1
Q
what is cognitive science?
A
study of mind and intelligence
2
Q
what is cognition?
A
mental processes for acquiring and using information
3
Q
why is cognitive science scientific?
A
provides formal (mathematical) theories and tests them with empirical (often experimental) evidence
4
Q
why is cognitive science interdisciplinary?
A
cuts across the biological, social, and computational sciences
5
Q
which philosophers provided early foundations for psychology?
A
- plato: innate knowledge
- aristotle: laws of rational inference
- decartes: mind-body problem
- hume: problem of induction
6
Q
when was psychology born as a scientific discipline?
A
mid 1800s
7
Q
which early methods did psychology include?
A
- introspection
- psychophysics
- comparative animal behaviour
- early psychometrics from survey data
8
Q
what did behaviourists reject?
A
most earlier methods, particularly introspection
9
Q
what were behaviourists principle?
A
any reference to unobservables is unscientific; only allowable data are observed behaviour
10
Q
what is classical conditioning?
A
UCS –> R
UCS paired with CS
now CS –> R
11
Q
what is operant conditioning?
A
rewarding/punishing a behaviour will increase/decrease that behaviour respectively
12
Q
what are the 3 problems of behaviourism?
A
- tolman’s ‘cognitive maps’ experiment with rats: group 1 (food always in the same place) learned faster, showing accurate mental maps - not just behaviour sequences
- miller (1956) showed capacity limits: common limit of ~7 across processing and memory; reinforcement doesn’t help much but chunking does
- chomsky (1959) argued against skinner’s view of language: language is structured (rules on categories), recursive (embedding), and productive (infinite sentences); reward/punishment can’t explain acquisition
chunking: C IAC BCF BIY YZD VD vs CIA CBC FBI YYZ DVD
skinner thought that language acquisition could be explained through operant conditioning
13
Q
what is the birthday of cognitive science?
A
sept 11, 1956 (conference of computer scientists, linguists, psychologists)
14
Q
what are the central characteristics of the cognitive revolution?
A
- computational theory of mind as paradigm (mental representations (info stored in our minds) and mental processes (how that info is stored))
- interdisciplinary methods
- fascination with classical philosophical questions
15
Q
what are tinbergen’s four questions?
A
- function: what is its adaptive value?
- ontogeny: how does it develop over the life of an individual?
- mechanism: how does it work?
- phylogeny: what sequence of evolutionary changes led to it?
16
Q
what are marr’s levels of analysis?
A
- computational: what is the goal of the computation (problem being solved)?
- implementational: what is the physical substrate that implements this algorithm?
- algorithmic: what cog mechs (reps and processes) carry out this computation?
17
Q
what are the 3 things involved in mental activity in computation?
A
- representation: a symbol that stands for something else
- process: a procedure that converts one rep into another
- computation: info-processing that uses cog processes to convert one representation into another
18
Q
in humans, how are computations implemented?
A
- in our nervous systems
- but they don’t have to be - transistors in computers maybe different biology in e.t. aliens
19
Q
what do our brains and the best AI systems have in common/uncommon?
A
- use neural networks (biological or stimulated)
- this doesn’t mean that the underlying representation are similar
20
Q
what puzzles remain for computation?
A
- emotion
- culture
- embodiment
- language relativity (are the thoughts wer think determined by the language we speak)
- consciousness
21
Q
how does a species become intelligent?
A
environmental survival pressures select individuals with adaptive problem-solving; possibly via ‘mental organs’ (modularity hypothesis)
22
Q
how does an individual become intelligent?
A
combination of innate knowledge and learning algorithms (e.g., language)
23
Q
how does a society become intelligent?
A
through cultural evolution (selection on ideas)
24
Q
what is bound rationality?
A
rationality subject to limitations in info, time, and processing capacity
we must be in some sense rational – our
computations correctly solve survival-
relevant problems
25
which four frameworks illuminate rationality?
- logic
- probability theory
- utility theory
- game theory
26
what is a representaiton of X?
something that stands for or refers to X
27
how can representations be linked?
- structure (map)
- resemblance (picture)
- convention (word)
28
what kinds of things can be represented?
- concrete (objects)
- abstract (numbers)
- even fictional
29
what is the lesson of the plato's cave?
we do not have direct access to reality; we must construct it
30
what does decartes' wax example show?
our representation of wax goes beyond its sensory properties
31
what is the physical symbol systems hypothesis?
all entities capable of intelligent action are PSSs
32
what does a physical symbol systems contain?
- symbols in the form of physical patterns
- represenations comprising structured combinations of symbols
- processes (rules) for manipulating symbols which are themselves represented in the systems
33
what is a symbol?
a physical token that can stand for something (letters, pixels, transistor states)
34
what allows minds to exist physically and be connected to the world?
symbols in the form of physical patterns
35
why are representations often recursive?
they can embed one representation within another, allowing arbitrarily complex states
36
why must rules be represented into the system?
to allow the system to derive new representations that track the represented world (this is computation)
37
why can a physical symbol system produce intelligent action?
- if X represents A, then with the right processes, the system behaves as if it had access to A
- so the 'right' PSS can draw useful inferences for navigating the world
38
what does the language of thought hypothesis propose?
thought uses: structured, language-like mental representations ('mentalese'), and logic-like rules for combining/manipulating them
39
what is a proposition?
a claim about the world that can be true or false
40
what are predicates and variables in the language of thought hypothesis?
