Cognitive Psychology Attention What do these have in common? You are driving to a lunch date, and accidentally take the route to your job. After you correct your route, as you are driving by the theatre, a red ball chased by a child suddenly appears on the street, and you screech your brakes. You get to the restaurant and try to find your friend, who has flaming red hair. The restaurant is packed, it’s hard to make-out faces, but you can see people’s hair so you look for red hair. When you get to your table your friend asks if you noticed the Star Wars promotion with two costumed people fighting with light sabers. As you talk about important but dull business, your mind keeps drifting to your exciting first date last night. You force yourself not to think about it, but it keeps coming back. Innatentional Blindness (original experiment) http://www.youtube.com/watch?v=vJG698U2Mvo Change Blindness (office) https://www.youtube.com/watch?v=diGV83xZwhQ Change Blindness Counter experiment: http://www.youtube.com/watch?v=mAnKvofPs0 Campus Door Demo: http://viscog.beckman.uiuc.edu/flashmovie/12.php Construction door http://viscog.beckman.uiuc.edu/flashmovie/10.php Gradual Change: http://viscog.beckman.uiuc.edu/flashmovie/1.php Aspects of Attention 1. Detection. 2. Filtering and selection. 3. Search. 4. Automatic processing. 5. Concentration. Architecture The box model: Sensory Store Filter Input (Environment) Pattern Selection Recognition STM Response LTM Attention In this model, attention is: The filter and selection boxes The arrows. The special job carried out by each of these boxes according to different theories of attention (Yes, this is cheating) In this model attention: Puts together information from various sources. Gets information into STM Works in imagery Detection Two kinds of thresholds: Absolute Threshold: Minimum amount of stimulation required for detection. Difference Threshold (“Just Noticeable Difference”): Amount of change necessary for two stimuli to be perceived as different. Detection Absolute Thresholds: Vision: One candle, on a mountain, perfectly dark, 30 miles. Hearing: A watch ticking 20 feet away. Smell: A single drop of perfume in a three room apartment. Touch: The wing of a bee on your cheek. Taste: One teaspoon of sugar in two gallons of water. Determining Thresholds How to determine thresholds: Method of limits: Ascending: Start with a value below the threshold, increase, ask for detection, increase… At the point a person says “detect,” average that stimulus value with the value from the previous trial. Repeat to estimate threshold. Descending: Same, but start above threshold and work down. Combining results from both directions will give you an estimate of the threshold. Determining Thresholds How to determine thresholds: Method of constant stimuli: Present a series of randomly selected stimulus values, ask for yes/no response for each. The value that’s detected 50% of the time is the threshold. These methods can be adapted to determine difference thresholds. Determining Thresholds We think thresholds work like a step function, but they don’t. They are sigmoid or ogive curves This graph represents a step function. Below the threshold there is 0% detection. Above the threshold, there is 100% detection. This is the way we normally believe our perception to work. This graph represents an ‘ogivecurve’ and how detection really changes – it is a gradual slope. The threshold is defined as a 50% detection rate. Determining Thresholds Difference Threshold: Weber’s Law: K = ΔI / I K is the Konstant Δ is the difference I is the stimulus intensity The formula states that the threshold for noticing a difference (whether it’s the length of a line or weight of a dumbell) is a constant ration between the ‘old’ / background stimulus and the ‘new’ / target stimulus. Determining Thresholds Early Researchers Noticed: Thresholds Shift! These are ogive curves for stimuli of the same intensity but with different signal to noise ratios or payoff matrix How to get around this problem: A model that accounts for signal to noise ratios and payoff matrixes Signal Detection Theory Signal Detection Can estimate detection (sensitivity) independent of bias. Two kinds of trials: Noise alone: Background noise only. Signal+noise: Background noise with signal. Two responses from observer: Detect. Don’t detect. Signal Detection: Four Situations State of the world Signal Noise Yes (Present) Hit