how do we understand the behavior of others?: the agency system clark barrett ucla...
TRANSCRIPT
What is agency?
• Agency = the capacity to act in a goal-directed (intentional) way
• Humans use the “intentional stance” (Dennett 1987) to interpret and make predictions about behavior.
• Main question: What cognitive mechanisms allow us to do this?
Talk outline
• Research goal: searching for building blocks of the agency system
1. Perceptual templates / schemas
2. Conceptual schemas
3. Switching the system on and off
4. Interactions with other systems: agency and social cognition
• Concluding speculations
Why should you care?
(Why is the agency system important for culture, norms, and evolution?)
• A) A security guard fails to detect a terrorist 50 people die.
• B) A security guard throws a grenade into a plane full of passengers 50 people die.
• Perhaps understanding intentions gets you more than just looking at outcomes?
(Intentions in behavioral econ: Blount, McCabe)
A “black box” account(non-computational)
Perceptual cues
Behavior predictions
INPUTS“Theory of mind”,
“Belief / desire reasoning”, “Intentional
stance”
OUTPUTS
Knowledge Judgments & Decisions
?What’s in here?
Inside the black box(a modular, or computational, account)
Perceptual cues
Predator-prey schema
Social Xch schema
Etc.Behavior prediction
s
INPUTS
Conceptual schemas
OUTPUTS
Scope restrictors / modifiers
John
Susan
Jim
Human
Dog
Lion
IndividualsTaxa
Processor (“ToMM”)
Knowledge
Perceptual
templates / schemas
Judgments &
Decisions
(AD, ID)
Inside the black box(a modular, or computational, account)
Perceptual cues
Predator-prey schema
Social Xch schema
Etc.Behavior prediction
s
INPUTS
Conceptual schemas
OUTPUTS
Scope restrictors
John
Susan
Jim
Human
Dog
Lion
IndividualsTaxa
Processor (“ToMM”)
Knowledge
Perceptual
templates / schemas
Judgments &
Decisions
Perceptual templates
• Achieving (many) goals entail certain kinds of motion: e.g., pursuit
• If intentions have motion signatures, “templates” can be made: detectors
• Does the mind contain such templates for detecting particular kinds of intentional behavior?
An experimental study of intentional motion perception
Todd, Barrett, Miller, & Blythe
• Question: can people reliably use motion to infer the intentions of agents, and categorize them?
• Categories: Pursuit, court, lead / follow, guard, fight, play• Generated by German adults in game context, evaluated
by second set of judges (free descriptions, forced choice)
• Within categories, no two exemplars alike; must rely on abstract qualities to make judgment
6-category study: German adults
00.10.20.30.40.50.60.70.8
0.9P
erc
en
t s
ele
cte
d
True category
Selected category
German 3-5 yr olds
Chase
Fight
Lead
Play
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
Pe
rce
nt
se
lec
ted
True category
Selected category
• But:• Are Germans just communicating a culturally shared
schema to other Germans?• Or are these motion schemas universal?
4-category cross-cultural study
Categories Shuar German
Shuar adults, horticulturalists
N=23
Berlin college students
N=40
Berlin kindergarteners
age 3-5, N=36
Chase Apapéatin Verfolgen
Fight Mániatin Kämpfen
Lead Jintíatin Führen
Play Nakurústin Spielen
Shuar and German adults
chase
fight
lead
play
00.10.20.30.4
0.5
0.6
0.7
0.8
0.9
Percent selected
True category
Selected category
chase
fight
lead
play
Motion perception: summary
• There appear to exist motion schemas for particular kinds of intention
• The same templates are present across cultures
Perceptual cues
Predator-prey schema
Social Xch schema
Etc.Behavior prediction
s
INPUTS
Conceptual schemas
OUTPUTS
Scope restrictors
John
Susan
Jim
Human
Dog
Lion
IndividualsTaxa
Processor (“ToMM”)
Knowledge
Perceptual
templates / schemas
Judgments &
Decisions
Perceptual cues
Predator-prey schema
Social Xch schema
Etc.Behavior prediction
s
INPUTS
Conceptual schemas
OUTPUTS
Scope restrictors
John
Susan
Jim
Human
Dog
Lion
IndividualsTaxa
Processor (“ToMM”)
Knowledge
Perceptual
templates / schemas
Judgments &
Decisions
Conceptual schemas
• Function: making inferences about particular kinds of intentional interaction.
• May be many, e.g. :– Social exchange, – mating, – parent / offspring, – predator / prey, – kin altruism– Fiske: relational models
• Is there evidence for them?