- predicates represent properties/relations (F(x), B(x,y))
- variables represent individuals (w = Wilbur)
## Footnote
F(x) = 'x can fly' (attribute)
B(x,y) = 'x is buddies with y' (relation)
F(w) = 'Wilbur can fly'
B(w,y) = 'wilbur is friends with y'
41
how can propositions be composed into more complex ones?
using logical operators
-P (not P)
P&Q (both need to be true, else false)
P|Q (holds if either or both is/are true)
P->Q (if P then Q, true if anytime P is true, Q is true)
41
what are inference rules?
syntactic symbol manipulations (proceeds independently of what the symbol stands for)
modus ponens/tollens and de morgan's law
42
give an example of modus ponens
from P and P->Q, we can derive Q
43
give an example of modus tollens
from -Q and P-> Q, we derive -P
44
why do we think that the language of thought hypothesis is true?
- productivity of thought: we can think infinitely many thoughts from a finite number of elements
- systematicity of representation: there are relationships among the different thoughts we can entertain
- compositionality of representations: the meaning of a representation is determined by the meaning of its parts (jerry is a brown cow = jerry is brown, jerry is a cow)
- coherence of inference: there are relationships among the different inference patterns we can use
45
what are the three questions that cog scientists ask about mental representations?
- format: organization, e.g., images v. propositions
- content: what info is stored, e.g., babies' intuitive physics
- processes: principles of inference, e.g., statistics v deduction
46
what factors can differ between solutions to the same problem?
- memory demands
- time efficiency
- error-proneness
47
what is semantic memory?
long-term memory for facts
48
what two possible organization exist for semantic memory?
- exhaustive model
- sparse model
49
what prediction does the sparse model make in semantic memory?
verification times should be longer for proporties stored at more distant nodes
50
what prediction does the exhaustive model make in semantic memory?
verification times should be independent of other nodes
51
what questions do mental imagery studies ask?
are mental images like perceptual images, or something esle entirely?
52
what is a depictive representation?
a picture-like format that specifies location/values in space
53
what is a propositional representation?
a description in a language-like format, e.g., ON(BALL,BOX)
54
what did shepard & metzler's (1971) mental rotation experiments show?
time required is a linear function of the angle of displacement (the more angles of rotation a mental image has, the longer it'll take to re-rotate it)
55
what are critiques of depictive imagery?
- tacit knowledge: people know that, in the real world, longer rotations take more time; maybe they are just matching this knowledge
- epiphenomonalism: we experience pictures in our minds - no one doubts this - but are they the underlying reps in which computations are carried out?
- sparsity of mental images: when you inspect a mental image in detail, you'll find it is remarkably sparse
56
why is causal knowledge important?
predicting consequences, explaining events, assigning blame
57
what are two ways to represent causual knowledge?
one big network vs. many small 'causal island'
A->B->C versus->B & B->C
58
what did johnson & ahn (2015) find?
people falsely remember transitive causal links in network, but not in islands
59
what is dualism?
the view that mental and physical things are fundamentally distinct
60
what was decartes' argument for dualism?
you can imagine existing without a body, but not without a mind - they seem like different things
61
what are philosophical zombies?
- beings physically identical to us (same brains, same behaviours) but not conscious
- so we have something that zombies lack despite being physically identical
62
what is mary's room thought experiment?
may knows all the science of colour but lives in black-and-white; when she sees colour, she learns something new
63
what is the problem for dualism?
| what question does it seek to answer?
how do minds and bodies interact?
64
what did phinease gage's cage show?
damage to the front lobe can drastically alter personality and behaviour
65
what did split-brain studies show (gazzinga et al., 1962)?
cutting the corpus callosum revelase specialized functions in each hemisphere
66
what did shen et al. (2019) demonstrate with deep image reconstruction?
brain activity patterns can be used to reconstruct images people are viewing
67
what is materialism?
- only physical things exist - the mental *is* the physical
- explains why our mental states correspond to brain states
68
what does materialism struggle to answer?
struggles to explain subjective experience (consciousness)
69
what is functionalism?
a mental state is defined by its function/role in the cognitive system, not by its physical makeup
70
what do most modern cognitive scientists believe?
they are functionalists; the mind is “software” that could, in principle, be implemented in different physical mediums
71
what is the mind-body problem?
mind-body dualism has been a historically influential view but is now rejected by scientists/philosophers
72
what is the nervous system?
an electrochemical signaling system that communicates information throughout the body
73
what are the nervous system's main cell types?
- neurons: signaling
- glial cells: support
74
what are the main divisions of the nervous system?
- peripheral: somatic (skeletal muscles + skins) vs. autonomic (internal organs) & afferent (sensory) vs. efferent (motor)
- central nervous system: brain and spinal cord
75
what are neurons?
the main computational units of the nervous system?
76
how many neurons does the human brain have?
80-90 billion
77
how do neurons process input?
- dendrites receive input
- if enough total input, an action potential is generated (all-or-none)
- a chemical signal is sent across the synapse to other neurons
- electric to chemical
78
how do neurons vary in signaling?
by firing rate, not graded potentials
79
what is a synapse?
the gap where neurons send chemical signals to other neurons
80
how does the brain represent and compute information?
through the pattern of trillions of connections among neurons
81
what is the hierarchical organization of the brain?
neurons --> regions (groups of neurons) --> networks (groups of regions)
82
what are the main gross anatomy structures of the brain?