False Alarm No (Absent) Miss Correct Rejection Response Hits (response “yes” on signal trial) Probability density Criterion N Say “no” S+N Say “yes” Internal response Correct rejects (response “no” on no-signal trial) Probability density Criterion N Say “no” S+N Say “yes” Internal response Misses (response “no” on signal trial) Probability density Criterion N Say “no” S+N Say “yes” Internal response False Alarms (response “yes” on no-signal trial) Probability density Criterion N Say “no” S+N Say “yes” Internal response Signal Detection: Sensitivity and Bias We can estimate two parameters from performance in this task: Sensitivity: Ability to detect. Good sensitivity = High hit rate + low false alarm rate. Poor sensitivity = About the same hit and false alarm rates. Response Bias: Willingness to say you detect. Can be liberal (too willing) or conservative (not willing enough). For example, if the true signal to noise ratio is 50% and you have a 75% detection rate, then your response bias is to be too liberal. Signal Detection: Sensitivity and Bias Computing sensitivity or d’ (“d-prime”) Is a measure of performance (like percent correct, or response time) Typical values are from 0 to 4 (greater than 4 is hard to measure because performance is so close to perfect) A d-prime value of 1.0 is often defined as threshold. d-Prime d-prime is the distance between the N and S+N Probability density distributions d-prime is measure in standard deviations (Z-Scores) In SDT, one usually assumes the two underlying distributions are normal with equal variance (i.e., both curves have the same standard deviation) d’ N S+N Internal response Signal Detection: Sensitivity and Bias Computing bias: The criterion is the point above which a person says “detect.” It can be unbiased (the point where the distributions cross; 1.0), liberally biased (< 1.0), or conservatively biased (> 1.0). Signal Detection: Sensitivity and Bias Since sensitivity and bias are independent, you can measure the effect of different biases on responding to a particular value for detectability. Influences on bias: Instructions (only say “yes” if you’re absolutely sure). Payoffs (big reward for hits, no penalty for false alarms). Probability of signal (higher probability leads to more liberal bias). Signal Detection: Sensitivity and Bias Receiver operating characteristic (ROC) curves: For a given detectability value, you can manipulate the hit and false alarm rates. An ROC curve shows the effect of changing bias for that level of detectability. Sample ROC Curves % of Hits Proportion H Very sensitive observer 1 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0 Moderately sensitive observer Zero sensitivity 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 Proportion FA 1 Zero Medium High Optimal Performance Depending on the probability of a signal trial and the payoff matrix, the optimal placement of the criterion will vary. p(N) opt = value (CR) - cost (FA) X p(S) value (H) - cost (M) You can compare performance to the ideal observer to assess the operator. Examples of Visual Search Is there a threat? Where’s Waldo? Search How do you use attention to locate items in a complicated array? Two kinds of search: Feature Search and Conjunction Search. Feature search: A single feature allows you to find the item you are searching for. Find the blue S. Search X X T T T T X T X S X X T X X T Search X T X T T T X T X T X X T X T T T X S T X X T X X X T X T X T X T T T X T X T X line (blob) orientation Julész & Bergen 83; Sagi & Julész 85a, Wolfe et al. 92; Weigle et al. 2000 curvature Treisman & Gormican 88 length, width closure Sagi & Julész 85b; Julész & Bergen Treisman & 83 Gormican 88 colour (hue) density, contrast Nagy & Sanchez Healey & Enns 98; 90; Healey 96; Healey & Enns 99 Bauer et al. 98; Healey & Enns 99 size Treisman & Gelade 80; Healey & Enns 98; Healey & Enns 99x Search How do you use attention to locate items in a complicated array? Conjunction search: You have to combine features to find the item you are searching for. This should take attention and be more difficult (Treisman, 1988). Find the green T. Search X X T T T T X T X T X X T X X T Search X T X X T X T T X T X X X T T XT X T X TT TT XX TX T X X XX T X T T X T X X T T T T X T X T T X T X T X T Simple feature search Look for an “O” T T T T T T O T T T T T T T T T T T T T T T TT T T T T T T T T T T Simple feature search Look for something red O T T T O T O T T T O T O