Cognitive🔢 Flashcards
1
Q
Cognitive machine challenge
A
Build a machine that reasons with logic: create new logic from facts already known (inferences)
Use symbolic logic (replace all words with symbols so machine can make inference automatically)
2
Q
Existing logical computers
Blocks and chess
A
Blocks 1970s- Computer solves spatial arrangements by starting with known facts and infers how to arrange
Chess 1990s- start with known facts of chess positions, generate new positions and consequences of moves, choose best outcome
3
Q
Learning with neural networks
A
Strengthen connections
Before- U to R
After-CS strengthened connections to U
4
Q
3 types of Cognitive psychology models
A
Data analysis model- data driven and descriptive (line graph)
Box and arrow model- information processing, conceptual. IMPLICIT assumption
Computational model-informational processing as simulation, various levels of abstraction. EXPLICIT assumption
Performs cognitive tasks to learn how it is implemented in the brain, without neurobiology
5
Q
Explicit vs implicit models
A
Cognitive process- implicit models
Computational models can make assumptions explicit and assumptions can then be tested (specific predictions for outcomes)
6
Q
Goals of modelling
A
Must be specific, models are often implicit
Can test assumptions
Predicts outcome, help to select which experiment to perform and distinguish between models
Can be exploratory (not predictive) trends difficult to show from the brain (oversimplified however)
Abstraction and idealised real concepts
All models wrong but some are useful but explanation does not imply prediction
7
Q
Marr’s 3 levels of understanding
How to think about info processing systems like the brain
A
Computation-WHY (problem) the goal e.g. recognise objects
Algorithm- WHAT (rules) represented approach, how carried out e.g. detect edges, outline
Implementation-HOW (solved physically) e.g. visual neurons sensitive to lines
8
Q
Bottom up approach
A
Implementation (neural circuits)
Algorithm (how generate algorithms)
Computation (what can we solve this with)
Neuroscience favours this
9
Q
Top down approach
A
Computational (specific problem)
Algorithm (to solve problem)
Implementation (how these representations can be implemented in neural circuits)
Experimental techniques favour this method (epistemological bias)
10
Q
Marr’s level favoured by neuroscience
A
Implementation (has epistemological bias towards it)
Bottom up
11
Q
Moravec’s paradox
A
Computers can do hard tasks to us e.g. solving tasks but not easy tasks to us e.g. perception
12
Q
Multisensory integration and touch/movement study
A
Seeing and feeling that you touch the object is important
Anaesthetise finger to block all touch sensations (does not affect motor abilities)
Much slower and less precise to pick up objects, need to integrate
13
Q
Reference frames
A
(representational schemas) represent different info from senses, transform representations to common representation
Need to know body orientation and position of object, relative positions(body schema) to external space to unify frames
14
Q
Reference frames of the senses
A
Vision- eye centred/retinal, location of stimulus on retina
Audition- head centred, location of sound with respect to ears
Touch-body centred, location of tactile stimulus on skin
Need to convert between reference frames and to external space (the world, irrespective of location and orientation of body)
15
Q
Reference frames and snake game
A
Need to know information about snake’s body layout to convert from player perspective to snake perspective
See space in top down view as player, but could control actions of snake in different view (first person)
16
Q
Reference point
A
Reference frames need to be in same reference point
‘Is the object I’m seeing the same object I’m touching’
17
Q
Transformations needed for us to convert between reference frames
A
Eye to head transformations- orientation of eyes needed
Head to body transformations- orientation of head needed
18
Q
Body schema
A
Position of body in space, relative position of body parts
19
Q
7 body schema factors
A
Spatially coded- body parts in external and relative space
Adaptive-changes over lifetime
Modular- body parts processed in different brain regions
Interpersonal- others movements make sense to you
Supramodal-combines input from proprioception, senses
Coherent- Keeps continuity so don’t feel disembodied
Updated with movement- continually tracks posture
20
Q
Body image
A
Body percept (‘feel’ you have a body) Body concept (of what a general body look like) Body affect (how we feel about our body)
Structural description- hand attached to arm
Body semantics- names for body parts
21
Q
Does body posture affect perception?
Temporal order judgement tasks
A
Participant stretches fingers on hand they think was stimulated first
Crossed arm condition-
worse at body perception, body schema INTERFERES with perception, confused about where touched
22
Q
How does body schema develop?