Predator-prey schema
Pre-contact
Predator Mutual Preydetects prey detection detects predator
() ()Approach Wait / Flee Wait / Hide Hide / Ambush
Pursuit
KeyAgents Parameters predator knowledge relation prey action relation
death
Study of children’s inferences about predator-prey interactions
Barrett, Cosmides, & Tooby
• Shuar (N=28) and German (N=38) 3 to 5 year olds• Simulated predator-prey encounter with plastic models
(Jaguar / horse; Lion / zebra)• At each stage, children predict what will happen next• Also infer mental states of predator and prey
Example question: When the lion sees the zebra, what does the lion
want to do ?• Schema-consistent:
– Chase zebra– Catch zebra– Bite zebra– Eat zebra– Kill zebra
• Inconsistent:– Go away (N=1 german)– Eat grass (N=1 shuar)
• Rest DK or no response0
10
20
30
40
50
60
70
80
90
100
Ger
man
Shuar
Ger
man
Shuar
Ger
man
Shuar
Schema-inconsistent
Schema-consistent
3 yrs 4 yrs 5 yrs
Q: When the lion catches the zebra, what will happen?
• Schema-consistent: – Lion hurts zebra– Lion kills and / or eats
zebra• Inconsistent:
– German 3yr: "then he wants to go to the hospital" (not clear if lion or zebra
– German 3 yr: "lion climbs up to the window and falls down“
• But: Few unrealistic or “fantasy” answers
• German + Shuar similar: surprising on cultural view
0
10
20
30
40
50
60
70
80
90
100
Ger
man
y
Shuar
Ger
man
y
Shuar
Ger
man
y
Shuar
Schema-inconsistent
Schema-consistent
3 yrs 4 yrs 5 yrs
0
10
20
30
40
50
60
70
80
90
100
German Shuar German Shuar German Shuar
Consistent
No response
Fantasy
Folk-psychological
Irrelevant
Violation
3 yrs 4 yrs 5 yrs
Summary of responses on predator-prey questions
Predator-prey schema: Summary
• Predator-prey schema is present at an early age across cultures
• Leads to realistic inferences about predator and prey behavior
• Similar in very different cultures, uncontaminated by cultural inputs like fairy tales and cartoons
• Other schemas?
Perceptual cues
Predator-prey schema
Social Xch schema
Etc.Behavior prediction
s
INPUTS
Conceptual schemas
OUTPUTS
Scope restrictors
John
Susan
Jim
Human
Dog
Lion
IndividualsTaxa
Processor (“ToMM”)
Knowledge
Perceptual
templates / schemas
Judgments &
Decisions
Development of social exchange schema
Barrett, Keller, Takezawa, Wichary
• Examined children’s judgments of violations of bilateral social contracts
• And predictions of reactions of different parties to violations
• German 1st and 4th graders
Children’s judgments of contract violations
0
20
40
60
80
100
friend sibling neighbor mother
Perc
en
t ju
dg
ing
co
ntr
act
bro
ken
mutualcompliance
childviolates
otherviolates
mutual non-compliance
Predicted reactions of victim of contract violation
Anger in Simon when other violates
0
20
40
60
80
100
friend sibling neighbor mother
Pe
rce
nt
att
rib
uti
ng
an
ge
r
1st graders
4th graders
Predicted reactions of violator
Guilt to Simon when Simon violates
0
20
40
60
80
100
friend sibling neighbor mother
Pe
rce
nt
gu
ilt
att
rib
uti
on
s
1st grade4th grade
Social contract schema: Summary
• Even young children can identify contract violations• Ability to predict other’s reactions in social exchange
situations may be useful for moderating one’s own behavior
• Future research: what other schemas are there? When do they schemas develop? (e.g. mating?)
Agency detection
• Some things are agents, and some are not• Agents require vigilance, and figuring out what
they are trying to do• Assuming everything is an agent entails costs• Selects for discrimination between agents and
non-agents
Agency detection activates appropriate inference systems
AGENT NON-AGENT
Activate agency system
Activate other object system
(e.g. substance)
Inferences licensed: can move, will react if touched, can hurt you… etc
Inferences licensed: can’t move, will not react if touched, can be subdivided into pieces that retain properties, etc…
Different patterns of inference:
Death as the cessation of agencyBarrett and Behne
• Hypothesis:
• Agency detection system contains a “remapping” routine
AGENT SUBSTANCE
COW STEAK
Activate agency system
Activate substance system
ALIVE DEAD
Different inference patterns
Deactivate agency system, activate substance system
Cross-cultural test of cessation of agency hypothesis
• 3 to 5 year old German and Shuar children• Sleep vs death: Animals and people• Target questions:
– Can it move?– If you touched it, could it move?– Could it hurt you?– If you made a noise, could it know you were there?– Could it be afraid?