T O TT O O T T O T T O O O T O O T T Conjunctive feature search Look for something red AND “O” O T T T O T T T O T T T O T O T O O T O T T T O T O T O T O O O T OT O T Response Time Conjunctive Search Simple feature search Number of Stimuli in Display Properties of searches: Feature searches: Conjunction Don’t require searches: attention (popout). No help from location cueing (don’t need it). Require attention. Affected by the number of distracters. Helped by cueing the location. Pop-Outs in Advertisement Scan Paths Feature Conjunction: Attention as Glue ~ The significance between conjunctive and disjunctive searches is that it means that individual features like color and size are loaded pre-attentively (attention is not required), but a conjunctive search requires attention to bind the two features to the object to a location in space. You need attention to know an object is both red and large and where it is. ~ The integration may happen in the visual cortex as a result of synchrony, with attention affecting the tuning properties of sensory neurons, and preparing other cognitive processes like working memory. Attention as Glue Keep your eye on the fixation point below. A screen with colored letters will be briefly flashed. Try to remember as many letters with their colors as you can. + L M T Q S P Q B O H U X V Z K Attention as a Glue What color was the X? Do you distinctly recall a particular letter being a different color? How did that happen? How did a color in one location get associated with an object in another location? This is “attention as a glue” Treisman’s Feature Integration Theory – A two-stage theory of visual attention. Stage 1. fast parallel for single features Stage 2. Slow serial for conjunctions of single features. Several primary visual features are processed and represented with separate feature maps that are later integrated in a saliency map that can be accessed in order to direct attention to the most conspicuous areas. A parallel search, a red circle amidst green circles, takes no time no matter how many green circles (it’s cheap). A serial search, with conjunction features, like red circles amidst black circle and red triangles, requires you to check each distractor serially. Automatic Processing After practice, some tasks no longer require attention. Three criteria for automatic tasks: Occur without intention. When the load is low Required reaction times are short The tasks are “over-learnt” or well-practiced No conscious awareness/Can’t be introspected. Don’t interfere with other activities. Fast processes -- the brain does them ‘automatically’, they are a basic feature You can tell how the process of automatization is going by doing dual task studies (primary and secondary). Automatic Processing Read the Words. Say the colors Which is harder? Automatic Processing You did the Stroop task. The interpretation is that you automatically read the word. If that’s the task, the color doesn’t interfere because you don’t automatically register that. If you’re supposed to name the color, automatic reading messes you up. Filtering So, thresholds shift…but once set, then what? What happens when something gets over the threshold wherever it is? When does meaning become involved? How do we choose what to attend to? Is the choice made early or late? Themes Early or Late? In other words, does something get chosen before or after (respectively) the “stimulus gets stamped with meaning” What is attention? Some sort of bottleneck or filter? A capacity or resource (or several kinds)? Can we learn something by looking for it in brains? Filtering Attended Sensory Store Unattended Early: Broadbent. Selection happens at the filter and sensory store before pattern recognition. The selection is made at the EARLY STAGE of crude physical analysis. Filter Pattern Recognition Selection Shortterm memory Filtering Early: Evidence: “7-4-1” Dichotic listening. Two messages, one to each ear, played simultaneously. Shadowing: Repeat out loud everything in one ear. What do people (or what don’t people) notice in the unattended ear? Miss change of speaker. Miss change of language. Miss change of direction. “3-2-5” Filtering Early: Evidence: “7-4-1” Filter flapping: Two sets of numbers come in, one set in each ear. Report by ear: Easy. Report in order: Hard. The argument is that the filter lets in all of one channel, then the other, no problem. To switch back and forth takes a lot of