Experiment-temporal order judgement task
A
4 month olds no difference if feet are crossed or not, no body schema interference BUT
6 month olds reach to incorrect foot half the time when crossed, Body schema/ posture now matters for perception. Interferes with tactile orientation
23
Q
Cross modal contingency- tactile discrimination task
A
Determine which finger was vibrated:
Visual light distractors on same finger that was vibrated OR other finger
Delay= incongruent reaction time - congruent reaction time
Longer delays when distractors on same hand vs different hand
Touch and vision, two reference frames are converted between
24
Q
Cross modal contingency task with crossed arms
A
Tactile stimulus on same side of body, visual stimulus on different side of body
Effect of visual distractor (delays) moved with hand when arm crossed as it switches visual hemispheres.
Cross modal interactions mediated by body schema
25
Peripersonal space
Space immediately around the body, objects can be immediately grasped, can contract and expand
26
Tool use
Tools are incorporated into body schema
| Crossed hands-some delay effects
27
Body schema disorders
Alice in wonderland syndrome- size perception distortion, body parts appear smaller or bigger associated with childhood and migraines
Autotopagnosia- inability to locate body parts, loss of spatial unity. Fused percept of fingers
Phantom limbs- body schema do not adapt, feel sensory input of missing limb. Often painful and can change in size over time
28
Neural basis of body schema
Cross modal neurons found in brain, respond to inputs from several senses
Neuron fires when body touched or if visual stimulus moves close to hand
Visual receptive area moves with hand, modified by body schema
29
What happens with integration when posture changes
Crossmodal effects remap
30
Neural basis of tool use
Neurons incorporate tools, expands space during tool use, encode the peripersonal space
Monkeys: neurons responded to stimuli at far end of tool which would have been too distant from the monkey to be triggered in peripersonal space (without tools)
Crossing tools- connect right hand to left visual field and becomes an extension of the hand
31
Mirror experiments
Tactile stimulation corresponds to visual stimulation seen indirectly in the mirror reflection at a distance
Reflections near participants hands re-coded as originating from peripersonal space near those hands. Peripersonal space extends if sees self in mirrors but not when see another persons same parts
32
The integration problem
Integrate into a common reference frame
Vision and audition in different reference frame, sensory conflict
(Can see and hear the dog so where is it?)
33
Reducing lens experiment-Testing conflicts between vision and touch
Participant manipulated object, look through reducing lens- object looks smaller but feels bigger
Asked to judge size by vision or touch
Visuals- trusted more than touch (VISUAL CAPTURE)
also if could draw or feel before (touch or motor)
Touch- small but consistent influence
34
Testing for a sensory hierarchy experiment
Report number of auditory beeps played during flashes of light
Number of beeps determined reported number of visual flashes
Audition can influence vision -NO strict hierarchy
Visual capture not universally true
35
Visual capture
Trusting visual info over other senses
36
Modality precision hypothesis
Use the modality with highest precision (lowest uncertainty) for task
Spatial task- vision
Temporal task- audition
37
Where do we get sensory uncertainty (affecting precision) from?
Perceptual limits: spacing of photoreceptors in the fovea
Neural noise in synapses
Cognitive resourse limits (attention)
38
Maximum likelihood estimation (two models)
| How to solve a problem
Normative model- establish bounds, is this the best method? Based on theory e.g. Minimise uncertainty
Process model- how a problem is actually solved, based on data
39
Normative model (sensory integration)
Pick integration method that minimises sensory uncertainty (maximum likelihood estimator)
Smallest sensory uncertainty is integrated (low variance = low sensory uncertainty
If one mode is more certain, can rely on it more
40
Optimal weights of senses
Haptic input is weighted higher due to lower sensory variance
Integrating information from multiple sources always causes uncertainty to decrease
41
Manipulating sensory uncertainty experiment
Virtual bar with sensory conflict between haptic and visual information
Asked to judge the size of the bar
Determine point of subjective equivalence, manipulate uncertainty
Discrepancies between visual and haptic can change uncertainty
Without visual noise- perception biased to visual input
With visual noise- perception determined by both haptic and visual
High visual noise- determined by haptics
Human performance follows optimal sensory integration rules
42
Do we always integrate info optimally?
