams with boxes and arrows can sometimes be interpreted in several ways – they can be ambiguous
What might be better? A working computer program?

Hierarchical theories of behavior: Information flows without conscious awareness at lower levels, but with awareness at higher levels
Transformational grammar: Algorithms for manipulation of information
Broadbent’s model: What is it that flows through the model, along the arrows? Information
What is it that computers and minds have in common?
They both work on the same kind of “stuff”: Information

For discussion: How would you define "information"?

Information is the opposite of uncertainty

Note, though, that cognitive scientists are less concerned with how information is measured and transmitted.
They care more about how it is transformed (processed). I. e. – changed from one form into another.
A concern with information processing leads to an emphasis on representation: What kind of mapping exists between information as it is received and information as it is used or stored?

Marr: Vision		Stanovich: Reasoning
Computational	What purpose is served by the behavior?	Intentional	What are the person’s goals, knowledge, and beliefs?
Algorithmic	How is the purpose achieved?	Algorithmic	What cognitive processes account for these intentions?
Implementational	How is the algorithm implemented ?	Biological	What biological processes operate to make this happen?

Class Notes: Chapter 1