This is going to be a discussion of 3-4 lines of research that illustrate the theme of identifying...
-
Upload
bryce-wilson -
Category
Documents
-
view
213 -
download
1
Transcript of This is going to be a discussion of 3-4 lines of research that illustrate the theme of identifying...
This is going to be a discussion of 3-4 lines of research that illustrate the theme of identifying and then using “mental representations” to explain, predict, and control decisions. The organization of the talk is loosely autobiographical … but, I’ll try to keep the focus on the behavioral science, not on the speaker.
Autobiography:“ … a man who gives a good account of himself is lying.”
“ … only to be trusted when it reveals something disgraceful.”
“ … the most respectable form of lying.”
“ … all autobiography is self-indulgent.”
“ … as common as adultery, an even more reprehensible.”
What is the essence of the cognitive approach?The “Computational Metaphor”
1. Assume that people are intelligent, goal-seeking, “calculating,” ...
2. thoughts cause behavior and computation is a good “model” of thought processes;
3. people have “mental models” of the external world inside their heads;
4. A human “cognitive architecture” is something like a typical computer’s.
A cognitive model of mental arithmetic
1. processes – subtract, carry, store-in-memory, retrieve-from-memory … (“the computational process”);
2. representations/information – numbers, locations, relations;
3. “a machine” that executes the processes (“the architecture”)
Part 1: A model of the calculation process based on “production rules”
P1: IF the goal is to do a subtraction problemTHEN the subgoal is to iterate through the columns of the problem …
… P11: IF the goal is to subtract a digit from a number and the number is of the form ‘string_digit’ and a result is the difference of two digits and the digit is less than the number 10 THEN the result is ‘string_sum’ and mark the column as processed and POP the goal …
The concept of a production system
production system notation – a kind of programming language – these are the ‘rules’ for thinking
a production rule: elementary cognitive process (the building block for bigger cognitive processes), condition-action data structure, makes cognitive processing “serial” (for a production to fire its condition must be satisfied “in working memory” and only one fires at a time), analogous to an s-r association or a neural unit (but composed of several neurons)
IF goal is to subtract 7 – 4 THEN answer is 3
Part 2: What about the mental representations?
(the “language of thought”)
- symbols and propositions- images- phonological/acoustic- grammar
n.b. there are endless empirical and conceptual arguments about these details among cognitivists …
ADDITION-FACT
FACT3+4ADDEND1 SUM
ADDEND2
THREE
FOUR
SEVEN
isa
isa
INTEGER
isa
VALUE VALUE
3 7
isa
Arithmetic Fact Chunk – This is what the inside of your head looks like
VALUE
4
Part 3: If productions are the process, what “executes” the process?
“Cognitive Architecture” – is organized like a computer – memories, buffers, cpus, …. (maybe not batteries, fans, …) n.b., here’s a major link between the cognitive analysis and neuroscience: e.g., Where’s working memory located? (in the DLPFC!)
John Anderson’s ACT-R Model (2007)
PROBLEM STATE:MENTALMODELOF THE
CURRENTSITUATION
CONTROL STATE:KEEPS
BEHAVIORALIGNED
WITH GOALS
PRODUCTION SYSTEM:WORKING MEMORY
(loosely speaking)
ACT-R 5.0
Environment
Pro
duct
ions
(Bas
al G
angl
ia)
Retrieval Buffer (VLPFC)
Matching (Striatum)
Selection (Pallidum)
Execution (Thalamus)
Goal Buffer (DLPFC)
Visual Buffer (Parietal)
Manual Buffer (Motor)
Manual Module (Motor/Cerebellum)
Visual Module (Occipital/etc)
Intentional Module (not identified)
Declarative Module (Temporal/Hippocampus)
Suppose we take the cognitive approach, how do we do research?