- brain stem: basic life functions
- cerebellum: motor coordination
- thalamus: relay center
- hypothalamus: homeostasis
- cerebrum --> cerebral cortex (sensation, control), limbic system (memory, emotion), basal ganglia (input integration)
83
what did brodmann identify?
52 areas based on cellular structure (layers, density, connections), many mapping to functional specializations
84
what are the functional specializations of the lobes?
- frontal: motor cortex (commans), prefrontal cortex (planning)
- parietal: somatosensory cortex (touch)
- occipital: visual cortex (vision)
- temporal: auditory cortex (hearing, language)
85
how does fMRI work?
measures changes in oxygenated blood
86
how does PET work?
measures changes in blood glucose
87
why is fMRI used more often than a PET scan?
better spatial resolution; no radioactive dye
88
why is subtraction used in imaging studies?
- because every region is always somewhat active; subtraction isolated activity due to the process of interest
- we compare 2 conditions that differ in only the cognitive process we are interested in
- the difference in activation between regions represents the additional neural activity caused by that activity
89
what did greene et al. (2001) find in moral dilemmas?
different brain regions activity in moral-personal vs moral-impersonal dilemmas (emotion vs reasoning)
## Footnote
pulling a railway switch vs actually pushing a person
90
what did kosslyn et al. (1993) show about imagery?
mental imagery activates primary visual cortex (V1) retinotopically (large vs small imagined 'A' recruits different areas)
## Footnote
neurons process information in a way that maps onto the spatial structure of the retina
'a' -> moer posterior activation (fovea)
'A' -> more anterior activation (periphery)
91
what is reverse inference?
inferring cognitive processes from brain activation, based on prior knowledge of regional functions
## Footnote
if we already know that a brain region does process X, then if that region is active when participants are doing task Y, we infer that
task Y involves process X
92
what is a morality example for neural correlations?
- angular gyrus = emotion & middle frontal gyrus = working memory
- t/f, when people think about 'personal' dilemmas, they rely more on emotion and less on memory
93
what is an imagery example for neural correlations?
- anterior regions of primary visual cortex (V1) = peripheral vision (i.e., V1 = retinotopically organized)
- t/f, when those regions are activated more just by imagining a large object, it implies that the 'mind's eye' is also retinotopically organized
94
what are the three problems with neural correlates?
- causation: are neural correlates causally linked to processing? imaging techniques are indirect (e.g., fMRI relies on assumptions about the
relationship between blood oxygen levels and cognitive processing). we don't know what parts of the activated regions are causally responsible
vs downstream consequences
- scale: how do individual neurons implement this processing? (imaging techniquestion have very coarse scales - a voxel in fMRI contains almost 1,000,000 neurons)
- mechanism: how does it work? if we knew what every neuron was doing, we wouldn’t understand why this
pattern implements the mental algorithms underlying cognition
95
what are the two main methods for studying causation?
- neuropsychology: examine deficits associated with particular patterns of brain injury (e.g., stroke or surgery)
- transcranial magnetic stimulation: interfering with signaling
96
give examples of neuropsychological evidence
- phineas gage, split brain patients
- 'face-blindness' , prosopagnosia
- cortical colour blindness
97
what is single-cell recordings?
measuring responses of individual neurons to stimuli
98
what did hubel & wiesal (1959) discover?
- cells in primry visual cortex (V1) sensitive to particular directions of motion in specific retinal regions
- e.g., cells that are sensitive to one particular direction of motion in one region of the retina
99
how do idealized neurons work in models?
they add up inputs to each neurons (mutliplied by weights) into a sum X and fire (S=1) if X passes a threshold T
100
what did mcculloch & pitts (1943) discover about simple networks?
they can implement logical operations like AND, OR, NOT
101
how does a simple network implement "NOT"?
when input I₁ = 0, neuron fires (S=1); when I₁ = 1, neuron does not fire (S=0). Therefore, S = -P
102
what limits single-layer networks?
they cannot compute all functions
103
how do hidden units solve the limitations of single-layer networks?
multi-layer networks with hiddent units can compute a vast range of functions, e.g., XOR (exclusive OR, P or Q but not both)
104
how is knowledge represented in neural networks?
it is distributed; information lied in the pattern of weights between units
105
what is graceful degradation?
- performance decreases gradually (not catastrophocally) when units are removed
- the system is damage resistent
106
how is memory accessed in neural networks?
by content (cued by related info), not by address
107
what two main types of learning exist?
- supervised learning: external teacher feedback
- unsupervised learning: self-organization
108
what is hebbian learning?
'neurons that fire together wire together' - neurally plausible associative learning
109
what is the delta learning rule?
adjusts weights up if the model overshot, down if undershot
110
what is backpropogation?
an algorithm for training multi-layer networks by propagating error backwards through hidden layers
111
what are convolutional neural networks (CNNs)?
networks that preserve input structure, inspired by retinotopy in the primary visual cortex (V1); used for vision tasks
112
what are recurretn neural networks (RNNs)?
networks with feedback connections that provide simple memory (e.g., sentence context)
113
what are autoencoders?
models trained to reconstruct inputs after compressing them through a bottleneck of hidden layers
114
what is spaun (eliasmith et al., 2012)?
