Class Notes: Chapter 3

Mind and Brain

Few would deny that the brain is the site of mental processes
Yet study of mind and brain have until recently been largely independent of each other

Some early discoveries: Damage to the brain leads to changes in behavior, sometimes predictable, sometimes not.

Recent technological developments: Neuroimaging (esp. PET and fMRI)

Brain Damage:A Tale of Two Neurologists

Antonio Damasio, Descartes’ Error
The relevance of emotions to intellectual functioning
Based on the classic study of Phineas Gage

V.S. Ramachandran, Phantoms in the Brain
The mind invents its own reality to explain information provided by the brain
Example: Capgras syndrome

An Old Puzzle

A run-away trolley will collide with and kill five workers.
You can divert it onto a side track, in which case it will kill only one

A run-away trolley will collide with and kill five workers.
You can stop it by pushing one very heavy person off the bridge

Brain Imaging:Solution of Old Puzzle

Greene and others
The two versions of the trolley problem engage quite different areas of the brain
This has led to a clearer understanding of judgment generally, and moral reasoning in particular

Chapter 3: Section 3.1

Explain the concept of multiple realizability. How does it figure into the argument that we can study cognition without studying the brain?

There are numerous systems that have functional properties which can be implemented in several ways
Marr: Study of implementation is separate from, and independent of, the study of functions or algorithms

Can you understand the mind without investigating the brain?

Can we fully describe the mind as a Turing machine?
Two schools of thought that may never be reconciled.
The issue pervades several of the issues we explore later.

Chapter 3: Section 3.2

The turn to the brain: Integrating two different levels of analysis in the study of vision

Functional analysis: identifying different and dissociable types of visual information-processing
Anatomical analysis: identifying different anatomical pathways within the brain

The logic of double dissociation, for any two psychological functions, A and B,

Functional analysis: two variables, X and Y
X affects A, not B (or X affects B differently)
Y affects B, not A (or Y affects A differently)

Anatomical analysis: two locations, M and N
Damage to M affects A, not B
Damage to N affects B, not A

Example: Two functions served by vision:

Where is it? What is it?

Anatomical analysis: Early evidence from brain-damaged patients:

Damage to parietal and temporal lobes produces different types of impairment
parietal = problems acting on and locating objects
temporal = problems identifying objects

Further evidence from brain-damaged patients

There is double dissociation between visual recognition and visuomotor control
Balint’s syndrome – visuomotor deficit (optic ataxia) with no recognition impairment
Visual form agnosia - impaired recognition with preserved visuomotor skills

Functional analysis: Evidence from normal subjects:

The Ebbinghaus Illusion
If asked to reach, subjects typically make an accurate-sized grasp response

But how do we combine the functional analysis and anatomical analysis?

It is dangerous to infer function of neural areas directly from what happens when they are damaged
Is the impairment due to the damaged area? Or to the fact that information fails to reach another area
How do we get a model of an information-processing pathway?

Explain Ungerleider and Mishkin’s distinction between the “what” and the “where” systems. Where is each system located?

Mishkin & Ungerleider provided experimental documentation of separate pathways for the different functions
Identifying the pathways is complicated because the corpus callosum may offer alternative routes

Dorsal –information relevant to object location
Ventral –information relevant to object identification

Ventral processing – recognize patterns. To prevent performance, three regions must be removed
Dorsal processing – object location. To prevent performance, only two regions need be removed

How does Mishkin and Ungerleider’s bottom-up approach to cognitive science differ from Marr’s top-down approach?

Marr: Begin with a determination of goals and determine the necessary algorithm. Then find out how it is implemented.
Mishkin & Ungerleider: Begin with an analysis of the implementation. Then postulate the functions.

Why do we need to take care when making inferences about cognitive function from neuropsychological evidence?

Problems interpreting brain damage results have been discussed
Results in humans are often correlational, not experimental
Problems with cross-species generalization of experimental results
There is always a need for converging evidence

Chapter 3: Section 3.3

Neural network models:

A theoretical response to renewed interest in the brain, and a partial rejection of the multiple realizability assumption

Two motivating concerns:

A need to consider coordination and timing
A need for graceful degradation

Note, though, that the result is still far from a “model of the brain”

Describe the key features of artificial neural networks

Parallel processing, not serial
Uniformity of the constituent units: nodes and connectionsconnections (associations between nodes)
Nodes arranged in layers
Training by means of modifying the weights for connections
Much more detail in chapters 8 and 9

For what type of tasks are neural networks particularly suited? Give your own example of such a task

Ideal for pattern recognition, especially involving complex rules for the patterns
Example: How do you tell a dog from a cat?
It seems so easy, but try specifying the algorithm!

Is neural network modeling a useful endeavor in cognitive science? Why or why not?

Neural networks are part of the associationist tradition that goes back to Aristotle
There are similarities with behaviorist, S-R theories: The notion that all complex behavior can be explained in terms of simple associations
A cognitive version of S-R theories?

Pictures of the Brain

Early brain recording methods: EEG
Good temporal resolution, but poor at localizing activity
“Like listening outside a factory to determine how it operates”

Recent developments:

Evoked potentials: EEG to a specific stimulus
PET: Measuring blood flow in the brain
fMRI: Measuring blood oxygenation

Using Brain Images to Test Models

In principle, PET and fMRI can identify brain regions that are active during various cognitive activities
Need to filter out background activities
Complex statistical manipulations that can sometimes produce artifacts of their own
In principle, an abstract model can lead to testable predictions about brain activity as well as predictions of behavior

What are the main differences between the neurological and the cognitive models of single word reading?

Two competing models of what happens when we read a word
Logic of the Petersen et al Study: Each function is associated with a particular brain region
Hierarchy of experimental tasks permits measurement of activity for each function

What were the conclusions of Petersen et al.’s study of single word reading?

Distinguishable activities in each brain region
No activation of auditory pathways following visual presentation and speaking task
Separate and parallel pathways for auditory processing, visual processing, and meanings (associations)

What role should neuroimaging play in cognitive science? Are there any criticisms of using it?

Pictures have a powerful effect on people
Sometimes too powerful!
Diane Beck (2010): People find brain images and neuroscience language more convincing
Greene’s work on moral judgments shows how careful research can clarify the theoretical issues

Think of neuroimaging as just another dependent variable
Compare, say, PET results with reaction time data:
Both work best if (a) we have models that make explicit predictions and (b) the predictions imply a double dissociation with some independent variable