Moral Babies: Preverbal infants know who and what is good and bad
J. Kiley Hamlin UBC Psychology
Automatic Evaluation
“Big bully” “Blinded by rage and frustration” “The villain”
“Sweet circle” “Lovers” “Innocent Young Things” Bad guy. “Our hero” Good guys.
Triangles and circles??
Heider & Simmel, 1944 (See also Haidt, 2001 for evidence for “intuitive” moral judgments) AutomaticHow do we Evaluation do this?
• Surely some is learned
• but which type of experience?
• deliberate, slow = result of teaching and cognitive development?
• easy, fast = based on every day observations and experiences in the world?
resulting automaticity/intuitive nature in adults due to practice (driving)
• Perhaps some aspects built-in?
My pursuit:
• Determine what, if anything, is built-in and/or develops in the absence of candidate required experiences • language, teaching, enculturation • specific personal experience • Examine a population with much less of these • To what extent do infants’ evaluations map onto adult moral judgments? Basic methodology:
3 character puppet shows: Protagonist, Helper, Hinderer
Goal: Reach hilltop Obtain object Obtain object
Problem: Too steep Someone else has it In box, can’t lift lid
Helpful Act:
Unhelpful Act:
Preference for Helpers over Hinderers?
• Depending on age/study, habituate or familiarize infants to helper/hinderer events
• We then examine their preference (choice or attention)
• moms close eyes
• choice experimenter blind to identity Some preference results:
Helper Hinderer 100 ** ** ** ** 75 *
50
25 Preference: % Infants 0 Hill: 6 months choice Hill: 3 months looking Box: 5 months choice Ball: 5 months choice Ball: 3 months looking
e.g., Hamlin, Wynn, & Bloom, 2007, Nature; Hamlin, Wynn, in Bloom, 2010, Dev Sci; Hamlin & Wynn, 2011, Cog Dev
But how are they doing this?
Helper = good because intends to facilitate what another intends
Hinderer = bad because intends to block what another intends How are they doing this?
• Helping and hindering not differentially familiar or expected • No looking time differences to helpful vs unhelpful events • But they do seem to react positively to helping and negatively to hindering • Blind coders correctly predict what kind of event infants just watched based on their facial expressions (5 & 6 mo conditions)
Hamlin, Wynn, & Bloom, in prep
How are they doing this?
But perhaps just something inherent to the stimuli?
Inanimate control conditions:
Pusher-Upper v. Ball-Giver v. Box-Opener v. • Pushing up Pusher-Downer Ball-Taker Box-Closer versus down • Ball leaving versus taking • Box opening versus closing
Nope: no preferences in any inanimate conditions: Action must be directed at an agent How are they doing this?
• Maybe they like goal-facilitators and Helper of agent = good dislike goal-blockers Hinderer of agent = bad • But adults interpret behavior in context: • e.g., mom blocking baby’s goal to reach light socket • e.g., reward and punishment • A human universal, theorized to stabilize cooperative social systems (Boyd et al, 2003; Fowler, 2005, Hauert et al, 2007; Henrich, 2006; Henrich & Henrich, 2007; Nowak & Sigmund, 2005)
Evaluating Rewarders and Punishers
• When adults see others as deserving punishment, we may positively evaluate those who perform it...do infants?
Protagonist with unfulfilled goal Help act Deserves Reward
Hindering act Deserves Punishment Evaluating Rewarders and Punishers
Prosocial Antisocial Target Target (2 separate conditions)
Helped ?
Hindered ?
Evaluating Rewarders and Punishers: Results
Helper Hinderer ** ** ** NS 100 ** ** ** ** ** ** 75
50
25 Choice - % Infants 0 Prosocial Target Antisocial Target Prosocial Target Antisocial Target Prosocial Target Antisocial Target
19-month-olds 8-month-olds 5-month-olds
Hamlin, Wynn, Bloom, & Mahajan, 2011, PNAS How are they doing this?
By 8 months, Helping good guys = good Hindering bad guys = good • Evaluating in context! • At least resembles adult intuitions of deservedness/reward/punishment • But maybe 8- and 19-month-olds are just valence matching?
Valence-matching?