2.5 million node model of the human brain, able to perform diverse tasks (recognition, counting, induction)
115
what are LLMs trained to do?
predict the next word
116
how are LLMs similar to smaller networks?
neurons sum inputs; trained with backpropagation
117
how are LLMs different from smaller networks?
they are bigger (billions of neurons), deeper (many layers), trained on massive text corpora, with embeddings and transformers, plus freinforcement fine-tuning
118
what is tokenization?
breaking text into tozens (unit of inputs)
119
what is the role of embeddings?
represent words as vectors in high-dimensional space, with similar meanings --> similar values
120
what is a transformer?
architecture that uses attention heads to weight tokens across a context window, processing meaning and context
121
why don't LLMs always give the same answer?
the 'temperature' setting allows lower-probability words to be chosen
122
how are LLMs trained?
predict tokens, backpropagate errors, repeat billions of times; then fine-tune with reinforcement learning from human feedback (RLHF)
123
why are LLMs 'black boxes'?
they are trained for correct outputs, but we don’t know how they internally achieve them
124
what do LLMs tell us about the mind? the 2 possibilities
- revolutionize our understanding of the mind - parallels between network architecture/training and evolution/development
- they don't tell us much - humans don't require billions of words to learn, so LLMs may work differently
125
what is artifical general intelligence (AGI)?
the ability to perform any cognitive task at a human-like level
126
what is superintelligent AI?
AI that can outperformance humans at any task
127
what is AI alignment?
the challenge of specifying human goals precisely enough for AI to reliably pursue them
128
what is bostrom's 'paperclip maximizer'?
- a thought experiment about misaligned goals leading to catastrophic consequences
- a machine that wants to make paperclips will do anything (including destroy the world for metal) to make them
129
what is the optimistic view of AI?
like past tech revolutions, AI may massively improve living standards and enable breakthrouhgs (longevity, renewable energy, space)
130
what is the skeptical view of AI?
progress has been overestimated for decades; intelligence may not scale with more neurons/data; concepts like “superintelligence” may not even make sense
131
what is thought made of?
- words are words in the thinker's natural language (why is it difficult to put a thought into words or say something different from what we meant)
- words are images (doesn't tell you which aspects are important)
- words are concepts
132
what is the difference between a concept and a category?
- many concepts are categories (groups of individuals)
- plato thought the real objects are imperfect copies of an ideal form of each concept
- we do not remember/talk about each object or event as unique but as an instance of a class or concept
- many concepts are not categories: abstract thoughts or propoerties/relations
133
what is the main idea of categories as definitions?
a set of necessary (everything in this category has each of these properties) and sufficient (if something satisifies all of these properties, it must be a category member) criteria for being in a category
134
what is the classical theory of categories?
1. categories are mentally represented as definitions
2. something is or isn't a category member
3. all members in the category are equally a member
135
why is categories as definitons an attractive idea?
- clear procedure for categorization and inference
- makes it easy for us to agree on what we are talking about
- not limited to concrete things
136
what are the 3 problems with categories as definitions?
- most categories don't have definitions (GAME?)
- what about borderline cases (tomato is a vegetable/fruit)
- some things seem to be 'better' examples of a category (robins are more birds than penguins)
137
what is the main idea of categories as prototypes?
- represented as a summary representation of category features (which vary based on their importance)
- more typical category members have the most important features and more of them
- to categorize a new item: we must compare it's similarity to the feature list (placing more weight on the important features)
138
what and how are weights given to certain features in the categories as prototype theory?
1.0 --> necessary features
combination of the probability and salience of a feature (how likely it is to come to mind)
139
what is the prototype theory of categories, and how does it address problems with the classical theory?
says categories are organized around best examples (prototypes), not strict definitions.
1. no necessary & sufficient conditions: features are probabilistic, not definitional
2. borderline cases: some members are closer to the prototype than others
3. typicality effects: even clear members vary in how many important features they share with the prototype
140
what are basic level categories in the prototype theory?
- tend to be the default level in communcation
- children learn them earlier than lower or higher levels
- balance imfortativeness against cognitive economy (higehr levels not be specific enough and lower levels take more effort to communicate/discriminate without gaining new info)
| dog (basic), golden retriever (subordinate), mammal (superordinate)
141
what 2 problems arose as a response to the basic level categories (prototype)?
1. we seem to represent some information in our categories that goes beyond feature lists (correlated (most small birds also sing) and variability)
2. categories could be highly disjunctive: if a summary representation has 6 features, two items could each match
on 3 features and both be seen as category members (unstructured)
142
what is the main idea in the categories as exemplar theory?
- there is no overall feature list for any category, just exemplars retrieved from memory
- each exemplar of a category has its own representation in memory
- category rep is a set of those reps
- to categorize a new item: we retrieve the most similar exemplars to the new item. once we retrieve enough exemplars from the same category, we put the new item into that category
143
how does the exemplar theory of categories address problems with the classical theory? what does it leave out?
1. no necessary & sufficient conditions: no need for shared features across all members
2. borderline cases: borderline items retrieve similar numbers of exemplars from different categories
3. typicality effects: typical exemplars are more numerous and similar, so they’re easier to retrieve
limitation: like prototypes, it leaves out important aspects of category representation and allows arbitrary/disjunctive categories.
144
why do researchers use artificial categpries when testing prototype vs exemplar theories?
- because prototype and exemplar theories make similar predictions for real categories
- artificial categories help tease them apart
145
how do people categorize the prototype in experiments?
often categorized easily even if it was never seen before
146
how do exemplar models explain why prototypes are easily categorized?
prototypes are usually similar to many stored exemplars, making them easy to recognize
147
which model usually performs better in experiments, prototype or exemplar? do people rely only on prototypes or only on exemplars?