effort. “3-2-5” Filtering Problem for early models: People detect their name on the unattended channel (cocktail party phenomenon). Treisman (1960): If a shadowed story switches ears, people follow it, and then correct. They have to be attending to meaning to follow the story. Filtering Problem for early models: Example 1: …I SAW THE GIRL/song was WISHING… …me that bird/JUMPING in the street… Example 2: …AT A MAHOGANY/three POSSIBILITIES… …look at these/TABLE with her head… Filtering Attended Sensory Store Filter Pattern Recognition Selection Shortterm memory Unattended Attenuation model: Everything in memory is active at some resting level. Some stuff that’s important has a high resting level, making it easier to respond to (e.g., your name). Other stuff has a low resting level, making it harder to respond to. As you think about something, you raise its activity level. Filtering Attenuation model: The unshadowed ear is attenuated (the volume is low). This little bit of attention can reach something with a high resting level (your name, a story you’re shadowing), but not some random bit of information. So, no filter, just attenuation. Filtering Capacity model: You have a certain amount of attention, you can spread it around as needed. If you spend a lot on one task, then you have less for others. Primary task: Do well on this no matter what (main focus of resources). Secondary task: Also do this. By manipulating the difficulty of the primary task and measuring the secondary task, we can see how attention allocation affects performance. Filtering Capacity model: For example, Johnston and Heinz (1978) had two tasks: Primary: Shadow one ear for a change that is easy (gender) or a change that is hard (category). Secondary: Detect a light. Filtering Capacity model: Johnston and Heinz (1978) Primary Secondary Shadow one list 1.4% error (control) Easy (gender) 5.3% error 310 ms Hard (category) 20.5% error 482 ms 370 ms Filtering Capacity model: What this implies is that the filter can be early (gender) or late (category), the amount of your resources that you allocate to it determines where the filter is. Emotion Driving Attention • Detecting a Snake in the Grass (Ohman, Flykt, Esteves, 2001) • Stimuli; snakes, spiders, mushrooms, flowers • Presented in 2x2 or 3x3 displays • Task; “Do all the pictures belong to the same category?” Emotion Drives Attention • Reaction time to detect target in ms. Fearful Neutral 2x2 950 1010 3x3 950 1010 Emotion Drives Attention • The Emotional Stroop Effect – You are slower at naming a color of emotionally charged words than neutral words • Taboo words vs neutral words Emotion Drives Attention • Classical Stroop RED GREEN BLACK YELLOW BLUE RED BLACK RED GREEN BLACK YELLOW BLUE RED BLACK Attic Bitch Shit Anus Frame Dyke Senate Note Bank Queer Scrotum Wife Emotion Drives Attention Emotion Drives Attention • Emotional Stroop effect occurs with Taboo words Alcohol words (beer) in alcoholics Smoking words (cigarettes) in smokers Spider words (web, crawl) in arachnophobics Food pictures in females with anorexia Threat words (disease) with people with anxiety disorders The Dot Task: Detection in PTSD Mood and Attention: Levels of Focus (Gasper & Clore, 2002) • Hypothesis: Affective cues may be experienced as task-relevant information, which then influences global versus local attention. • Mood Manipulation: Subjects randomly assigned to write about a happy or sad event in their lives • Each participant randomly assigned to a drawing chain, where the first person in each group saw a drawing of an African shield with the title of “Portrait of a Man.” In a later session, a 2nd person saw the 1st person’s reproduction from memory, and so on. Local Global Lesions in LH produce deficits in local perception Lesions in RH produce deficits in Global Mood and Attention: Levels of Focus (Gasper & Clore, 2002) • Happy Mood condition more likely than Sad Mood to contain schema-relevant details like title and facial features • Sad Mood drawings became less face-like down the chain but not Happy Mood drawings • Sad Mood drawing looked less like original Mood and Attention: Levels of Focus (Gasper & Clore, 2002) • Experiment 2 employed a task in which the same objects were sometimes the global and sometimes the local stimulus (Kimchi & Palmer, 1982). Participants saw an overall shape (e.g., a triangle) made up of smaller geometric figures (e.g., triangles). Their task was to indicate which