Foundations of Cognitive Psychology

Question 1

What is the main goal of cognitive psychology?

Answer 1

To understand cognition and how the mind works using formal and testable theories

Cognitive psychology focuses on the mental processes involved in perception, memory, language, and problem-solving.

Question 2

What does the quantitative approach in cognitive psychology emphasize?

Answer 2

*Development of formal theories
*Theories that explain and predict cognitive phenomena
*Theories must be scientifically testable

This approach seeks to quantify mental processes and behaviors.

Question 3

Define a theory in the context of psychology.

Answer 3

A principle or set of principles that explain a body of facts

A good theory specifies causal relationships and allows for predictions.

Question 4

What are the key characteristics of a good theory?

Answer 4

• Aims for understanding and explanation
• Is a set of principles
• Must predict future observations/data
• Helps simplify narratives
• Must specify structures of interest
• Must include transformative functions
• Must explain how or why something happens

These characteristics ensure that theories are robust and useful.

Question 5

What is NOT considered a theory?

Answer 5

• Not a description
• Not just data
• Not just a diagram

Theoretical frameworks go beyond mere observations or representations.

Question 6

What is the science of mind in psychology?

Answer 6

Psychology is a science of the mind that requires good, testable theories of mental processes

This involves understanding how minds respond to environments and adapt.

Question 7

Describe the Lancet Fluke example.

Answer 7

A parasite that hijacks an ant’s brain to increase chances of being eaten by sheep, demonstrating a lack of true intelligence due to inability to adapt

This example illustrates cognitive insight regarding adaptability.

Question 8

What are the three levels of analysis proposed by David Marr?

Answer 8

• Computational Theory Level
• Representation and Algorithm Level
• Physical Implementation Level

Each level offers a different perspective on understanding cognitive systems.

Question 9

What does the Computational Theory Level focus on?

Answer 9

*Describes input-output functions
*What the problem the system is solving
*Why is it solving this problem
*Constraints on its solution

This level addresses the purpose and limitations of cognitive processes.

Question 10

What is the Representation and Algorithm Level concerned with?

Answer 10

• How the function is computed
• What information is represented
• How it is represented
• What algorithm is run on the information
• How input is transformed into output

This level delves into the mechanics of cognitive processes.

Question 11

What does the Physical Implementation Level describe?

Answer 11

How the system is physically implemented in the brain, including how neurons compute and learn

This level focuses on the biological basis of cognitive functions.

Question 12

Which level of analysis does Marr argue is best for understanding cognition and why?

Answer 12

Representation and Algorithm Level because it’s abstract enough to understand and test mechanisms, but concrete enough to explain the model

This level balances abstraction with practical applicability.

Question 13

What is the Rescorla Wagner Model?

Answer 13

A theory of learning through association that maps input state to output state

This model is foundational in understanding associative learning.

Question 14

In the Rescorla Wagner Model, what does Vij represent?

Answer 14

Change in associative strength between US (i) and CS (j)

This notation is crucial for understanding the dynamics of learning associations.

Question 15

In the Rescorla Wagner Model, what does λ represent?

Answer 15

Learnability of the association

This notation is crucial for understanding the dynamics of learning associations.

Question 16

In the Rescorla Wagner Model, what does ∑k(Vik) represent?

Answer 16

Total associative strength between i and all other stimuli k

Question 17

Fill in the blank: The salience of US (i) in the Rescorla Wagner Model is represented by _______.

Answer 17

α

Salience refers to the importance or relevance of a stimulus in the learning process.

Question 18

Fill in the blank: The salience of CS (j) in the Rescorla Wagner Model is represented by _______.

Answer 18

β

Salience refers to the importance or relevance of a stimulus in the learning process.

Question 19

What is the purpose of the Representation and Algorithm Level in the Rescorla Wagner Model?

Answer 19

• Specify content (what is α?)
• Specify format (how is α represented?)
• Define the algorithm (how is salience used?)