Causal role in valenced acts critical for evaluation: Only moral agents deserve punishment, not patients If they’re valence matching, they should also prefer hinderer of hindered target
They don’t. (they like the helper)
Receives bad Does bad behavior; behavior; does not deserve deserves punishment punishment So...how are they doing this?
By 8 months, Helping • Resembles adult intuitions of good guys = good deservedness/reward/punishment Hindering bad guys = • But maybe evaluations based on good analyzing outcomes only • A first-order goal account: infants analyze the goal of the Protagonist only; like those who bring it about (if P = good) and dislike those who prevent it (if P=bad) • In particular, no intention to help/ harm on the part of the Helper/ Hinderer
Intention v Outcome?
• Even within a social analysis that incorporates contextual info, infants could evaluate:
• The outcome only (causes Protagonist to get goal (+) or not (-))
• The intention only (intends to be nice (+); or mean (-))
• Adults’ judgments incorporate both; kids tend to rely on outcome (e.g., Piaget, 1932, a zillion others)
• Some evidence infants capable of distinguishing intention from outcome:
• in analysis of others’ goal-directed acts (e.g. Brandone & Wellman, 2009; Carpenter et al, 1998; Hamlin et al, 2008, 2009; Meltzoff, 1995)
• in evaluations of first-party interactions
• Failed attempts to help (Behne et al, 2005, Dunfield & Kuhlmeier, 2010) Failed Attempts: Puppet Shows
Failed Helper Failed Hinderer
Failed Attempts: Design
• 5- and 8-month-olds
• Set 1: Can they use intention only? (Outcomes equal, intentions differ) Failed Helper versus Hinderer Failed Hinderer versus Helper • Set 2: Intention versus outcome? (Intention pitted against outcome) Failed Helper versus Failed Hinderer
• Set 3: Do they use outcome only? (Outcomes differ, intentions equal) Failed Helper versus Helper Failed Hinderer versus Hinderer Failed Attempts: Results; 8-mo-olds
* Nicer/Better Response Meaner/Worse Response 100 ** ** **
75 Using Intention NS NS Privileging Intention 50
25 Not Using Outcome Choice - Percent Infants
(H+) (FH-) (FH+) (H-) (FH+)(FH-) (H+) (FH+) (FH-) (H-) 0 [Help v Failed Hinder] [Failed Help v Hinder] [Failed Help v Failed Hinder] [Help v Failed Help] [Failed Hinder v Hinder]
Outcomes both good Outcomes both bad Outcomes oppose Intention Intentions both good Intentions both bad
8-month-olds, n=16 per condition Hamlin, under review
Failed Attempts: Results; 5- and 8-mo-olds
Nicer/Better 100 Meaner/Worse (replicated 5-month-old success with * * * Successful Helper v Successful Hinderer) 75 NS NS NS NS NS NS NS 50
Choice - Percent Infants 25
0 [Help v Failed Hinder] [Failed Help v Hinder][Failed Help v Failed Hinder] [Help v Failed Help] [Failed Hinder v Hinder] Outcomes both good Outcomes both bad Outcome opposes Intention Intentions both good Intentions both bad n=16 per condition Hamlin, under review Failed Attempts: Conclusions
• (Also have a control for movement - you can ask me later) Positive intention = good Negative intention = bad • 8-month-olds uniquely use intention
• 5-month-olds need both
• Neither uniquely utilize outcome
• although characters didn’t play a direct 8-month-old causal role in outcomes here - that would be accidental help/harm
• But they are definitely not only going with outcome, nor merely evaluating goal blocking/facilitating
But...adults are still more mentalistic!
• Giving Brussels Sprouts = Good?
• Depends who you ask.