Need to know uncertainty for optimal integration- hard to estimate, easier in sensory perception than cognitive reasoning
Calculations can be intractable or take long time- heuristics are subopitmal but fast
43
Estimating sensory noise vs cognitive noise
Estimating tilt of lines with sensory noise (low contrast) or cognitive noise (random tilts to different lines)
Integrate sensory noise OPTIMALLY but cognitive SUBOPTIMALLY
43
How might probabilities be encoded in the brain
Mean and variance
Full distribution
Samples
44
The correspondence problem
In external space: if hear a woof and see a dog how do you know it is one dog or two?
45
Conflict tasks
Stroop-Word reading (automatic) interferes with controlled colour naming
Flanker- respond to central arrow >><>>
Simon- press left or right button with congruent or incongruent hand
Go/no go task- assess inhibitory control
46
Response manipulations: reverse stroop
Point at coloured patch, colour word appears in middle in a different ink
(Reverse stroop test)
Associations between stimulus and response (response compatibility) causes automacity, is not just a sensory process
Colours interfered with word naming not the other way around
47
Attentional manipulations
Stroop effect eliminated or reduced when only one letter was coloured
Automatic processes do not operate independent of attention
48
Stimulus onset asynchrony manipulations
Onset of colour and word (of stroop) presented at different times
Flash up coloured background then written word
No amount of head start for colour information produced interference on word reading
Automatic processes are not simply faster
49
Training on stroop task
Trained to name shapes as a colour
Shape then presented in a contrasting colour ink to its ‘name’
Initially colour ink interferes with naming the shape (as a colour)
Later- colours interfere with shapes and vice versa
Training- shapes interfere with naming colour
Automacity as a continuum, with practice anything can become automatic, a process in not either (completely) automatic or controlled
50
Making skills automatic and controlled processes
Skills are overlearnt stimulus response pairs, triggered by environment, rapid and stereotypes
People can alter skills, less habit like, elaborate cognitions
51
Logan and crump typing skills experiment
Participants type words, feedback on screen
Errors corrected OR shows error that wasn’t made by participant
When errors inserted, participants believed they made it (illusion of authorship) Sensitive to feedback not actions themselves.
Slowed when made real error (even when not told) and didn’t slow when told they made one
52
Typing skill hierarchical loops
Skill controlled by hierarchical loops. Automate some parts to pay attention to harder parts. Complex behaviours neither entirely automatic or controlled
Outer loop- language comprehension and generation decode on words to type, output on screen
Inner loop- finger and keyboard interactions sensitive to feedback from fingers
53
Yerkes-Dodson and skill
Optimal performance reached with arousal but too much reduces it
54
Choking (football players dribbling)
High arousal leads to poor performance usually performed well
Dribbling- footballer vs beginner either focus on task OR an interference task with dominant OR non dominant foot
Beginners better when not distracted but experts best if slightly distracted
55
Typists misallocation of attention
Typists told to leave out letters typed by other hand (misallocation of attention)
Disrupted automatic skill, had to think about how it was done
Anxiety can cause this
56
Ironic processing
Must not think about how the skill is performed
| But trying not to makes you think of it
57
Ericsson’s theory of deliberate practice
Improvement slows but does not stop with practice, immense skill:
```
Effortful, extensive practice must be deliberate
Breaks skills into components
Focus on reducing errors
Use targets (evolve as skill increases)
Individually tailored training
```
58
Training for elite violinists and figure skaters
Violin- 4-5 hours practice a day, naps and feedback from coaches, focus on difficult elements
Figure skaters- gold medallists made more training errors, more mistakes practicing beyond current limit
59
Replicating Ericsson
Practice alone does not make an expert (students)
| Coaching less important
60
Rationality
Defined in terms of norms- rules of action or thought which define optimality
Consistency (coherence)
Correspond to reality
61
Availability bias and framing bias
Availability bias - overestimate frequency of an event from how easy it is to come to mind
Framing bias-identical choices presented as losses or gains can affect decision
62
Conjunction fallacy (Linda)
Linda fallacy- choosing that Linda (left wing philosophy student) is more likely to be a feminist and a bank teller than only a bank teller
63
Decision calculus
Calculate the option with highest utility
| Compare value of different options, consider uncertainty
64
Marginal value of utility
From nothing to something = happy
But more than this won’t increase to same amount
A method for contrasting uncertainty
65
Calculating expected utility (decision making method)
E= P x U
expected utility = probability x utility
Lottery ticket is £15 with 20% chance of £50 and 10% chance of £100