This level aims to clarify the computational mechanisms involved in learning.

Question 20

What does the Implementation Level of the Rescorla Wagner Model investigate?

Answer 20

• How neurons encode items
• How they learn associations
• How associations are represented physically in the brain

This level links theoretical constructs to biological underpinnings.

Question 21

What is Alan Newell’s critique of cognitive psychology?

Answer 21

It often tests phenomena through binary oppositions, which can generate endless questions leading to fragmented findings rather than integrated theories.

Newell’s metaphor illustrates the limitations of this approach.

Question 22

What does Newell suggest is needed in cognitive psychology?

Answer 22

Formal theories to constrain what is worth testing. Theories guide experiments and experiments shoudl falsify theories

Theories should guide experiments, which should ideally falsify theories.

Question 23

Define a theory in science.

Answer 23

A well supported explanation of some aspect of the natural world, built to be explained and predicted, tested and refined through the scientific method.

Question 24

What are the characteristics of a good theory?

Answer 24

*Prohibit certain outcomes
*Verified by genuine attempts at falsification.
*Unity- Explains many things
*Fecundity- Produces new questions

Question 25

What is a model in the context of theories?

Answer 25

A formal representation of a system, usually simplified, written in a formal language (math, logic, code).

Question 26

Define cognition.

Answer 26

A set of processes that generate intelligent behaviour, characterized as flexible, goal-directed, and adaptive.

Question 27

What key question must a good theory of cognition answer?

Answer 27

What problems the mind solves them (how it represents the world and what it does with that representation)

Question 28

Who was Alan Turing?

Answer 28

A British mathematician and key WWII codebreaker who developed the concept of a Turing machine.

Question 29

What are the four components of a Turing machine?

Answer 29

* Machine Table - program/algorithm * Machine State - current state * Tape - infinite strip divided into cells * Read/Write Head - reads tape, writes new symbol, moves left/right

Question 30

What can a universal Turing machine do?

Answer 30

*Anything that can be done by a finite number of precise steps can be done by a Turing machine *Simulate any other Turing machine and compute anything computable.

Question 31

What is the computational theory of mind?

Answer 31

*The mind is a computational system that processes information through algorithms. *Suggests that if the mind only does what is doable and what it does isn't random then a UTM could imitate what the mind does

Question 32

What does it mean if the mind is computational?

Answer 32

*It is describable through an algorithm, and all algorithms can be executed by some Turing machine *So the mind can be modeled as an algorithm on a UTM

Question 33

What are the consequences if the mind is not computational?

Answer 33

* It must be random * Cannot be understood or scientifically studied * Requires the whole brain to represent the brain

Question 34

What is partial randomness in the context of the mind?

Answer 34

Some parts of the mind may be random while others are orderly, justifying the study of computational aspects of cognition.

Question 35

What is the Turing Test?

Answer 35

A test where an interrogator must determine who is human among a machine and a human by asking questions. If the machine fools the interrogator, the machine passes the test

Question 36

What are some problems with the Turing Test?

Answer 36

* Tests performance, not competence * Can be gamed with preprogrammed answers * Depends on interrogator skill

Question 37

What was the purpose of John Searle's Chinese Room Argument?

Answer 37

It challenges the idea that passing the Turing test equates to true understanding or intelligence.

Question 38

What was the Chinese Room Arguments?

Answer 38

1) A person in a room receives Chinese characters as input and follows a rulebook with instruction (in English) to produce Chinese output 2) The person has no understanding of Chinese, just manipulating symbols 3)Yet outside observers think they are conversing with a native speaker (in reality you are just manipulating symbols, not understanding their meaning)

Question 39

What conclusion can be drawn from the Chinese Room Argument?

Answer 39

*Neither the person nor the system has real understanding, so passing the Turing test does not equate to true intelligence. *Computers just manipulate symbols not intentionally so they can't have human intelligence

Question 40

What is the nature of consciousness according to the text?

