The flashcards below were created by user
betweenloveandhate
on FreezingBlue Flashcards.
-
What is a network computational architecture? How does it differ from a serial computational architecture? What advantages does this provide for understanding cognition?
- Network - multiple things can be processed at the same time
- Might be mutually reinforcing
- Resembles what brain might do
- Serial - one thing happens after another
-
How do symbolic networks differ from connectionist networks?
- Symbolic networks:
- Each node represents a conceptLinks and connections are relational - network encodes propositions
- Connectionist
- Neurons that fire together wire together
- Develop connections that are either excitatory or inhibitory - connections don’t mean anything themselves
-
How are two-layer connectionist networks limited (I.e., what sorts of concepts can’t they capture)?
- Can’t learn when similar inputs don’t map onto similar outputs
- Exclusive (OR) cannot be learned
-
Explain the difference between localist and distributed representations in a connectionist network.
- Localist - one unit per concept
- Nodes are concepts
- Connections are not symbolic
- Advantage - very clear; know what’s going on
- Disadvantage - every concept needs separate unit
- Can’t generalize because you’ve learned something for that particular concept
- Distributed
- Nodes are subsymbolic
- Takes a whole collection of nodes to represent 1 concept
- Connections are not symbolic
- Representations emerge from learning and constraints
-
What are PDP networks good at? (Be able to define and/or give examples for 5 traits)
- Allows for more concepts
- Network can make predictions depending on similar inputs
- Graded output
- Can be activated by different features
- Some things can positively and negatively affect total number (to threshold)
- Auto-associators (have generalizations from features to features; reconstructive, content-addressable memory; graceful degradation; simultaneous constraint satisfaction; generalization - go beyond inputs; learning from example)
-
What are PDP networks not good at? (Be able to define and/or give examples for 5 traits)
- Differentiating different individuals
- Two birds vs. one big bird - identical twins
- Compositionality
- Barack beat Mitt. vs Mitt. beat Barack
- Crisp category boundaries
- 3 vs. 257
- Narrow generalization
- Explicit teacher
-
What is content addressable memory? Are symbolic variables content addressable? Why or Why not?
- Information is not retrieved by knowing a (content-less) address, but instead by using some of the content as a cue to retrieve the remainder of the information
- Symbols are not content addressable - in order to activate a concept, you have to already know features of the concept
-
What is dead reckoning? Why does Gallistel argue it requires symbolic representations? What evidence exists that ants and bees can dead reckon?
- Dead reckoning - no features on environment, figure out how you’ve moved and accelerated and figure out where you are relative to starting location
- Tunisian ant:
- Leaves nest to go on foraging journey; finds food and takes it back to nest
- Takes a reading of sun and figure out how far they’ve gone from the reading
- Ant has to store 2 variables: distance and angle from nest
- Requires cognitive mapping (symbol-processing)
-
Why and how do bees dance?
- Dance to communicate where they are and where they’ve been
- Dance:
- Circle (waggle dance when you make figure 8 short enough that it becomes a circle)
- Waggle - angle tells what direction according to position of sun
- Duration of dance and number of vibrations = distance
- Communicate distance, direction, and quality of a resource
- Seem to be representing certain things symbolically
- Can also remember angles and distances between multiple sources of food
-
What is a cognitive map? What evidence is there that insects and humans use a cognitive map to navigate?
- Cognitive map - a mental encoding of the relative spatial positions of goals and landmarks
- Bees keep track of allocentric angles (between locations they are not at currently) when planning foraging trips
-
What type of information is most influential when a human reorients? What's the evidence?
Humans can combine landmark info with geometry
-
Be able to understand the results and implications of Wang & Simons (1998).
- Had people look at circular table with objects on it in particular arrangement
- People closed their eyes and moved or stayed in same place while turning table
- Moved one object
- Given new viewpoint, did an object move and where?
- People did better when the table position was the same
-
True or false: our ability to use maps to navigate is a cultural invention?
- False
- Munduruku - can still use maps (at American children level)
- Maps = intrinsic
-
What is subitizing? What creatures do it? What's the evidence? What would happen if a person/animal were unable to subitize?
- Subitizing - counting small groups of things very rapidly (breaks down at about 4 items)
- Animals and humans
- Children and box
- Crackers hidden in box; child reaches in box and grabs one at a time
- Will reach into the box the same number of times they see crackers going into a box
- Keeping track of number of objects
- Tamarin monkeys
- Hide eggplants in box and they watch
- Look at looking times of monkeys and see if they are surprised by number of eggplants hidden in box after reveal
- If unable to subitize, have to count even small numbers of things
- Estimating abilities are intact
-
What is estimating? What creatures do it? What's the evidence? What would happen if a person/animal were unable to estimate?