1. Conduct behavioral studies;
2. goal is to model the cognitive processes that occur when someone does addition;
3. measures: answers (accuracy), intermediate responses, time to respond, eye movements, brain activity … think-alouds;
4. model (within a framework) – test – revise … apply
The problem of “constraining” the model
1. More and denser behavioral data …
2. help from outside1: rationality (optimal adaptation, evolutionary selection);
3. help from outside2: neuroscience
…
Then “everything” converges at once … maybe
Some big successes for cognitive analysis
1. attention and perception (e.g., air traffic control);
2. language (spoken, heard, read);
3. classroom instruction (arithmetic, geometry, programming, …);
4. explicating “routinized” expert skills (chess, music, mnemonists, … ) …
Why is cognitive psychology necessary?
1. Often the mental representation of a situation is not a perfect, veridical reflection of the world outside the person - so a strictly behaviorist approach will be inadequate.
2. information is not processed optimally, rationally - so we can’t rely solely on rational models;
3. theoretical analyses at “adjacent” levels of scientific explanation are facilitated by having a cognitive account - e.g., recent developments in adaptive analysis (Behavioral Game Theory, Evolutionary Psychology) and in Cognitive Neuroscience;
4. demonstrations by successful application of cognitive analysis - e.g., machine tutors for mathematics, applications in user-machine interface design …
Disclaimer …
Few cognitive analyses are as complete as the mental arithmetic example – most analyses are “partial” - many explicitly address only one of the 3 parts of the full analysis; are often ‘informal’ verbal sketches of parts of the full analysis; etc. And, there is still widespread disagreement about the details of the general model, the best behavioral research paradigms, etc.
Nonetheless, the cognitive approach is the underlying conceptual paradigm in most of Psychology today.
THOMAS KUHN
HERBERT SIMON
JOHN ANDERSON
Humble Beginnings:Person Memory Networks
The goal of the research was to describe the mental representations that were created when college students “formed an impression of a (fictional) person based on sentences or short film clips depicting that person’s actions.
Belief-sampling Model for Candidate Evaluations(Hastie & Park, 1986; Tourangeau, Rips, & Rasinski, 1988, 2000;
Zaller & Feldman, 1992; Lodge, 1995 … and several others)
1. Memory store of “beliefs” - ideas, images, and evaluations associated with a (political) concept, e.g., Hilary Clinton, national health care, Iraq War, Bill Clinton, Yale Law School, …
2. A sampling process whereby beliefs are sampled from memory when an evaluation is needed;
3. An evaluation process whereby the elementary evaluations from the sampled beliefs are integrated into a summary evaluation;
4. [outside the model] … evaluation is used to take an action (e.g., to vote).
The Candidate’s Memory Representation …
Three kinds of structure: - croutons in a cognitive soup
- associative network
- arguments and narratives
Vary the structure to simulatedifferences in expertise, partisan-ship, etc.
Ancient hypothesis about an associativenetwork of ideas about Ronald Reagan (Hastie, 1986)
Ideas about the Idea Sampling Process …
- depends on the memory structure
- easy to model mathematically - sampling from an urn (“cognitive soup”), or activation of a network (graph)
- bias the sampling process to explain “context effects”
- vary extent of sampling to simulate accountability consequentiality effects
EVALUATIONINTEGRATOR
Ideas about the evaluation-integration process
- usually modeled as a weighted average
of sampled, elementary evaluations -
interpreted as an anchor-and-adjust
procedure
- many variations on the basic model,
e.g., weights depend on order, recency,
extremity of the elementary evaluation,
capacity of Working Memory, etc.
- various stopping rules, e.g., one-reason-
evaluations, confidence threshold, etc.
EVALUATOR
w1(paternal)
+ w2(cuts education loans)
+ w3(Christian)
.
.
.
= SUMMARY EVALUATION
Could this be a more general model of preference construction (of all kinds)?
Captures the notion that preferences are based on a few attributes
Captures context-dependency and instability of preferences
Question: How can the sampling and integration processes be made goal-dependent?
Question: Can the model account for classic “preference phenomena”: framing, preference reversals (procedural variance), risk attitudes, etc.?
Question: How does it need to be modified to handle on-line evaluations?