Answer 40

Consciousness cannot be proven or operationalized as it has no observable consequences (no testable hypothesis)

Question 41

What are the two main types of models in psychology?

Answer 41

Mathematical Models and Process Models ## Footnote These models serve different purposes in understanding psychological phenomena.

Question 42

What the aims of Mathematical Models?

Answer 42

* Common in personality psychology *Where personality models specify which traits interact to produce things like preferences (e.g. Rescorla Wagner) *Concerned with computational level in Marr's hierarchy ## Footnote They specify which traits interact to produce preferences or behaviors.

Question 43

What is the aim of Process Models?

Answer 43

*Common in cognitive and behavioral neuroscience *Concerned with algorithmic level in Marr's hierarchy *To describe how inputs are turned into outputs (what is represented an how it's represented) ## Footnote This includes questions about representation and manipulation of that representation.

Question 44

What are the two broad classes of Process Models?

Answer 44

Symbolic models and connectionist models ## Footnote Each class has distinct approaches to processing information.

Question 45

What do symbolic models represent?

Answer 45

Knowledge as symbolic data structures ## Footnote Symbolic models use structured rules to manipulate symbols.

Question 46

What are symbolic data strucutres composed of? What are they comparable to

Answer 46

*Composed of basic elements (variables) *Comparable to formal language or simplified programming language ## Footnote They utilize variables and operators to create propositions.

Question 47

Examples of Symbolic Language?

Answer 47

*Variables: S, T, U, V *Operators: & (and), | (or) *Rules: Any variable is a legal proposition, and any 2 legal propositions can be combined using an operator to create a new proposition *Insight: Despite having only 4 variable and 2 operators, the number of possible sentences is infinite due to repeated application (e.g. if S is a legal proposition, then S & S is also legal, etc.) ## Footnote These variables can be combined with operators to form propositions.

Question 48

What defines the processes in symbolic models?

Answer 48

Symbolic operation on data structures defined by if-then rules (E.g. If larger than (x, y) then can occlude (x, y) –> mean if x is larger than y then x can block y from view) ## Footnote These resemble computer programs.

Question 49

What are production systems? What are their 3 components?

Answer 49

*A prototypical symbolic model used in cognitive science *Components: Base set of known facts, set of inferences, executive control structure

Question 50

What does the base set of known facts represent?

Answer 50

*The current knowledge state or database *E.g. larger (elephant, lion) larger (lion, housecat) ## Footnote For example, larger (elephant, lion) and larger (lion, housecat).

Question 51

What is the role of inference rules in a production system?

Answer 51

*Rules can be fired when the left-hand side matches the facts in the database * E.g. If larger (x, y) and larger (y, z) –> larger (x, z) ## Footnote Firing adds new statements to the database.

Question 52

What is the executive control structure responsible for?

Answer 52

*Decides which rules and in what order to fire *needed fro systems with many rules/facts for efficiency (can't check every rule each time) ## Footnote It is crucial for efficiency when many facts and rules are present.

Question 53

Operation of production system

Answer 53

1) Current State: What the system knows now 2) State Space: All possible states the system could reach 3) Goal State: Desired knowledge or outcome 4) State Transition: Application of rules to move from one state to another 5) Search: Algorithm to find efficient rule applications that move toward the goal ## Footnote It represents the current knowledge of the system.

Question 54

What are the advantages of symbolic systems?

Answer 54

* Computational power * Can represent infinite propositions with a small set of components and rules * Generalization- Can represent new elements even if never encountered before ## Footnote They can define general rules and reason about new elements.

Question 55

What are some disadvantages of symbolic systems?