- Estimation - ability to recognize amounts of things roughly when they exceed 3-4
- Present in animals, children, and adults
- Macaque monkey brains
- Fixate on dot; see array of dots, see nothing, given one of several possibilities
- Either sees complete match or not
- Has to say whether it's a match or not
-
How does language affect numerical cognition? What's the evidence?
- Subitizing - distinct from language or math
- Counting - requires language
- Frank et al 2008; 2012
- English speaking people were fine unless they had to do a verbal distraction task
-
What are naturally occurring animal communication systems like? How do they differ from human language?
- Communicate thru displays and calls
- Usually finite in number
- Each display has a holistic meaning, can't be recombined to mean something new
- Humans an re-assign sounds to different meanings
- Can't combine sounds like we can
- Limited, finite vocabulary
- Aren't learned; typically innate
- Human languages are learned to a certain extent
-
To what extent can animals acquire human communication systems? How are they limited?
- Nim Chimpsky:
- Able to learn words and use them appropriately
- Had trouble recombining words
- Seemed like he was communicating thoughts, but when comparing Nim to children, number of terms he used were very constant
- Something that is particularly human is necessary in learning human language
-
Define and characterize the advantages of arbitrariness and discreteness.
- Arbitrary:
- Actual signs in language are arbitrarily associated with the thing
- The sounds themselves are randomly mapped to some meaning
- Opposite of iconicity
- Makes referring and abstract concepts possible
- Discreteness:
- Made up of whole bits as opposed to continuous bits
- Need to add whole words up in order to make up sentence; need whole sound in order to make up word
- Allows us to combine units in predictable ways which can give ways to novel meanings
- Allows you to break out of combinatorial system
-
What are the basic levels of linguistic representation? What type of rules govern each level of representation?
- Phonology - rules for sound combinations
- Formation rules - what are possible sound sequences in a language?
- Adjustment rules - how sounds are altered in context
- Morphology - how words are formed
- Applying things like past tense to made-up words
- Syntax
- Rules determine grammatical acceptability
- Two main aspects:
- Word order
- Phrase - group of words that act together as a grammatical unit
-
What's the difference between descriptive and prescriptive rules? Provide examples. "Fantast-fucking-ic"
- Descriptive rules - describe how you use language
- More that what grammar teachers can teach but less than you can explain
- We know where to put things even though we weren't explicitly taught the words (fantast-fucking-ic)
- Prescriptive rules - prescribe what you ought to do (who vs. whom)
-
Given a sentence, what is the subject and what is the predicate?
- Subject = noun phrase
- Predicate = verb phrase
-
Provide evidence that sentences have internal structure above and beyond being ordered strings of words
Structural ambiguity - the only thing that differs is their phrase structure
-
What evidence is there that sign languages operate like spoken languages?
- Structural ambiguity - the only thing that differs is their phrase structure
- Signs = arbitrary
- Different signs depending on origin of sign language
- Distinctive features: configurations
- Distinctive features: location
-
What is induction?
- Induction - figuring out a rule based on a finite set of examples
- Trying to generate rules for things we've never seen before
- Meno's paradox
-
What kinds of evidence are theoretically available for learning novel concepts? In what situations is each kind of information most useful? Which is more important for language acquisition?
- Positive evidence
- Evidence that an utterance is well-formed
- Over 99% of the input is grammatical
- Negative evidence
- Evidence that an utterance is not grammatical
-
Why do children babble?
Experimenting with the mouth and sound formation
-
How does the learning trajectory of language production and comprehension abilities of children differ?
- Children know much more than they say
- Bottlenecks in production - children have trouble pronouncing certain strings of phonemes
- Recognize only the sound distinctions that are important to their native language by 1 year
- Children understand words as early as 9 months
- Some knowledge of word order by 18 months
-
How does the child's knowledge of language sounds develop? What's the evidence?
- Initial state - prepared for all the languages of the world
- Organized around an initial set of pre-linguistic categories
- Extremely plastic, responsive to environmental language
-
How do words and syntax develop?
- At birth, infants can discriminate:
- Mother's voice from another woman
- Dr. Seuss story read during pregnancy from another story
- Native language relative to another language only if they differ in rhythmic properties
- Syntactic Comprehension - "Look, Big Bird is tickling Cookie Monster"
-
What learning takes place in the womb? How do we know?