- exemplar models often do better, especially with unusual or “weird” category structures
- both theories seem to work at times
- some studies show participants shifting strategies during experiments, while others consistently favor one approach
148
what happens when categories contain atypical items (in the prototype vs exemplar)?
people fail to learn the atypical items even after many trials which is more consistent with prototype theory
149
what is the main idea of categories as theories theory?
- categories are represented as a structured set of relationships between features: wings and flight go together because wings enable flight --> theory of birds
- to categorize new items: compare not only the item’s features to the theorized features, but also whether its features are related to one another in the way predicted by the theory
150
what are causual theories? they are related to which type of category theory?
- category as theories
- common cause structure: if F1 causes F2, F3, F4 then F2, F3, F4 are correlated
- common effect structure: F1, F2, F3 all cause F4 but F1, F2, F3 do not need to be correlated
151
what does the theory-based (causal) view of categories say about how people learn and use categories?
- people learn categories by understanding the causal relationships among features
- new item judgments rely on features and their correlations as predicted by a causal model
- common-cause or common-effect structures guide categorization
- causal correlations matter most, especially in common-cause structures
152
what are essence theories? they are related to which type of category theory?
category of theories theory
psychological essentialism: belief that categories are defined
by a hidden ‘essence’ that causes its superficial features
* biological kinds
* artificats - intended function
* people - ideal (scientists)
153
what are goal-derived theories? they are related to which type of category theory?
category of theories theory
ad hoc categories: categories constructed on the fly (things to eat on a diet, take the connections game in NYT)
* might have very few properties in common but we can use our background knowledge to determine whether they fit the goal
154
define transduction
tranforming one form of energy into another
155
how does the eye turn light into energy? what are rods and cones?
- photoreceptor cells in the retina turn light into energy
- rods: more sensitive in low light; low spatial acuity
- cones: less sensitive in low light; high spatial acuity
156
how does the ear turn sound waves into energy?
hair cells in the cochlea turn sound waves (pressure changes in the air) into electricity
157
how does the skin turn tactile pressure into energy?
mechanoreceptors turn tactile pressure into electricity
158
how does the nose and tongue turn chemical signals into energy?
taste buds and olfactory receptors turn chemical signals into electricity
159
what is psychophysics?
quantifies how physical quantities are transformed into psychological quantities
160
what is the just noticeable difference?
smallest difference that is seen as different from the reference point
## Footnote
if I turn on 1 (vs 20) additional light bulbs in a room, would you notice?
161
what is the weber-fechner law?
jnd is bigger at larger magnitudes
162
what is the weber fraction?
percentage increase required for detecting distinctions across different types of stimuli
163
what are the stages of visual processing?
1. retinal image: objects reflect light which is detected by our retinas. isn't literally a 2d coord system but close enought that it is usually treated this way in models/computer vision algorithms
2. image-based processing: primal sketch. primitive elements are parts of 2d images (edges, bars, blobs) defined by difference in light intensity. retinal reference fram (coordinate system is relative to the eye not enviro)
3. surface-based processing: 2.5D sketch. 2D surfaces embedded in 3D spaces
4. object-based processing: 3d geometry, each object have its own coordinate system. primitive elements are 3d volumes including unseen surfaces
5. category-based processing: affordances (the visible functions of an object), identifying the kind of object
164
what is the receptive field?
the area of the retina to which a neuron is sensitive
165
when and where does the visual information arrive?
after initial processing by several layers of cells in the retina and neurons in the thalamus, ‘raw’ visual information arrives in V1
166
what are simple cells, complex cells, and end-stopped cells?
- simple cells: neurons in V1
sensitive to a particular pattern for a
given receptive field in the retina (static)
- complex cells: sensitive to motion in a particular direction
- end-stopped cells: sensitive to movi ng corners, angles, or lines of a specific length
167
t or f? feature detectors are not involved in perception
false, they are involved
168
what did selective rearing experiments show in cats/dogs?
cats raised in an environment with lots of horizontal lines had more (more sensitive) horizontal simple cells and the same for vertical
169
what are selective adaptation experiments?
- participants are exposed to a repetitive stimulus that is constant in a particular feature
- after adapting to downward motion, the participant might perceive a stationary object as moving upward
170
how does edge detection work in light to edges?
> -1, +1
^ +1
-1
171
how is the primary visual cortex organized? what about v2-v5?
- v1 - retinotopically: neurons in the visual cortex are arranged to mirror the spatial layout of the retina
- v2-v5 have larger receptive fields
and become less retinotopically
organized. these areas have complex
connections to one another, including
feedback loops (forward and backward
connections among regions)
172
what is the main idea of the canonical perspective?
we recognize objects easier from canonically perspectives
173
what are the problems with the canonical perspective?
* insufficient information to recover a 3D representation of an object
* if we are just “mentally rotating” a misaligned object to canonical view, this already presupposes a 3D representation
* many objects have multiple canonical perspectives
174
what is the main idea of aspect graphs?
- represent the topological structure of objects
- reduce the number of stored viewpoints required
- includes a model of an object's 3D structure
175
what are the problems with aspect graphs?
- number of viewpoints becomes large very fast
- cannot explain discriminations between objects with the same topological structure
- requires that the 3d struture is already known (cannot guess the 3d structure of unfamilar objects)
176
what is the main idea for the recognition by components theory?