Next step … Narratives as Representations
How do jurors makesense of the complex,incomplete, contradictory, collection of evidence that they hear in a typical criminal trial?
Is it just another case of cognitive soup?
Demonstration
Listen carefully to the sentences
the speaker reads …
Try to understand what’s going on
Which sentences did you actually hear?
1. The struggling company developed a drug to treat high blood pressure.
2. The rock crushed the tiny hut.
3. The macho construction foreman gave the female welder defective oxygen tanks.
4. The huge rock rolled down the steep mountain and crushed the tiny hut.
5. The female welder went to repair the rusted roof supports.
6. The company received FDA approval for the expensive drug to treat high blood pressure.
7. Valerie and Chuck said that Fuzzy Trace Theory is boring and stupid.
Explanation-based comprehension is natural, automatic, almost irresistible
The mind glues fragments of experience together into narratives (and other situation models) and it is almost impossible to “deconstruct” them.
The most natural means of comprehending and communicating everyday experience is the construction of narrative scenarios; if people are involved, goals are the central “motors” for actions and events.
The more the fragments of experience that instantiate a familiar story template – fit our expectations – the more complete and believable the story.
Conversely, unusual stories are poorly remembered and they are not “glued together” as tightly as familiar stories.
Distinctiveness Effect: Exceptions to a coherent story are especially well-remembered.
Vividness Effect: The more “details” in the test sentence, the more likely it seems that the event occurred.
“Story Model” for Juror Decisions
Jurors [in U.S. criminal trials] reach their verdicts by solving 3 cognitive sub-tasks:
1. constructing a mental model summarizing the evidence in the form of a narrative, “story;”
2. learning the verdict categories (from the judge’s instructions);
3. classifying the story into a verdict-category.
Why do jurors reach different verdicts?They have constructed different stories
"INNOCENT STORY"
Initiating events: Afternoon quarrel
between ² and victim (Embedded episode)
Physical states: Friend comes over (² has "back up")
Psychological states: ² angry at victim
Goals: ² intends to find victim ² intends to confront victim ² intends to kill victim
Actions: ² gets knife ² searches for victim ² goes to bar ² stabs victim
Consequences: Victim dies
² arrested
"GUILTY STORY"
Victim inv ites out Victim & go outs ide Victim hits Victim has razor Victim pulls razor Victim lunges at
& v ic tim in bar Girlfriend asks for ride Victim gets angry Victim intends to do something to Victim threatens leaves bar
intends to show knife intends to protect self
Friend comes over Friend suggests go to bar wants to avoid v ictim doesn't want to go intends to not go in if v ic tim at bar Victim not at bar & friend go into bar
falls against wall woozy
carries knife by habit carry ing knife
pulls knife comes back off wall & v ic tim scuffle
Knife goes into v ic tim Victim is wounded
afraid of v ictim sees razor
CONVICT
ACQUIT
TRIAL EVIDENCE BACKGROUND KNOWLEDGE PRETRIAL PUBLICITY
Not a surprise to trial lawyers …
Defense Gambled, and Lost, With a Minimal Presentation
Lawyers for both sides told jurors that the outcome of the trial of Martha Stewart and Peter E. Baconvoic should turn on which version of events they believed. But in the end, the jurors seem to have heard only one story, that of the prosecution. … However much they criticized the government's version of the events that led to Ms. Stewart's prosecution – that she received a tip to sell her 3,928 shares in ImClone Systems, then conspired to deceive investigators about the sale – the defense never offered a complete story of its own … "The defense would've benefited, by calling a witness who could tell the entire story," said Roland Riopelle, a former federal prosecutor. (NYT, 03 06 04)
Whatever commentators may say, a trial is a struggle between competing stories. What juries require is a story into whose outline they can plug the testimony and evidence with which they are relentlessly bombarded. (Johnnie Cochran – Journey to Justice, 1997)
One key theoretical question:
Do explanations (stories) cause verdicts?
Hypothesis: If the evidence favoring one side of a case is presented in an order that facilitates the construction of one story, the side of the case, favored by the easy-to-construct story will have an advantage in the juror’s decision process.