- Sound travels through tissue and fluid
- Particularly low frequencies
- Babies identify certain aspects of their native language even immediately after birth
-
What evidence suggests that children are born with an innate capacity for language?
- Born recognizing the distinctions that are important in any language in the world
- Lose ability to distinguish sounds that are not important to their native language
- Shifting of sound distinction, not addition
-
Describe 3 functions that concepts serve for cognition. Illustrate each with an example.
- Classification: ability to take two things and say that they are the same type
- Understanding/explanation: why a thing is as it is
- Prediction: we know how it's going to behave in the future
- Reasoning: can use knowledge of category to draw inference on this category
- Communication: allows us to convey knowledge that allows someone else to do all these things without them having to learn it themselves
-
Give an example of how mental concepts might contribute to stereotyping.
- People underestimate differences within a group and enhance differences between groups
- The presence of a category is:
- Driving apart - people make bigger disparities between categories
- Driving together units in one category
- Group effects
- Negative traits assigned to other group, positive traits assigned to their group
- Ingroup/outgroup effect
-
What is a basic level concept? What's the evidence for it? Why is there a basic level organization?
- First-learned
- Natural level for naming
- Cross-cultural consistency - first words are usually basic level
- Share both parts and shape
- Biggest grouping for which there are lots of shared features
-
What is the classical theory of concepts? Name 3 problems with this view. Illustrate with examples.
- Bundles of necessary and sufficient features
- Triangle - enclosed polygon with 3 straight sides
- Bachelor - unmarried, adult, human, male
- Bird - has wings, lays eggs
- All or none
- Problems:
- Hard to come up with proper definitions for categories (not inclusive enough)
- Definitions not exclusive enough either
- Which properties are really atomic?
- Tough to come up with a set of features for many categories
- Most categories are not all or none
-
What characterizes probabilistic theories? Describe two pieces of experimental evidence that support them.
- Categories organized around typical, not defining, properties, or features
- Category membership is graded
- Some members are better than others
- Category boundaries are fuzzy
-
What evidence is there for prototype organization of concepts?
- Family resemblance
- Ideal is a prototype
- Input is matched against prototype
- "Central tendency" extracted from multiple experiences
- Prediction - equal or better performance for ideal than for actually experienced examples
- Subjects learned to classify dot patterns formed by distorting prototypes
-
What are Armstrong et al.'s main arguments against probabilistic theories?
- We can reason absolutely with all or none categories, even if they act graded sometimes
- Some mentally stored identification function used to sort things quickly, and also a mentally categorial description of the category that determines membership in it
-
Describe Armstrong's dual position theory. What is the Theory Based Theory of concepts? How does it work differently for natural kinds and artifacts? Give examples. Be able to recognize examples of different theories in action and to explain them.
- Dual aspect of concepts
- Core description - definitional = compositionality (how wholes get meanings from their parts)
- Identification procedure - probabilistic
-
Be able to recognize examples of different theories in actions and to explain them.
- Prototype models
- What you store in memory is central tendency
- Ideal exemplar of tendency (bird -> robin, not penguin)
- Exemplar models
- Think of all the examples and store them simultaneously
- Any additional token is compared to all your experiences
- Average relationship with any token
-
What is induction? Why is it a puzzle for cognition?
- Drawing a general conclusion from specific facts or observations
- No answer is sure to be right, only more probable
-
What is the difference between a normative and a descriptive theory? What is the normative theory of induction?
- Normative accounts - how we ought to reason
- Past experience
- Most rational/optimal
- Choose outcome that gives you the right probability of success
- Descriptive accounts - what we actually do
-
What's a heuristic? Why do humans reason using heuristic strategies?
- Rule of thumb or strategy for problem solving
- Faster; gives you the right answer most of the time
-
Confirmation bias
Tendency for people to favor information that confirms their preconceptions or hypotheses regardless of whether the information is true
-
Availability
The easier it is to come up with an example of something, the higher likelihood you're going to give it
-
Representativeness
Over-reliance on small numbers, regression to the mean (base rate neglect, conjunction fallacy, gambler's fallacy)
-
Law of small numbers
Judgmental bias which occurs when it is assumed that the characteristics of a sample population can be estimated from a small number of observations or data points
-
Base rate neglect
If presented with related base rate information and specific information, the mind tends to ignore the former and focus on the latter
-
What the two kinds of information are critical in reasoning about the likelihood of an event given some evidence? Which do people tend to ignore in their judgments?
- Diagnostic information - likelihood of the data given a hypothesis
- Base rate
- People don’t consider the base rate - prior likelihood of the hypothesis
-
Describe two reasons why heuristical reasoning is typically sufficient for our purposes.