of two other figures (e.g., a square made of triangles or a triangle made of squares) was more similar to this target figure. Result: Sadder people base their decisions on the local features, and report doing so. Posner Cueing Task Central cue peripheral cue cue ISI target Central Cue condition triggers endogenous attention / voluntary attention Top-down Peripheral Cue condition triggers exogenous attention/ Reflexive attention Bottom-up Inhibition of Return The “Been There, Done That” Reflex We are faster at unpredicted cues after a long enough pause IOR prevents going over the same ground, promotes searches for novel stimuli • The findings from patients with brain damage led Posner to construct a model for attention that involves three separate mental operations: • Disengaging of attention from the current location • Moving attention to a new location • Engaging attention in a new location to facilitate processing in that location. Psychological Refractory Period (PRP) Timing the Central Bottleneck A: Multiple sensory input B: Serial Decision maker C: Multiple action output Stimulus 1 RT ms Stimulus 1 Stimulus 2 A A SOA Stimulus 2 Slope = 1 25 150 400 SOA 900 B Time C PRP B C Embodied cognition of attention: is Cognition Time-Pressured? • If we were designed to perform under pressure, we would be good at it. • But, the reality is that, under time pressure we fall apart • We actively avoid operating under conditions where we are time pressured • Most of daily behavior consists of mundane, routine behavior Embodied Cognition is Time Pressured Wilson, 2002 • Summary – Perceptuomotor processes are time-pressured, but that does not illuminate cognitive performance under time pressure – Difficult to interpret whether “cognition is time pressured” means we evolved to perform under pressure or that our cognitive abilities must be understood in the context of coping with (unsuccessfully) or avoiding time pressures Line Bisection and flower drawing are examples of spatialbased neglect. The dumb-bells are an example of object based neglect IMPLICIT ASSOCIATION TEST CATEGORY SWITCH Insects Bad O O O Flowers Good Get Kristin's wonderfuldemo Roach O O nasty O O Daisy O O joyful O O Tulip O O terrible O WORD CATEGORIZATION Flowers Bad O O O O Insects Good O O wonderful Roach nasty O Daisy O O joyful O O Tulip O O terrible O IMPLICIT ATTITUDES 25 Number of Items Correctly Classified 20 15 10 5 0 Insects + Bad Insects + Good IMPLICIT BELIEFS 2000 Reaction Time 1500 1000 500 0 Insects + Good Insects + Bad Stereotype Threat (Beilock & McConnell, 2004) People perform in compliance with stereotypes In one of the first studies on stereotype threat, Steele and Aronson (1995) had high-achieving African American and Caucasian students at Stanford University complete a portion of the graduate record exam (GRE). Prior to doing so, some students were told that the test was diagnostic of intellectual ability whereas others were told that the test was a laboratory problem-solving task not diagnostic of intellectual ability. Stereotype Threat (Beilock & McConnell, 2004) Results demonstrated that after controlling for SAT scores (to equate past academic performance), there was no difference in GRE performance between White and Black students for whom the test was not framed as diagnostic of intellectual ability. Of those students who were told that the test was diagnostic of intellectual ability, however, African Americans performed significantly worse than Caucasians. Steele and Aronson argued that informing students about the diagnosticity of the test activated the negative cultural stereotype that “Blacks are not as intelligent as Whites,” which contributed to the less-than-optimal performance of African Americans on a test assumed to gauge intelligence. Stereotype Threat: Memory or Attention (Beilock & McConnell, 2004) How does stereotype threat work? One proposal is that stereotype threat produces reduces working memory capacity But past research has shown that stereotype threat effects are most pronounced for expert athletes, for whose abilities are highly proceduralized, relying little on working memory. On the other hand, expert athletes “choke” when they start to pay too much attention to the steps of their automatized processes; this increased attention can backfire and disrupt what should have been fluent, proceduralized execution. This idea is often termed the “explicit monitoring hypothesis.” Stereotype