LE 11: Social Perception: The Illusion of Animacy and Facial Trait Inference

Interaction and the Illusion of Animacy

• The Power of Interaction: The internal sense of an object’s animacy is significantly heightened when multiple objects interact in a way that suggests they are responding to one another.

• The Narrative Constraint: Humans possess a nearly irresistible urge to assign narratives to moving shapes, even when they are merely pixels on a screen.

• Heider-Simmel Style Demonstration (Shapes Narrative)

  • The Visuals: A video featuring a large triangle, a small triangle, a circle, and a box (representative of a room with a door).

  • The Perceived Plot: The large triangle is interpreted as ‘mean,’ ‘aggressive,’ or ‘abusive.’         * The circle and small triangle appear to be victims or targets of the large triangle's ‘torment.’         * Initially, the circle hides at the edge of the box, appearing ‘skittish.’         * The small triangle and circle ‘team up’ or ‘unite through a common enemy’ to trick the big triangle.         * The narrative concludes with the victims locking the big triangle in the box and ‘partying’ or celebrating before escaping.         * The final state of the large triangle is perceived as ‘frustrated’ or ‘angry,’ leading him to break his surroundings.

• Technical Insight (Causation Without Contact): The illusion of animacy is most powerful when there is perceived causation without direct physical contact. Because the shapes seem to respond to each other’s movements across distances, they are granted intentionality and life.

Biological Motion and Species Identification

• Eadweard Muybridge: Credited with arguably creating the first motion picture,

  • Muybridge captured animal motion to reveal how creatures move through the world.

• Point Light Displays: By placing lights on the joints of an animal (such as a pig), researchers can remove all visual information except for the motion itself.

• Confusion Matrix Results:

  • When humans view point light displays, they can identify the specific animal species based purely on movement patterns.

  • The ‘diagonal’ of the confusion matrix represents high success rates in correct animal assignments.

  • Common confusions occur (e.g., mistaking a camel for a pig), but generally, humans discern ‘species-typical physiological signatures’ through motion.

• Beyond Motion: Other physiological signatures, such as the white of the eye (sclera), are used alongside motion to make judgments about whether an object is a person, a specific animal, or a causal entity.

Social Perception and Identification of Individuals

• Defining Individual Recognition: Social life requires the ability to identify unique individuals (e.g., recognizing a specific student named Audrey) despite changes in their hairstyle, orientation, or environmental lighting.

• Constancy of Perception: We perceive a ‘stable individual’ despite physical fluctuations.

• Modeling Others: To survive and thrive socially, humans must create internal models of others' attributes, including:

  • Interests and goals.

  • Abilities (what they are good or bad at).

  • Trait dispositions (confidence, introversion, propensity for anger).

• Social Predictions: These models allow us to predict behavior and arrange our social world favorably. This aligns with Ervin Goffman’s work in The Presentation of Self in Everyday Life.

The ‘Fire Hose’ of Social Information

Humans use a deluge of data to infer internal states. Key channels include:

• Physical Features: Build, height, weight, mode of dress, eye color, hair color, and skin color.

• Demographics: Age, sex, gender, and race.

• Language and Auditory Cues: Words used, accent, vocabulary size, syntax errors, and speaking speed.

• Facial Expressions: Frowns, smiles, furrowed eyebrows, and eye contact (signaling attention or avoidance).

• Nonverbal Behavior and Context: Body language and posture. Interpretation varies by context (e.g., a motionless person in a bed vs. a motionless person on a classroom floor).

• Clusters and Congruence: Whether different channels of information (e.g., being quiet while hunched over a phone) add up to a consistent explanation.

• Interactions: Direct one-on-one experiences or observations of how a person treats a third party.

• Reputation: Third-party information about an individual's past behavior.

Sutherland et al. Face Study: Dimensions of Facial Traits

• Methodology:

• Researchers used hundreds of photos of Caucasian adult faces from the Internet.

  • Participants rated these on multiple traits: Aggression, approachability, trustworthiness, confidence, dominance, intelligence, health, attractiveness, age, baby-facedness, masculinity, femininity, and skin tone.

• Statistical Morphing: Researchers identified the top 20 and bottom 20 faces for each trait and created a statistical ‘average’ prototype for the high and low ends. Blends were created in $10\%$ increments to form a spectrum.