Answer 55

* Lack of Nuance: Can't capture all meaning (e.g. sarcasm) * Rigidity: Human behaviour is more flexible, we don't always apply every rule we know * Non Automatic Processing: Humans react faster *We learn through symbolic representations but models assume atomic elements already exist, so It's difficult to explain how we learn representations *Removing or damaging 1 element can crash the whole system *Poor Neural Plausibility- Unclear how symbolic systems wold be implements in the brain ## Footnote Symbolic systems may fail to capture complex human behaviors.

Question 56

What are connectionist models?

Answer 56

Connectionist models are computational models inspired by how the brain works

Question 57

What do connectionist models consist of?

Answer 57

Networks of nodes similar to individual neurons or populations of neurons

Question 58

How is knowledge represented in connectionist models?

Answer 58

As patterns of activation across nodes in a network

Question 59

What does processing involve?

Answer 59

Passing activation between nodes

Question 60

What determines how much activation is passed between nodes in a connectionist network?

Answer 60

Connection weight

Question 61

What is the structure of a connectionist network?

Answer 61

1) Nodes: Can be input, output or hidden nodes. They are simple processing units that have activations (how they are firing) 2) Connections: Represented by weights between nodes that can be positive (excitatory), negative (inhibitory) or zero (no influence) 3) Representation: A concept isn't a single node, it's a patter of activation across multiple feature nodes (e.g. Lion = activation of feline, wild, carnivore –> each of the descriptions are feature nodes)

Question 62

Explain processing in the network?

Answer 62

1) Activation Mechanics: Each node has an activation value and an input value (sum of activations received from connected nodes) 2) Input Calculation: aj = Activation of node j (sending node) wij = Weight of the connection from node j to node i

Question 63

Example of input calculation

Answer 63

*Activate ‘feline’ (a=1), ‘wild’ (a=1), ‘carnivore’ (a=1) *If these all connect to ‘likes–antelope’ with weight of 1 *Input to ‘like–antelope’ = (1x1) + (1x1) + (1x1) *All of this happens in parallel, not sequentially

Question 64

What is the goal when teaching the network a concept like 'lions like antelope'?

Answer 64

1) Create positive weighted connections (e.g. weight = +1) between the nodes that represent the concept/lion (features) and the output node ‘likes–antelopes 2) When the concept/lion feature pattern is activated, activation flows to a specific/likes–antelope output node 3) As a result, the network activates that specific/likes–antelope output node

Question 65

Generalization example

Answer 65

1) Network has never seen house cat in training 2) But it shares features with lion (e.g. ‘feline’) 3) As a result, activating the house cat pattern will lead to a partial activation of ‘like–antelope’) 4)Network generalizes that house cats might like antelope because it recognizes feature similarity, not symbolic rules

Question 66

What is the goal when teaching the network a concept like “octopi do not like antelope”? (negative association)

Answer 66

1) Build negative weights (w = -1) from octopus features to ‘likes–antelopes’ 2) Dashed lines represent negative connections 3) E.g. Tentacle –> likes–antelope: -1 4) If ‘wild’ is shared with lion 5) now has positive from lion, negative from octopus –> total weight = 0 6) Activating octopus features –> negative input to ‘likes–antelope’ 7) If input < 0, the activation = 0 8) The node ‘likes–antelope’ remains off

Question 67

What are the advantages of connectionist models?

Answer 67

*Flexible Representations: Representations can expresses graded similarity (e.g. cat is more like a lion than an octopus because of similar features) *Automatic Generalization: Generalization occurs due to feature overlap, no need to explicitly write the rule *Graceful Degradation: If some nodes or connections are damaged the network still functions *Parallel Processing: All node activations and inputs are calculated simultaneously, which is more brain like *Neural Plausibility: The architecture and process resemble real neural systems, making it easier to implement the model

Question 68

What are the disadvantage of connectionist models?

Answer 68

*Lack of Symbolism: No clear way to represent structured rules, so generalization is only based on similarity not logical structure *Limited Generalization to Dissimilar Cases: Children can generalize to completely new categories, connectionist networks generalize only based on similarity to known cases