Threat: Memory or Attention (Beilock & McConnell, 2004) Do stereotype threats reduce working memory capacity, or do stereotype threats prompt explicit monitoring of automated procedures Expert male golfers perform a putt, before or after hearing a stereotype (“men are poorer putters than women”) or receiving control information (“putting performance differs as a function of skill level”). Experts who received stereotype did worse Stereotype Threat: Memory or Attention (Beilock & McConnell, 2004) Now how to determine it is attention? Introduce a dual-task Experiment 2 (Beiock et al, 2003) Two groups with stereotyped and non-stereotyped One group performs putting alone Second group putts while listening for target word Results. Performance was the same for putters in the dual & single task who had no stereotype threat. For putters with stereotype threat, performance was better in dual-task condition (Stereo-type Threat affects attention, not memory) The End Quizz You drove home and did not stopping at the store. a) This was due to a search failure because the sign for the store did not pop-out. b) You have a lot on your mind, and you are easily distracted c) Going to the store is a conscious decision, but you were filtering based on perceptual features d) Driving home is an automatic process e) Driving home is a conditioned response Quizz You’re walking to class and thinking about a quiz that’s coming up. Someone calls your name, but you don’t hear them. a) Your ROC curve is high. b) This counts as a hit c) You are filtering for perceptual features d) You are filtering for categorical or semantic information e) You didn’t study and you can’t hear people while throwing-up Concentration Our last topic has to do with the task of “paying attention.” Sometimes you have to concentrate on something in which you have no interest. Sometimes you have to not think about something in which you have an interest. Concentration Wegner, Schneider, Carter, and White (1987). Try not to think of a white bear. Five minutes, measure the number of times people do it. Or, try to think of it. Both are hard, with less activity later on. Concentration Wegner, Schneider, Carter, and White (1987). After suppression, it’s easier to keep thinking about a white bear. After expression, it’s still hard not to think of a white bear at first, but people adapt. Embodied cognition of attention: is Cognition Time-Pressured? • Cognition happens in “real-time” or “runtime” – Cognition “must cope with predators, prey, stationary objects and terrain as fast as the situation dishes them out.” – How do you get robots to think about walking on uneven terrain, or to swing from branch to branch, or looking around a crowded room looking for a soda without bumping into something – Story of legalizing sightdogs. Embodied cognition of attention: is Cognition Time-Pressured? • Examples for: – Skilled hand movements, or time-locked perceptuomotor activity (catching, throwing, tying, walking) – Inhibition Of Return – Exogenously driven attention • Examples Against: – PRP – Task-switching – Trade offs between speed and accuracy in attention Quizz You are looking for a friend at a party. This person has brown hair and is very tall. a) You are performing a serial search, that will be affected by the number of people there b) You are performing two serial searches, and the person will “pop-out” c) You are performing a conjunction search which will be affected by the number of people there d) You are performing a conjunction search and the person will “pop-out” because there is nobody else there with those two qualities. Quizz A. B. C. D. The Titanic hitting an iceberg would be a pretty good example of a hit miss false alarm correct rejection Quizz Bob, an electrician, is trying to see how faint he can make a light. He starts by turning the light ON to its maximum, then turning it down until he cannot see it. a) Difference detection; methods of limits; ascending b) Difference detection; method of constant stimuli; random c) Absolute thresholds; method of limits; descending d) Absolute thresholds; constant stimuli; random Quizz In Signal Detection Theory, which of the following is not true: A. attention requires more hits than false alarms B. there is a normal distribution for signal and one for noise with the distance apart measured in Zscores C. d-prime measures the difference between signal and noise D. bias and sensitivity are independent