• Sleep / death is a strong test
Patterns of inference for sleep vs death
0
0.2
0.4
0.6
0.8
1
3 4 5 3 4 5 3 4 5 3 4 5 3 4 5
Per
cen
t ye
s
Sleep
Death
0
0.2
0.4
0.6
0.8
1
3 4 5 3 4 5 3 4 5 3 4 5 3 4 5
Per
cen
t ye
s
Sleep
Death
Move? If touched?
Hurt you?Detect you?
Be afraid?GERMAN
SHUAR
Mean # correct responses by population and age
0
1
2
3
4
5
6
7
8
9
10
3 4 5 3 4 5
Nu
mb
er c
orr
ect (
of 1
0)
Human condition
Animal condition
German Shuar
Cessation of agency: summary
• Agency inferences can be switched on and off for a particular object
• This aspect of agency detection present by age 4 or earlier
• Same developmental trajectory across cultures: suggests core feature of agency system
Intentions and social contract violation
• Cosmides (1989: social contracts are agreements to exchange benefits
• Cheating = accepting benefit without paying cost
• But:
Suppose you agree to give your friend $1000 if he will give you his car next week. However, his car is stolen. Has he cheated you?
Perhaps intentions are an important part of social contracts.
Social contract Wason
Social contract rule:
“If you give me your watch, then I will give you $10”
Watch no watch $10 $5
P ~p q ~q
About 75% of people pick violation cards
Manipulating intent, incentive, ability
• Cover story manipulated so that potential violator either had:
1. Intent to violate, or violated by mistake
2. Incentive to violate, or no incentive
3. Ability to violate, or no ability (except at random)
• How do the presence or absence of these factors affect subject’s vigilance for cheaters (card turning patterns)?
Benefit / Intent / Ability: You supervise four women who volunteered to help out at the local Board of
Education. Your volunteers were supposed to follow certain rules for assigning students from various towns to the appropriate school district. Each volunteer is the mother of a teenager who is about to enter high school, and each processed her own child’s documents. You overheard that some of your volunteers intended to break the rules when it came to assigning their own children to a school. Here is the situation:
Although both communities are equally prosperous, the parents in Dover City have always cared about the quality of their schools, including Dover High, and have been willing to pay for it. In contrast, the parents in the neighboring town of Hanover have never wanted to spend the money, and have opposed any taxes to improve Hanover High. The Board of Education created this rule: “If a student is to be assigned to Dover High School, then that student must live in Dover City.”
A. Dover High School B. Dover CityC. Hanover High School D. town of Hanover
Operationalizing intent
Without intent
• Benefit / Ability
• You supervise four women who volunteered to help out at the local Board of Education. Your volunteers were supposed to follow certain rules for assigning students from various towns to the appropriate school district. Each volunteer is the mother of a teenager who is about to enter high school, and each processed her own child’s documents. You know your volunteers are honest, but you suspect that they may have made some innocent mistakes: they may have broken the rules for assigning each child to a particular school...
Operationalizing incentive, ability
• Incentive: parents sort their own students into schools, or only students of others
• Ability: students names are written on sheet, or students identified only by code numbers
Agency and social contracts: summary
• Vigilance for cheating affected ~ 20% for each factor• Additive
• The difference between intentional and accidental violation of norms may be important in many other contexts as well: killing, politeness, fairness, etc..
Summary of findings
• The agency system is not a single “ability,” but is comprised of many components
• Perceptual templates for identifying agents and specific intentions of agents
• Conceptual schemas for reasoning about intentions: provide the content for theory of mind
• Agency system can be turned on and off• Agency system can influence social attribution and
decision making processes
Perceptual cues
Behavior predictions
INPUTS“Theory of mind”,
“Belief / desire reasoning”, “Intentional
stance”
OUTPUTS
Knowledge Judgments & Decisions
?What’s in here?
Eventually, we will need a fully computational account…
(Which might or might not look something like this.)
Perceptual cues
Predator-prey schema
Social Xch schema
Etc.Behavior prediction
s
INPUTS
Conceptual schemas
OUTPUTS
Scope restrictors
John
Susan
Jim
Human
Dog
Lion
IndividualsTaxa
Processor (“ToMM”)
Knowledge
Perceptual
templates / schemas
Judgments &
Decisions
Understanding the agency system might have many implications for understanding social evolution.
How important is the ability to understand intent for the evolution of cooperation,
norms, etc?