• Prosocial/Antisocial acts don’t have to look any different
• So, you need to know something about the mental states of the target (like their preferences) to know what is helpful/unhelpful for him/her
• If you don’t know, you can’t really be praised/blamed
• AND preferences tend to be relative: I like brussels sprouts, but I LOVE chocolate Preference Facilitation Study: Methods
• 10-month-olds • Phase 1: 2 elephants watch as Protagonist grasps the same toy 4X (which one counterbalanced) • Phase 2: Protagonist loses access to toys • Phase 3: One puppet gives access to toy previously grasped, other gives access to toy previously ungrasped
Preference Facilitation Study: Puppet Shows
Preference Events repeated 4 times; objects switch each time Preference Facilitation Study: Puppet Shows
Lost Access Event
For half the babies the previously-grasped object switches sides, for half it stays
Preference Facilitation Study: Puppet Shows
Door-opening events
Grasped object always where was during the Lost Access Event, only one puppet on stage at a time Preference Facilitation Study: Design
• Preference-Knowledge (previously shown)
• Grasping one toy and not other (preference inferrable)
• Door-Lifters watch (have knowledge of preference)
• Preference-NoKnowledge
• Grasping one toy and not other (preference inferrable)
• Door-Lifters absent (no knowledge of preference)
• NoPreference-Knowledge
• Grasping only toy available (preference not inferrable)
• Door-Lifters watch (have knowledge of (lack of) preference)
Preference Facilitation Study: Design
Different accounts make different predictions: • Lowest-level (brute physical cues present in stimuli): • Prefer Grasped-Door-Lifter in none (because actions same) or all 3 (if there’s something we’re missing)
• Low-level (end-state facilitation w/o mental states): • Prefer Grasped-Door-Lifter in all 3 • Low-Mid-level: First-order goal attribution: • Prefer Grasped-Door-Lifter in Preference-Knowledge and Preference-NoKnowledge • High-Mid-level: Second-order goal attribution without knowledge/ignorance of goal (assume that if they themselves know goal, everyone knows goal - non-representational mentalistic account):
• Prefer Grasped-Door-Lifter in Preference-Knowledge and Preference-NoKnowledge • High-level mentalistic: Second order goal account that includes knowledge/ignorance of goal (representational mentalistic account):
• Prefer Grasped-Door-Lifter only in Preference-Knowledge condition Preference Facilitation Study: Results
Grasped Door-Lifter Ungrasped Door-Lifter * 100
* 75 Supports fully mentalistic account!
ns ns 50
25 Choice - Percent Infants
0 Preference-Knowledge Preference-NoKnowledge NoPreference-Knowledge n=26 n=26 n=26
Hamlin, Ullman, Tenenbaum, Goodman, & Baker, in revision
Interim Summary/Conclusions
• By 3 months of age, infants prefer helpers to hinderers • By 5 months, they react more positively to helping than hindering • By 8-10 months, they: • evaluate behavior in context • privilege intention • incorporate helpers’/hinderers’ representational mental states • This is certainly adult-like! So what’s with development?
• (At least) 2 big unanswered questions:
1. What’s with kids < 8 mos? Domain-specific or domain-general failures?
• maybe they don’t use context, intention, etc
• or maybe they’re just generally limited
2. What’s with kids failing at a whole slew of things like this??
• Perhaps the failure is with implicit/explicit knowledge and/or verbalization
• Many have suggested this for asymmetries in infants’ v. children’s performance in other domains, like object permanence and theory of mind
Beyond forced choice
• Young infants can’t do much more than look at stuff and choose between puppets • But toddlers can do much more! • In the second year: • Understand many words (most don’t speak much) • Give things • Take things away Adults use moral language
• Don’t have to infer toddler’s conceptual representations of helpers and hinderers
• just ask them!!
• "Hey (baby), Do you remember the GOOD guy? Can you find the GOOD guy? See these guys? Show me the GOOD guy."
• 18-month-olds
• Study 1: Hill, distinct intonations, varied the words, between-subjects
• Study 2: Ball, same neutral voice for both, all “good/bad,” within-subjects
Language Study: Results
Effect with mean may be weaker, but certainly distinguishing Helper them (possibly don’t like touching mean) Hinderer 100 ** ** ** **
75 ns ns
50
25 Choice - Percent Infants
0 Study 1 Nice Study 1 Mean Study 2 Nice Study 2 Mean n=8 n=8 n=9 n=11
18-month-olds Hamlin, Wynn, & Bloom, in prep Actual Reward and Punishment
• Toddlers can follow simple directions • They can give things and take things away • Infant studies suggest they should interpret these as valenced actions • Do they give to good puppets and take from bad puppets?