Experimental Method
Mock-jurors listened to a tape recorded trial and render verdicts. Every mock-juror heard exactly the same evidence.
There are two sides to the case: Prosecution versus Defense (and the Prosecution always goes first).
The experimental manipulation is that within each “side” the evidence is presented in one of two orders:
“Witness Order” — the order from the original trial, which is not chronological.
or“Story Order” — i.e., chronological order
Results
DEFENSECASE
WITNESSORDER
STORYORDER
STORYORDER
PROSECUTIONCASE
WITNESSORDER
percentage of mock-jurorssaying “guilty”
59% 78%
31% 63%
The General “Explanation-based” Judgment Framework
Yet, another mental representation analysis:How do people represent conceptual categories
in long-term memory?
The classic viewis that we have “lists of essentialand “associatedfeatures … the feature lists canbe used to classifyand reason aboutcategories and their members
But, there is more than just “features” … we have “theories” about how the features are interrelated… maybe we can represent those interrelationships as “causal dependencies” in a dynamic “Bayesian Causal Network” (Rehder & Hastie, 2001; Sloman, Love, & Ahn, 1998)
Fictional categories with contrivedcausal dependencies between attribute values
Schemas representing causal dependencies - as Bayesian Networks
The causal roleof an attributewithin the largercausal schemadetermined the importance of thatattribute in categoryclassification andcategory-basedinferences
An answer to the “riddle of induction”?An explanation-based account of
inductive generalizations …
Explanation-based account for “inductive strength”
Inductive inferences are based on a two-stage reasoning process:
1. Why do the base exemplars have the novel property? (Why did these houses get burgled? Why did these particular HP products break?)
2. How widely distributed is the mechanism that produces the novel property?
Final example of “mental representation analysis”
How do people predict what’s next in a sequence of related events?
The now classic examples: “Random Events” - coins, roulette, births (gender) and athletic performance - basketball goals, football team wins, …
What’s next?
When the events are generated by a random mechanism people often exhibit an odd “Gambler’s Fallacy” judgment bias: expectation of negative recency - that a string of one outcome will be followed by “reversal” …
No 53 puts Italy out of its lottery agony Sophie Arie in The Guardian, 02 11 05
Thousands dreamed of it. Hundreds lost sleep over it. Some apparently even died for it. At last, Italy has been put out of one of its worst cases of collective lottery agony. After not showing for almost two years, number 53 was pulled out of the basket in the Venice lottery on Wednesday night. All over Italy people had placed increasingly huge bets on the elusive number in recent months, more and more convinced that it had to appear. In a frenzy that even lottery-mad Italy has rarely seen, some 53 addicts ran up debts, went bankrupt, and lost their homes to the bailiffs. Four died in 53-related incidents. A woman drowned herself in the sea off Tuscany leaving a note admitting that she had spent her family's savings on the number. A man from Signa near Florence shot his wife and son before killing himself. A man was arrested in Sicily this week for beating his wife out of frustration at debts incurred by his 53 habit. In all, more than €3.5bn (£2.4bn) was spent on 53, an average of €227 for each family. In January alone, €671.9m was spent. Although 53 had come up in other regional lotteries, it had not appeared in Venice since May 2003, and Wednesday was the 153th draw
Betting on winning lottery numbers
Does Your iPod Play Favorites?My first iPod seemed to have a fondness for Steely Dan, while other artists were sent into exile. By Steven Levy NewsweekLast spring it dawned on Apple CEO Steve Jobs that the heart of his hit iPod digital music player was the "shuffle." This feature allows users to mix up their entire song collections—thousands of tunes—and play them back in a jumbled order, like a private radio station. Jobs not only moved the popular shuffle option to an exalted place on the top menu of the iPod, he also used the idea as the design principle of the new low-cost iPod Shuffle. Its ad slogan celebrates the serendipity music lovers embrace when their songs are reordered by chance—"Life is random.“
But just about everyone who has an iPod has wondered how random the iPod shuffle function really is. From the day I loaded up my first Pod, it was as if the little devil liked to play favorites. It had a particular fondness for Steely Dan, whose songs always seemed to pop up two or three times in the first hour of play. Other songs seemed to be exiled to …
But, when the events are generated by an intentional agent, observers expect to see streaks: expectation of positive recency - that one or more ‘successes’ will be followed by ‘success’ (“positive recency”, “hot hand”, “autocorrelation”, …)
What’s next?