0.2 x 50 + 0.1 x 100 = £20
Expected utility has to be more than what you are paying to begin with so you SHOULD buy since you payed £15
66
Why we use heuristics
Bounded rationality
The world is complex
Decisions need to be made quickly and time is limited
Cognitive capabilities are limited
67
Ecological rationality
Rationality- designed to fit with the environment
Sometimes coherence is sacrificed for useful decisions
Can predict when fails and succeeds
68
Adaptive value
Value of action across evolution (if successful in the long run)
Long term expected value:
Errors are permissible if average benefits avoid costly mistakes
69
Environmental structure and heuristics
Decisions mesh with environment
| Heuristics make assumptions about the environment, what goes together (association) and probabilities (risks)
70
Recognition heuristic
Estimate larger frequency with familiar things
Works only with small amount of info, too much interferes with probability judgement
Memphis assumed to be bigger, Missoula unknown to us
71
Wason’s selection task and 2-4-6 task
Wason- confirmation bias of what know already (evolutionary)
should turn over 1 and E
2-4-6 Mostly confirmation bias, generate number sequences which fit to verify hypothesis (positive tests)
72
Biases are not mere errors or flaws
Mistakes are one off, errors are consistent
Biases are defined by norms (formal logic, probability etc) or evolutionary adaptiveness- so not flaws
Errors- heuristic system (fast and effortless) gets out of hand instead of rational system. Make wrong choice but for a reason, different cost/benefit analysis
73
Why might we make systematic errors
Use strategy optimised for different environment
Consider different choices
Use different cost/benefit analysis
74
Dual process theory
When two processes work at the same time, or support the same capacity
E.g. rational vs heuristic thinking work together, switch between them in the right circumstances
75
Heuristics
Robust, reduce chance of noise
Exploit free capabilities and provide solutions different from logic and probability
Customised to solve diverse problems
Ecological rationality- can predict when it fails and succeeds
76
Hill small group decision making
3-6 people in short tasks
Process loss and gains
Mostly groups performed at accuracy of second best member (attempt to find average)
77
4 factors why comparing groups is hard:
Task type
Standards of comparison
Coordination methods
Individual differences
78
Task types
Intellective vs judgement
Well defined vs ill defined goals
Insight task
Require background knowledge? Provoke strong intuitions or biases?)
79
Standards of comparisons for groups
Group is considered ‘good’ e.g. better than the worst participant
80
4 Coordination methods (for group to come to an answer)
No discussion- averaging answers
Iterative (Delphi) - see others’ answers and adjust answer
Dictator -Pick best individual to answer
Discussion (consensus, dialectic model)
81
Individual differences in a group
Information sources (cues)
Ability (working memory)
Other capabilities
Make individual decision about the ‘truth’ of the world by picking up on cues
82
Gigone and Hatsie- individual decisions in a group
‘The truth’ based on cues, different people pay attention to different things
Weightings are different
Summed up in group differently depending on their method (of who to converge on for final answer)
More accurate if members are highly correlated with environment
83
When averaging works in group decision
Individuals provide:
```
Independent estimates
without systematic biases (no strong intuition)
uncorrelated errors (no coordination between members so that estimates can average out to find answer)
```
84
Uncorrelated and correlated errors
Uncorrelated errors- Average out the error/noise. Won’t work if all have strong intuition
Corrected errors- from limited information, individual biases and group conformity
86
Group think
Polarisation in decision making
Obsessed with one answer, overconfident, blind to errors
Conformity skews decision making
86
Groupthink negative factors criticisms
Doesn’t always happen, move towards right answer
Not distinct phenomenon, a series
Wisdom of crowds-can make judgement with cues, better when see them easily
88
Interactionist approach- evolution of reasoning
Capacity to reason and argue evolved to PERSUADE others, not to solve problems (individualist account)
Confirmation bias to advance better arguments to persuade others in groups
89
Truth wins strategy
Best member straightforward and demonstrates the truth of solution to group to persuade them
80% groups converted to correct answer despite being against intuition/bias. Arguments change people’s representation of the task
(Intellective tasks)
90
Collective intelligence
Individuals’ intelligence does not predict high collective intelligence much
Groups of smart people do not outperform non smart
Successful group judgement depends on way errors are distributed across the group
91
Reading the mind in the eyes study
Had to guess the emotion of eyes
Soduku tests online or offline to test intelligence
Better groups at reading eyes in mind task (theory of mind and perception) were better at sudoku online (didn’t see persons eyes) vs in person which is surprising
Better reading mind in eyes task predicted better collective intelligence= more coordinated