• Findings on Overlapping Dimensions:

  • Masculinity/Femininity: Dominant faces often appeared traditionally masculine. Trustworthy, intelligent, and attractive faces often appeared more traditionally feminine.

  • The ‘Average’ Effect: Morphing many faces together typically results in a face that people perceive as attractive.

• Three Fundamental Dimensions (Factor Analysis): Perceptions of complex traits are built upon three lower-order dimensions:

  1. Approachability

  2. Youthfulness / Attractiveness

  3. Dominance.

• Example Inference: ‘Trustworthiness’ is primarily a combination of high approachability, some youthfulness/attractiveness, and some dominance.

The Reliability and Speed of Face Judgments

• Snap Judgments: First impressions are made extremely rapidly, yet show high consensus (agreement) among different raters.

• Reliability Statistics (Correlations):

  • At $1.0\,s$ exposure: Trustworthiness correlation is $r = .74$; Competence is $r = .59$.

  • At $0.5\,s$ exposure: Trustworthiness drops only slightly to $r = .66$.

  • At $0.1\,s$ ($100\,ms$) exposure Judgments remain highly correlated with longer exposures. This speed is near the ‘limit of detection’ where a person can barely confirm they saw something.

• Reliability vs. Validity: High reliability means we all agree on who looks trustworthy, but it does not prove that those people actually are trustworthy.

Case Study: Ewing et al. Investment Game

• The Game: Participants are given $10$ tokens. They decide how many to give to a partner (‘trustee’), which will be multiplied. The partner then decides how much to return.

• Conditions: Silhouette only (no info), seeing a face, or reputational information.

• Results by Age:     * 5-year-olds: Gave $3.4$ tokens to trustworthy-looking faces and $2.4$ to untrustworthy ones.     * 10-year-olds: Gave $3.6$ tokens to trustworthy-looking faces and $2.0$ to untrustworthy ones (a $1.6$ token gap).     * Adults: Showed the largest gap, roughly $4.0$ tokens for trustworthy faces vs. $1.5$ for untrustworthy ones.

• Reputation vs. Face: While faces have a large effect, reputation (e.g., ‘this person played fairly $1/10$ times’) is a much more powerful driver of decision-making for all ages.

Real-World Consequences: Wilson and Nick Rule (Florida Prison Study)

• Investigation: Researchers had independent raters evaluate the faces of convicted murderers in Florida on traits like trustworthiness and attractiveness.

• Key Finding: There was a significant correlation between rated facial trustworthiness and the severity of the sentence received (Life in Prison vs. the Death Penalty).

• Statistics:

  • Facial trustworthiness was a predictor of sentencing ($\beta = -0.36$).

  • As rated trustworthiness increased, the odds of being assigned the death penalty decreased by approximately $30\%$.

  • The effect persisted even for prisoners who were later exonerated, meaning their actual innocence did not protect them from the bias of their facial appearance.

  • The effect remained significant even when controlling for attractiveness, maturity, glasses, and tattoos.

Validity of Facial Inferences: Yeager et al.

• Does Face Rating Predict Behavior? Yeager et al. tested whether relying on facial trustworthiness judgments helped participants win more money in investment games.

• Performance Metrics:

  • Using facial information leads to a payoff that is no better than random guessing ($r \approx 0$ validity).

  • Optimal Strategies: In the specific study environment, players who used a policy of ‘Always Trust’ performed worse than those who used a policy of ‘Always Defect’ (distrusting everyone).

  • Taking action based on facial snap judgments can actually be less effective than having a fixed blind policy, because the information on the face rarely reflects actual behavioral intent.

Questions & Discussion

• Student Question (Tom): Noted that in the shapes video, it seemed like the circle and small triangle weren't originally acquainted but united through a common enemy to trick the big triangle.

• Student Question (Unidentified): Asked about the variability in statistics for trustworthiness (e.g., why $0.1\,s$ might differ from $1.0\,s$).

• Response: Variations are likely due to small subject numbers (sample error) rather than a true loss of reliability. Statistically, the scores at $1.0\,s$, $0.5\,s$, and $0.1\,s$ do not differ significantly.

• Student Question (Unidentified): Inquired if race was omitted from the Sutherland study to avoid discrimination.     * Response: The researcher believed it was likely to restrict the range or variation in the data set to maintain a focused control group (Caucasian adult faces), though the exact justification was not recalled.