In sports this expectation is often referred to as “the hot hand”…
The “hot hand” expectation is also anomalous … at least in sports like basketball
Tom Gilovich and Amos Tversky conducted statistical analyses of basketball shooting (college, pros, field goals, free throws) and found that the sequences of hits and misses were indistinguishable from a probabilistic (i. i.d., Bernoulian) sequence - like a sequence of coin tosses!
Gilovich and Tversky accounted for the “hot hand” bias with the following explanation: People watch the player and conclude he is not shooting “randomly.” They conclude this because they believe coins “reverse” more frequently than they really do (“the gambler’s fallacy”). Although the player is reversing “as frequently as a coin” - this seems too streaky … so observers are too ready to conclude the player is “hot” (having a “non-random” streak of “hits”) or “cold” (having a “non-random” streak of “misses”).
N.b., this explanation is incomplete - but, for present purposes, I want to focus on the assumption that when people watch the basketball player, they begin by “testing the hypothesis” that the player is “random”
A better account?
I believe that those initial expectations about the sequence are driven by the observer’s mental model of the “mechanism” that generates the events (a mental model of a “random coin,” of a “motivated, intentional player,” etc.)
Some preliminary results with basketball fans suggest this interpretation is correct - that the initial expectation is not “random mechanism”(research with An Oskarsson, Leaf Van Boven, and Gary McClelland)
Fans’ “control” expectations are that a player is likely to be “hot” (not “random”)
4
4.5
5
5.5
6
6.5
7
Less-Hot Hotter
Player's Hotness in Video
Pe
rce
ive
d H
otn
ess
(1
-7)
Control
Recruit
Don't Recruit
The point is, again, we need a mental representation analysis to understand the
judgment phenomenon
The big challenge is to come up with a good theoretical notation with which to describe the dynamic mechanisms (we think are in people’s heads) … oh, yeah, and to conduct some behavioral experiments that verify this interpretation …
Tale of Another Italian Lottery
… This was not (the “Number 53” drought of 2005), however, the longest losing streak in the history of the Italian lotteries … In 1941 the number 8 kept people waiting for 201 draws in Rome, raising the suspicion that Mussolini was somehow spiriting the number away each fortnight to keep the bets coming in to help finance Italy's entry into the second world war … and this rumor kept the number 8 from ever becoming a big favorite of the bettors.
… The point? It’s how you explain it that matters
Summary …
1. Mental representations of other people;
2. Mental representations of evidence in legal trials;
3. Mental representations of category concepts;
4. Mental models of “sequence generating mechanisms”
In each case, I’ve tried to make a case that understanding the mental representation, gives some insights into judgments based on the representation …
… but, does the cognitive approach have any value outside of the cognitive psychology community?
The Situation: 5+ behavioral sciences: … each busily working with its own behavioral data; … different methods; … and different theories.
ECONOMICS
POLITICAL SCIENCE
PSYCHOLOGY“The Cognitive Level”
NEUROSCIENCE
SOCIOLOGY
B E H A V I O R
The remarkable fact is how independent the fields are; how they each progress with their own “private” samples of behavior … how little they interact with one another (the “behaviors” are often identical).
When and how should the fields interact “vertically” with one another?
ECONOMICS
POLITICAL SCIENCE
PSYCHOLOGY
NEUROSCIENCE
SOCIOLOGY
B E H A V I O R
Value of the Cognitive Analysis:
Elsewhere in Psychology?
To neuro-scientific analyses?
To behavioral analyses at “higher levels” … Economics, Law, Political Science?
Does the mental representation cause behavior?
Which way to Gate #8?