- geons: a vocabulary of volumetric primitives that can be composed into objects
- solve a hard problem (identifying objects in a viewpoint-neutral way) through an easier problem (identifyig geons and matching them to geon-based templates)
- easier to recognize: typical category members (shared geons); objects when their parts are primed
177
what are the main processing stages in recognition-by-components (RBC) theory of object recognition?
1. edge extraction
2. feature detection (curved vs straight edges, Y/K/L vertices, parallelism, symmetry)
3. object parsing (breaking into parts)
4. geon categorization (basic 3D shapes)
5. category matching (identifying the object)
178
what are some problems/points against the recognition by components theory?
- can't distinguish between objects with the same geon structure (there are only 108 geons)
- different breeds of dogs? clouds? faces?
179
what is the difference between objects and scenes in visual recognition?
- objects: acted upon
- scenes: acted within - made up of a background + organized objects
180
how quickly do we recognize the gist of a scene compared to specific objects?
we extract the overall scene gist faster than we can identify individual objects
181
what are the main global image features used to perceive scene gist?
naturalness, openness, roughness, expansion, and color
182
in the “nuts and bolts of language” framework, what is the web?
- our propositional knowledge stored in the brain
- multidimensional, abstract “language of thought”
- messy but internal to the mind
183
in the “nuts and bolts of language” framework, what is the string?
- a series of sounds in linear order
- exists in the world outside the mind
- speakers translate web → string; listeners string → web
184
in the “nuts and bolts of language” framework, what is the tree?
- the structure mediating between thought (web) and sound (string)
- can be produced from the web and recovered from the string
- governed by rules that make language: productive (infinite ideas from limited sounds), compositional (meanings built systematically from parts)
185
what is the discrete combinatorial system?
small bits can be combined into larger bits, which in turn can be combined into even larger bits
* grammar: Rules for combining bits into bigger bits
186
what is the hierarchy of language?
- sounds
- words
- sentences
- pragmatics
187
what is a phoneme?
- smalled unit of sounds
- correspond roughly to written letters, but with many exceptions
* ‘s’ in “boots” vs “mugs”
* vary across languages (english has some that french doesn't and vise versa)
188
what are sound waves, specifically from the hearer's perspective?
- changes in air pressure that are transmitted through the air
- bigger disturbances in pressure (amplitude dB) are perceived as louds
- faster changes in pressure (frequency Hz) are perceived as higher pitch
189
can complex waves be broken down?
yes they can be broken down into simple waves, resulting in a spectrum of frequencies
190
how does our mind use waves to identify sounds?
formants (F1, F2, F3) are regions of the spectrum with greater
amplitude, which are among the features hearers use to distinguish
speech sounds
191
how do we produce sounds?
- pushing air from our lungs through our articulators (larynx, tongue, lips, etc.)
- we use these articulators to alter the waveform
192
how do vowels and consonants differ in air flow?
- vowel: air flow is unobstructed from larynx
- consonants: air flow is obstructed
193
what are the 2 properties of vowels?
- height: position of the tongue (beat vs bait)
- roundness: whether the lips are rounded
194
what are properties of consonants?
- voicing: whether the vocal folds are vibrating (that vs path)
- place of articulation: where the major constriction happens (p, k, t)
- manner of articulation: how the constriction happens: stop (no air flow like d), fricative (narrow constiction like z), nasal (air flows thru nose like n)
195
what are the 2 sound features not in english
- tones (rising or falling pitch)
- clicks with 2 places of articulation
196
what are syllables comprised of?
- an initial group of consonants
(’onset’)
- a vowel with optional consonants (‘rime’)
197
what are feet in relation to syllables?
- rhythmic units that determine the stress pattern
- feet into words
198
what are morphemes?
- smallest unit of meaning: roots (chair, quack, glamour), inflection (-s, -ed), affixes (counter-, pre-, -ism)
- since they can’t be broken down further into meaningful units,
they have to be memorized by rote
199
what are words? what rules of syntax do they follow?
- atoms of syntax
- they cannot be broken apart by syntactic rules
rules apply to whole words, not parts:
✔ This monster eats mice / What did this monster eat? / This monster eats mice quickly This monster is a mice-eater
✘ This monster is a mice-eater / What is this monster an -eater? / This monster is a mice-quickly-eater
| forming questions, inserting adverbs
200
what is inflectional morphology?
modifying a word to fit the syntax of the sentence
* plural, tense, person (i play, he plays)
* noun --> noun-stem + noun-inflection
201
what is derivational morphology?
- using prefixes and suffixes to create a new word (possibly with a different syntactic category)
- the meaning is predictable from its parts
- -ness forms nouns from adj (redness) = the quality of being red
- cannot attach to verbs ^
- un- in adj means NOT X and un- in verbs mean REVERSE X
202
what is syntax?
rules governing how sentences are formed from words
203
what is syntactic category?
'parts of speech' like nouns, verb, adjective
* Defined by grammatical function, not meaning
* noun isn’t a “person, place, or thing,” but rather a word that combines in certain ways with other types of words
204
what do the rules of syntax operate on? what evidence supports this?