92
Predictors of group intelligence
Coordination (reading the eyes in the mind task)
Average social sensitivity
Amount and distribution communication
Collective intelligence (correlated with theory of mind ability)
93
Diversity of editing in Wikipedia articles
Longer more complex discussion with diverse editing produced higher quality Wikipedia articles
94
Group homogeneity
Group members share sources or info and or biases
Collective decisions are less likely to incorporate all relevant info and more likely to be biased
95
Intellective tasks vs judgemental tasks
Intellective (eureka)- demonstrates solution
Judgemental- no demonstrable solution
96
Why group judgement can fail
Group’s accuracy depends on the accuracy of individual member judgements
Cues can influence group
Difficult to identify best member in non demonstrable solution, have to rely on intuitions about credibility
Systematic bias or persuasive individual
97
Social attention
Drawn to social information in a scene
98
Eye tracking
Desktop-powerful, less portable
Mobile- not as powerful
Records saccades, microsaccades, fixations, pupil dilation
Shows what captures the attention, insights to underlying cognition
99
Eye fixations (Yarbus)
Yarbus- participants mostly fixated on people in a painting and moved around social information
100
Where humans prefer to look
For social information
Attend to others and movement
Body parts to learn intentions, insight into behaviour
Interaction is crucial to development, understanding cues and learning behaviour helps us integrate into a social group
101
Foreign language exposure study
Infants with English as a first language taught Chinese by native speaker or on tv
Exposed to language from Chinese speaker=significant learning
102
Eyes and gaze in babies
First week-direct attention to eyes. Look at image with direct gaze for longer, looked more times
3 months-follow gaze
12 months- orient attention to gaze location
103
Monsters eyes study
Compare eyes in humanoid figure, monster and human figures
| Fixated on eyes themselves not the configuration of the face
104
Human eyes
Smaller areas of dark than primates so can follow our eyes
Contrast between light and dark allows us to communicate
Receive and send information, understand people’s thought processes
When reduce polarity, difficult to tell where eye looks since dark area shows where
105
EEG social attention
N170 associated with faces, greater amplitude
| Sensitive to eyes but slightly different ERP, different process
106
Insights from fMRI social attention
Many regions implicated in social attention:
Gaze direction, eye contact- amygdala
Facial identification- fusiform gyrus
theory of mind processing-prefrontal cortex
107
Direct and averted gaze on fMRI and eye tracking
fMRI and eye tracking- greater activation in fusiform gyrus to direct gaze
Greater attention to eyes and mouth to direct gaze (eye tracking)
108
Direct and averted gaze in conversation study
Live conditions- real person interviews with direct and averted gaze
Video condition-same but video recording
REAL they looked at face when asked and background to answer. Reduced cognitive load by looking away, averted gaze shows haven’t finished speaking and direct gaze shows we are listening
VIDEO look at face and background
109
Gaze cueing
Develops in infancy, move own gaze when someone moves theirs to look at something
Gaze following important prerequisite for transgenerational learning
110
Cueing paradigms (Posner)
Faster to find target if at same location as the cue
Significantly faster if cue was on the same side
When on different, have to move eyes to the cue at different location
111
Real world implications of moving eyes to cues
Aware of important information in the environment and adapt behaviour
Helps plan our own actions
Insight into other’s intentions
Reciprocal eye contact and attention to social information allows us to fit in as a member of social group
112
Gaze cueing and status
Shown picture and CV of high or low social status
Larger gaze cueing for high status faces
Gaze cueing not just bottom up processing, responds to environment (top down processing)
Also more if male, resemble onlooker, shared political alliance
113
Social context eye tracker study
Bought coffee with eye tracker
Pedestrians far away more likely to be fixed on in real life vs video
Less likely to look at pedestrian when near in real life vs video
114
Social attention in autism study
Film clip shown to autistic people
= more attention to the mouth than the eyes
SO misinterpreted the scene from focusing on the mouth
Look less at faces when people look directly at them
Attend to confederate longer when told they are begin watched
Evidence of no spontaneous gaze cueing
115
Social attention
| Testing nuerotypical people on Autistic spectrum
No relationship between autistic traits and attention to interviewer-live condition
Those with higher amounts of autistic traits looked LESS at the interviewer in the DIRECT eye contact condition- video
BUT research differs so Autism different for everyone (heterogenous)
116
Social attention and prosopagnosia
| Gaze cuing
Don’t recognise faces
No gaze cueing effects at shorter duration when viewing full face, may have inability to process whole face at once.