Reward and Punishment: Method
• 21-month-olds • 3 steps: 1. Giving game 2. Puppet show 3. Reward OR Punishment induction: “Oh no! There’s not enough treats left!” Reward and Punishment: Method
2 separate conditions:
$ Reward – Give a treat Punish – Take a treat
There’s only one treat left! There’s a puppy who didn’t get a treat!
Reward and Punishment: Results
Helper Hinderer 100 **
** ** 75
50
25
Target of Behavior - Percent Infants 0 Reward Condition Punish Condition n=16 n=16
Hamlin, Wynn, Bloom, & Mahajan, in press, PNAS Valence Matching Control: Results
Helpee Hinderee 100 **
NS * 75 Actually avoid taking from guy 50 who was hindered before! 25
Target of Behavior - Percent Infants 0 Reward Condition Punish Condition n=16 n=16
Hamlin, Wynn, Bloom, & Mahajan, in press, PNAS
Reward and Punishment?
• This is more than forced preference
• But not much more - we’re entirely scaffolding their behaviors
• They must decide that taking is negative on their own, but we kinda already knew that (even 3-month-olds dislike takers)
• So...in what sense are they motivated to punish? Some anecdotal evidence= (very bad):
Motivated to reward/punish?: Methods
1. Giving game - practice giving cars back and forth to “Mr. Lion”
2. Puppet show:
• Protagonist (boy) tries to get toy off high shelf
• Helper (girl with blond or brown hair) grabs toy and drops to boy
• Hinderer (other girl) grabs toy and runs offstage
3. Puppet shows preference through statistical sampling (e.g., 22% ducks, 78% frogs in a box, picks up only ducks, as in Kushnir, Xu, & Wellman, 2010
4. Puppet asks toddler to “give me one” up to 3 times
Are toddlers more likely to If the unhelpful puppet asks, give, and to give a preferred will they refuse to give, or give toy, if the helpful puppet asks? an unpreferred toy? Motivated to reward/punish?
Nice$Blond Mean$Brown
Nice$Brown Mean$Brown
Motivated to reward/punish?: Results
• About 75% give right away in response to first request Both Toys • Infants (who give) are not slower in hinderer condition • Suggests refusal because they don’t want to give, not because Preferred Toy inhibited around mean guy Dispreferred Toy • Infants across conditions aren’t slower to give disliked toy Neither Toy 20 ** * 15
10
5
Target of Giving - # of infants 0 Helpful Puppet Requests Unhelpful Puppet Requests n=25 n=22 Hamlin & Kushnir, in prep Toddler Studies: Summary & Conclusions
• They map the adult labels for good and bad onto puppet show characters @ 18 months
• They direct their own giving behavior toward helpers and taking behavior toward hinderers @ 21 months
• Something that isn’t a forced puppet choice scenario, they selectively give helper the toy helper (apparently) likes; do not do so for hinderer @ 21 months
Overall Conclusions
• Suggests that some of the tendencies we generally attribute to a sophisticated moral reasoner/actor are already operating in infancy & toddlerhood
• Perhaps adult intuition/automaticity is due to these evaluations being built in - even the ones that seem pretty complex (So) many unanswered questions
• What’s built in?
• Emotional responses? Cognitive rules about what’s good/bad?
• Individual differences?
• What’s going on with those 2 kids who choose the mean guy? Not getting it? Actually liking bad?
• Is this a stable difference that persists across the lifespan?
• Relation between moral judgment and moral action?
• Reward/punishment are moral actions, sort of, but they rely critically on preceding moral judgments
• What is the underlying definition of a good act? Does it incorporate group membership? Kin psychology?
• (A growing amount of) preliminary evidence says it does...even in babies
Thanks!
• Parents and Infants in • RA’s at UBC, Yale Vancouver and New Haven, CT • Janine Gellerman • Collaborators • Zoe Liberman • Karen Wynn • Annie Spokes • Paul Bloom • You • Tomer Ullman • Josh Tenenbaum • Noah Goodman • Chris Baker • Neha Mahajan • Tamar Kushnir