- Faster
- Gives right answer most of the time
-
Describe one way to get people to pay attention to base rates. Illustrate with an example.
- If chance is highlighted, performance improves
- If 100 pieces of paper with thumbnail sketches, 70 lawyers and 30 engineers
-
Provide two examples of the influence of unconscious reasoning on conscious decisions.
- HM and the tower of Hanoi
- Moving the disks is part of unconscious memory
- HM kept on getting better at it even though he doesn’t remember seeing the same
- Korsakoff’s and learned aversion
- Vitamin D deficiency that leads to problems similar to HM’s
- Could still learn implicit things
-
What is the “products vs. processes” theory of the conscious/unconscious distinction? What evidence militates for and against this hypothesis?
- What’s your mother’s name?
- Remember and retrieve
- Aware of product, not process
- Theeuwes, Kramer, Hahn, Irwin
- Had people look at array of objects on screen that initially started grey w/ an s in them
- At some point, a circle with red appear
- New circle also appeared in the array that was also red
- People say they didn’t see addition of circle; eye movements tell another story
- Just as many fixation to original red circle and the new red circle
- Problem with products vs processes theory
-
What are the differences between an automatic and a controlled process? How can a process become automatic?
- Automatic:
- Things that you don’t need to consciously allocate attention to
- No intention
- Highly efficient
- Controlled:
- Things you don’t have to do - naming a color
- Happens with intention
- More flexible
- Learning to make something automatic
- Trained people to read an essay and later answer questions about the essay
- At the same time they were reading, were taking dictation
- Through headphones, the person is hearing a list of words and is asked to write them down
- Initially, this task is impossible
- After months of practice, they get so rehearsed at taking dictation after reading that when they took a test, they did just as well as people without the audio
- Dictation task - people were unaware of it; happened below level of conscious awareness
-
What are the advantages and disadvantages of the conscious/unconscious distinction?
- Unconsciousness - advantages
- Efficiency - automatic things are shunted off to automatic system
- Other things need to be flexible; information that’s useful at one level might not be useful at another level
- Disadvantages
- Errors of introspection
- Causal misattributions
- Blindsight
- Errors of automaticity
- Stroop
-
What is Dehaene’s neuronal workspace hypothesis of conscious awareness? What are the three ways that a stimulus might be processed from this perspective? What are the neuronal and cognitive consequences of each?
- 3-way distinction in terms of phenomena and awareness: subliminal, preconscious, and conscious
- 2 things are necessary for you to be conscious of something:
- Bottom-up stimulus
- Subliminal - if weak or interrupted, there isn’t much activity
- Top-down attention
- Subliminal - if there’s no top-down, not going to get any real effect that is reportable at all
- Subliminal - short-lived priming
- Can have facilitation of processing of a familiar stimulus
- Doesn’t really seem to change things fundamentally
- Low-level inputs bottom-up
- Preconscious
- Sufficiently strong to reach conscious
- Top-down controlled-attention towards stimulus
- Might not be directing top-down attention towards stimulus activity; don’t get reportability
- Card trick/gorilla - could notice if you were orienting towards them, but you’re not
-
What are the easy problems of consciousness according to Chalmers?
- Some kinds of information in the brain are accessible - perceived objects: actions, contents of sentences
- Others are not - transforming retinal image, muscle movements, rules of syntax
-
What are the hard problems of consciousness?
- Can’t reduce consciousness to particular number of neurons
- Inherently subjective; can’t study it in the same way as we normally do
- Principle of structural coherence
- Things you are aware of are the only things you can have phenomenal awareness of - have qualia associated with them
- Same fxnal organization - same kind of conscious experiences
- Some information has 2 aspects - physical system, phenomenal aspect that is subjectively reportable
-
Behaviorism (sentience)
- Sentience makes no difference to anything; therefore it is an illusion
- The fact that you are aware of something = illusion
- Problem 1: Descartes’ disproof
- The only thing we can be sure of is our thoughts
- There’s still a you, and the only way you know that is because of your thoughts
- Problem 2: why shouldn’t we hurt people?
- Pain = byproduct of something; not real
-
Computation (sentience)
- Information/informational processing
- Problem 1: How?
- Problem 2: Panpsychism
- Everything you come across has some informational properties in some way; do they all have some degree of subjective awareness?
-
Neural Reductionism (sentience)
- The brain secretes consciousness
- Only possible with organic material
- Problem 1: How?
- Problem 2: Rules out sentient robots and aliens that aren’t made of same matter as us before we’ve even seen them
|
|