Gaze cueing at short duration when view eyes only
117
Parts of the neuron
Dendrites - receive messages from other cells
Axons- pass message along cell body to other neurons, muscles or glands
Myelin sheath speeds up electrical impulse (EEG detects electrical activity generated)
118
How fMRI works
Oxygen delivered to neurons by haemoglobin to replenish them after firing
More blood flow and oxygen transferred when activated
This is recorded in the fMRI (where blood flow is directed)
119
Diamagnetic or paramagnetic blood
Haemoglobin is:
Diamagnetic when oxygenated - strong magnetic field
Paramagnetic when deoxygenated- weak magnetic field
BOLD response- fMRI measures this
120
Negatives of fMRI
Delay between responses to stimulus and the measurement (around 4secs)
Initial dip in blood flow followed by re-compensation so overshoots until levels off
121
Different slices of structures
Caronal-across crown ➡️
Sagittal- from side⬇️
Axial
122
Impact of culture on facial expression processing study
White Americans and Japanese participants-reading in minds eye
Own culture bias for facial expressions when stimulus is own culture
On fMRI, same brain regions were responsible (superior temporal sulcus) but more activation for own culture
123
Autistic individuals and processing emotional images
| Kana et al
3 videos- emotion (identify it) object (say what is missing, not emotional) and neutral (control)
Object condition: emotion still unconsciously processed when not asked to judge it. Autistic people reduced activation in (medial prefrontal cortex) for implicit processing, do explicit better. Different activity in neurotypical participants
124
Kana et al study implications for autistic people
Recruit task specific brain regions for processing emotions when EXPLICT asked SO use specific instructions
Reduced activity in the brain region, especially when doing another task SO may be why they struggle to recognise subtle emotion
125
Combining fMRI and EEG
Complement weaknesses (fMRI weak temporal and EEG weak spatial)
Can see when and where something happens
BUT costly, timely, lots of equipment, high drop out
126
Early visual processing and EEG
Immediate response to stimulus
ERP- consistent pattern in response to stimulus/electrical trace
P100 (positive spike at 100ms when see visual stimulus)
The P100 spike happens in P07 electrode when stimuli is presented on the right and in P08 when on the left (visual evoked potential)
Look to fMRI to pinpoint which structures are involved
127
fMRI and EEG combined responses for faces vs non faces
Different responses: minus the responses to show regions involved only for faces
Faces processed differently to non faces - N170
129
Hyperscanning experiment
Monitor speaker and receiver
Both sides of interaction, reciprocal pattern
6 couples: females express joy, anger, fear, disgust, sadness
Males watch and try to feel the same
Brain scans- activation in same areas (SHARED NETWORK EFFECT) relationships share networks. Can influence activity in another’s brain
129
Brain as a predictor framework
1 Hypothesise regions involved in cognitive interest
2 Measure behavioural outcomes in these areas
3 Test whether step 1 predicts step 2
Integrates traditional neuroimaging methods with behavioural outcomes beyond immediate experiment, ecologically valid
130
Evaluate second person cognition
Shows brain regions involved in social cognition and how brain activity of one person affects another’s
Ask new questions about social behaviour and why some have social interaction difficulties
131
Second person cognitive neuroscience and hyperscanning
fMRI and EEG lack ecological validity
Use pre recorded gaze contingent avatar in real time instead of passive 3rd person recording
Networks activate even when not doing a mentalising task (trying to engage with avatar) unlike when in 3rd person
132
Communicative interpretation Linda problem
Could be seen as not a rational (logic) problem
| Find relevant information through communication
133
Positive test strategies and negative
Positive verifies the hypothesis
| Negative falsifies the hypothesis