- syntactic categories, not words
- exchanges: inverting two words from the speech plan usually follow syntactic rules (a paper full of holes --> a hole full of papers NOT a holes full of paper)
205
how is a noun phrase made?
noun phrase --> determinant adjective noun
np --> (det) a+n
| the happy boy
206
how is a verb phrase made?
verb --> verb + noun phrase
vp --> v np
| eats icecream
207
how is a prepositional phrase made?
prepositional phrase --> preposition + noun phrase
pp --> p np
| at the beach
208
how can sentences be embedded in other sentences? there are 3 ways
- s -> either s or s
- s -> if s then s
- s -> np thinks that s
209
why is a tree a good engineering solution to the web/string problem? there are 2 main reasons
1. branches of a tree contain information that is related in the web
* “If the girl eats ice cream, the boy eats candy”
* the boy is eating candy, not the girl
* the boy’s candy-eating is contingent on the girl’s ice-cream-eating, not vice versa
2. a listener can recover the tree from the string using a set of predictable rules
* parsing strategies: principles listeners use to reconstruct the tree from the string – identifying each new word’s syntactic category and how it fits with the existing tree
210
what is late closure?
| parsing strategy
- when possible, attach new words to the current phrase rather than starting a new phrase
- “Before the police stopped the driver...”
- is the driver being “stopped” (object of existing phrase), or is the driver doing something before the police stopped (subject of a new clause)
- eyetracking finds pauses and backtracking when this expectation is violated
211
what is minimal attachment?
| parsing strategy
- attach new words using a few new nodes as possible
- karen knew the schedule...
- by heart
- was missing
212
in linguistics, grammar is ____ not ____
descriptive not prescriptive test for whether a sentence is grammatical is whether most native speakers of a language agree that it is grammatical
213
what is an example of the following: some grammatical “rules” taught in school are not scientifically valid rules of grammar
E.g., ending sentences with prepositions in English
* “I went with Sally”
* “Who did you go with?”
* this rule was invented by 19th century writers of style manuals, based on a false analogy to Latin (in which it is not possible to end a sentence with a preposition)
214
what is pragmatics?
interpretation goes beyond literal meaning
215
what is the difference between semantics and pragmatics?
* sentence meaning: the literal meaning of a sentence, as determined ‘bottom-up’ by its words and syntax (semantics)
* speaker meaning: the intended meaning of a sentence, as determined by the conversational context (pragmatics)
216
what is the relationship between sentence meaning and truth conditions in pragmatics?
* sentence’s meaning = the states of the world that make it true.
* example: “Snow is white” is true if snow is white in the real world.
If truth conditions aren’t met, the sentence is false
217
what are speech acts in pragmatics?
speaking is not just saying something, it's doing something
* asserting, questioning, scolding, commanding, promising, thanking greeting, swearing, warning
218
what are felicity conditions for speech acts?
conditions that must hold for a speech act to 'work'
for speaker S to request hearer H to perform act A:
* A must be in the future
* H must be capable of A
* S must want H to do A
* It must be obvious H won’t do A anyway
* If conditions aren’t met, the act is infelicitous
219
what is the difference between conventional implicature and conversational implicature?
- conventional: typical way a word is used by most people, beyond a word's logical meaning (implied contrasts, orders, etc)
- conversational: arise due to the specific context of use
220
how do coversational implicatures work?
cooperative principle: participants in a conversation work together manage their contributions and interpretations efficiently
221
what are the four main gricean maxims?
- maxim of quality: be truthful and do not provide information that is false or lacks evidence
- maxim of quantity: provide as much information as is needed, but not more than is required.
- maxim of relation/relevance: make your contribution relevant to the ongoing discussion
- maxim of manner: be clear, brief, and orderly, avoiding obscurity and ambiguity
222
what are children's inferences from disfluent speech?
- speakers are likelier to pause if searching for a word for an unfamiliar object
- eye-tracking studies of 2- year-olds show that they expect disfluent speech will refer to unfamiliar objects
223
what is the combinatorial structure of every level of language?
Features -> Phonemes -> Syllables -> Feet -> Words
Morphemes -> Words
Words -> Phrases -> Sentences
224
how do we encode and decode thought?
systematic rules for converting between thoughts and trees and
between trees and strings
225
do the algothrims we learned work?
yes they work well but are imperfect
226
what is light?
electromagnetic radiation that is visible to the human eye, essential for vision
227
what is pinhole camera?
simple camera without a lens that uses a small aperture to project an inverted image
228
what is refraction?
bending of light as it passes from one medium to another, crucial for image formation in lenses
229
what are photons?
particles of light that carry energy and are fundamental to the quantum theory of light
230
what is the retina?
light-sensitive layer at the back of the eye that converts light into neural signals
231
what did isaac newton propose?
the particle theory of light and conducted experiments demonstrating that white light is composed of various colors
232
what did christiaan huygens suggest?
that light travels as waves through a medium called aether
233
olaus roemer was the first to do what?
measure the speed of light based on observations of Jupiter's moons
234
willebrod snell formulated what?
the law of refraction, explaining how light bends when entering a different medium
235
johannes kepler recognized what?
the retina's role in vision and the projection of images
236
what is the wave-particle duality?
theory that light exhibits both wave-like and particle-like properties, explaining various opitcla phenomena
237
what does the theory of evolution say about the development of the eye?
the development of the eye through natural selection, addressing the complexity of eye evolution
238
what is the sine law of refraction?
describes the relationship between the angles of incidence and refraction when light passes between different media
239
what did hecht, schlaer and pirenne do?
conducted experiments to determine the number of photons required for the human eye to detect light, revealing the sensitivity of retinal receptors.
240
what did yarbus do?
pioneered the study of eye movements and how they relate to visual perception and attention
241
what is accomodation (in relation to the eye)?
ability to change shape for focusing on near or distant objects
242
what is pupil reaction to (in relation to the eye)?
pupil constricts in bright light and dilates in dim light, controlling the amount of light entering the eye
243
what is retinal sensitivity (in relation to the eye)?
retina contains rods and cones, with rods being more sensitive in low light and cones responsible for color vision
244
what is a blind spot (in relation to the eye)?
area on the retina where the optic nerve exits, lacking photoreceptors, resulting in a blind spot in vision
245
what are after-images (in relation to the eye)?
visual phenomena that occur when the photopigments in the retina are bleached, leading to temporary visual impressions after the stimulus is removed
246
what is intelligence according to pinker?
ability to attain goals in the face of obstacles by means of decisions based on rational rules
247
what is consciousness according to pinker?
state of being aware of and able to think about one's own existence, thoughts, and surroundings
248
what is the turing machine?
machine that manipulates symbols on a strip of tape according to a set of rules, used to define computability
249
what is the computational theory of mind according to pinker?
theory that human thought processes can be understood as computations performed by the brain
250
what is mentalese according to pinker?
hypothetical language of thought that represents concepts in the mind
251
what is the causal theory of meaning according to pinker?
idea that the meaning of a symbol is determined by its causal relationship to the object it refers to
252
what is the inferential role theory according to pinker?
theory that the meaning of a symbol is determined by its role in inferences and the relationships it has with other symbols
253
what is the chinese room?
* a thought experiment designed to argue that mere symbol manipulation—which is what computers do—is not equivalent to true understanding or consciousness
* depicts a person in a room who, despite lacking any knowledge of Chinese, can produce perfectly comprehensible Chinese responses by following a rule book of instructions
254
what is the twilight zone in the pinker exerpt?
an episode that raises questions about consciousness and the emotional connection to artificial beings
255
according to pinker, what is his criticim for behaviourism?
- B.F. Skinner: “the question is not whether machines think, but whether men do”
- Behaviourism: responses explained by stimuli + reinforcement
- Pinker’s critique:
- Example of Sally escaping a fire:
- She runs out because she believes the building is on fire and does not want to die
- Not a simple stimulus-response: depends on beliefs/desires (e.g., smoke from toaster != escape)
- Conclusion: beliefs and desires are indispensable explanatory tools
256
what are phantom limbs?
sensation that an amputated or missing limb is still attached and functioning, often accompanied by pain/discomfort
257
what is neuroplasticity?
brain's ability to reorganize itself by forming new neural connections throughout life, allowing for recovery and adaptation after injury
258
what is mirror visual feedback?
therapeutic technique using mirrors to create the illusion of movement in a phantom limb, helping to alleviate pain and improve function
259
what is the sprouting theory?
suggests that nerve fibers from the face invade the area of the brain that corresponds to the missing limb, leading to sensations in the phantom limb when the face is touched
260
what is the unmasking theory?
proposes that prior to amputation, sensory input from the face partially activates the hand region in the brain, which becomes more pronounced after the limb is lost
261
what is chronic pain and how does it relate to learned paralysis?
- development of learned pain, where movement becomes associated with pain
- carryover of paralysis sensation to the phantom limb after amputation
262
what is the penfield map? what is it also refered to as?
- cortical homunculus
- functional maps of the cerebral cortex
263
what is chatgpt doing at each step?
choosing the next toxen based on probabilities given the text so far
264
what is temperature in chatgpt generation?
a control of randomness in token choice low temperature is deterministic high temperature is more creative
265
why not just use n-gram models
because the number of possible word sequences grows too fast and n-grams cannot capture long range structure
## Footnote
n-gram models are one of the older, simpler ways computers tried to “do language” before modern neural networks came along
an n-gram is just a chunk of n items in a sequence — in text, those items are usually words (though sometimes characters)
266
what brain structure gives the raw sensation of pain?
insula deep beneath the temporal lobe
267
what brain structure gives pain its unpleasantness and sense of danger?
the anterior cingulate in the frontal lobes
268
why did mikhey and dorothy laugh from pain?
damage cut the pathway from insula to anterior cingulate so pain was felt without agony creating alarm plus calm the recipe for laughter
269
what does a neural net provide that n-grams cannot?
generalization from training data through parameterized functions that approximate probabilities
270
how does a neural net learn?
through gradient descent adjusting weights to reduce error between predicted and actual next tokens
271
what are embeddings? how do they help?
- high dimensional numeric representations (vectors) of words where similar contexts cluster together
- they let the model infer meaning and context even for new or unseen word combinations
272
what is the idea of semantic laws of motion?
generating text can be seen as a trajectory through meaning space guided by learned probabilities
273
why does chatgpt work so well?
language is structured and redundant so statistical prediction captures much of its flow
274
what can chatgpt not easily do?
deep algorithmic reasoning or tasks requiring step by step irreducible computation
275
what is computational irreducibility?
some processes cannot be shortcut and must be run step by step
276
what are future directions wolfram suggests?
hybrid systems linking llms with symbolic tools or new architectures that integrate memory and computation
277
according to pinker, what does a production system contain?
- production system contains a memory and a set of reflexes, sometimes called "demons" because they are simple, self-contained entities that sit around waiting to spring into action
- the memory is like a bulletin board on which notices are posted
- each demon is a knee-jerk reflex that waits for a particular notice on the board and responds by posting a notice of its own.
278
what is de morgan's law?
from –(P&Q), we derive –P | –Q
PSYCH 256 - Cognitive Science
flashcards
Decks in class (4)
# Cards
