Lecture 6: and decision-making

CSCI 534() – Lecture by Jonathan Gratch IF THE SLIDE CHANGED

CSCI 534(Affective Computing) – Lecture by Jonathan Gratch Lecture 6: Emotion and decision-making

CSCI 534(Affective Computing) – Lecture by Jonathan Gratch = =

CSCI 534(Affective Computing) – Lecture by Jonathan Gratch ?

Should I wear a mask?

Should schools reopen? Should I get vaccinated?

CSCI 534(Affective Computing) – Lecture by Jonathan Gratch Can make us smart? Example

Task: Pick a poster to take home

• One group just grabbed one they feel good about • Other group asked to think carefully and write down their reasons for choosing

• Got to reconsider 6-months later

Wilson, T. D., Lisle, D., Schooler, J. W., Hodges, S. D., Klaaren, K. J., & LaFleur, S. J. (1993). Introspecting about reasons can reduce post-choice satisfaction. Personality and Social Psychology Bulletin, 19, 331–339.

CSCI 534(Affective Computing) – Lecture by Jonathan Gratch This lecture

▪ Emotion and decision-making – Examine what emotions do: how they shape decisions – Introduce “rational” models of decision-making ▪ Review rational choice theory (model that underlies economics)

– Contrast emotional from “rational” decisions ▪ Illustrate “emotional” departures from rational choice ▪ Describe “behavioral economic” models that capture this

CSCI 534(Affective Computing) – Lecture by Jonathan Gratch First an experiment

▪ Need 3 volunteers (that consent to being filmed) – Opportunity to win a reward – Have a chance of winning up to $20

▪ Note on Economic vs. Psychological research – Deception heavily discouraged in economics (can’t publish)

CSCI 534(Affective Computing) – Lecture by Jonathan Gratch Blue Player Red Player Gold Player

First count your pulse for 20 seconds

Write it down

CSCI 534(Affective Computing) – Lecture by Jonathan Gratch CSCI 534(Affective Computing) – Lecture by Jonathan Gratch Blue Player Red Player Gold Player

Finally, count your pulse for 20 seconds

Write it down

CSCI 534(Affective Computing) – Lecture by Jonathan Gratch Which gamble would you prefer?

Gamble 1 Gamble 2

CSCI 534(Affective Computing) – Lecture by Jonathan Gratch Another Illustration

CSCI 534(Affective Computing) – Lecture by Jonathan Gratch a) b)

$5 $9 $1 $1

c) d) -$22 -$54 $10 $10

In each case, expected value is $2

CSCI 534(Affective Computing) – Lecture by Jonathan Gratch Emotion a benefit or curse? A very old debate

▪ Plato argues emotion and intellect are in opposition

Reason lover of honor and modesty and temperance, and the follower of true glory; he needs no touch of the whip, but is The Allegory of the Chariot guided by word and admonition only

Emotion crooked lumbering animal, … the mate of insolence and , shag-eared and deaf, hardly yielding to whip and spur.”

CSCI 534(Affective Computing) – Lecture by Jonathan Gratch Emotion and Rationality

▪ Emotion is often said to “distort” reason. What does that mean?

▪ It means there is a theory of how people “should” make decisions (Rational Choice Theory) – Follows from a set of principles that seem an irrefutable characteristic of good decision-making

▪ And people don’t follow that theory

▪ And emotions help explain departures from rational choice theory

CSCI 534(Affective Computing) – Lecture by Jonathan Gratch Why should we care that people not “rational”

▪ Rational models are used to predict human decisions and make policy decisions across wide range of applications – Economic decisions: how individuals buy and invest – Public policy decisions: how will programs impact the and well-being of populations – Consumer choice: will consumers be satisfied by a product – Market mechanisms design: will policies governing transactions in an online marketplace be efficient – Technology

▪ Rational models are used to guide automated systems – Bargaining agents – Security agents – Navigation systems

CSCI 534(Affective Computing) – Lecture by Jonathan Gratch Example: Security

Milind Tambe

Rational models (game theory) can help us build decision-aids for such efficient security resource allocation. Use computational methods to predict the decision-making of potential criminals

CSCI 534(Affective Computing) – Lecture by Jonathan Gratch Example: Navigation

Sarit Kraus

Rational models can help us recommender systems such as driver navigation systems. Challenge is to recommend high quality routes that satisfy user preferences while achieving other goals (energy efficiency)

CSCI 534(Affective Computing) – Lecture by Jonathan Gratch Rational Choice Theory

▪ Developed over centuries

▪ Central foundation of economic decision-making

▪ Claimed to serve two basic purposes – Normative: how people (and machines) should act and think ▪ Helps us avoid confused, poor thinking ▪ Helps us analyze arguments ▪ Aids in design of “optimal” artificial decision-makers

– Descriptive: how people (and machines) actually act and think? ▪ Fundamental postulate of economics: people act rationally ▪ Allows that individuals may not be rational but this can be viewed as noise so that the population will act rationally (rationality “of the crowd”)

CSCI 534(Affective Computing) – Lecture by Jonathan Gratch Variants of Rational Choice Theory

▪ Decision theory centers on cost-benefit calculations that individuals make without reference to anyone else’s plans

▪ Game theory analyzes how people make choices based on what they expect other individuals to do. – We will discuss this when we consider

CSCI 534(Affective Computing) – Lecture by Jonathan Gratch Why is decision theory called rational

▪ What makes a decision rational?

▪ How would we develop definition?

CSCI 534(Affective Computing) – Lecture by Jonathan Gratch Axioms of Decision-Theory

1. Completeness:

– All actions (or objects) can be ranked in an order of preference; indifference between two or more alternatives is possible

= Tesla Model S BMW M6

2. Transitivity:

– If action (or object) a1 is preferred to action a2 and action a2 is preferred to a3, then a1 is preferred to a3. AMC Pacer

CSCI 534(Affective Computing) – Lecture by Jonathan Gratch Axioms of Decision-Theory

3. Continuity:

– When there are three lotteries (X, Y and Z), X preferred to Y and Y preferred to Z, then there should be a mixture of X and Z such that an individual is indifferent between this mix and Y

X Y Z Y

90% + 10% = 100%

4. Independence: – If we mix two lotteries (X, Y) with a third one, the ordering of the two mixtures will not change regardless of the particular third lottery used

50% + 50% 50% + 50%

CSCI 534(Affective Computing) – Lecture by Jonathan Gratch Decision Theory (or Expected Utility Theory)

▪ Decision theory satisfies these axioms

▪ Outcomes can be described by a utility function – The value (or happiness) derived from achieving this state – E.G. Money could be a person’s measure of happiness The value of winning a $1,000,000 lottery ticket is $1,000,000

▪ Outcomes can be described by a probability fn. – The likelihood that this state might be achieved in the future

▪ Decisions are then driven by Expected Utility

CSCI 534(Affective Computing) – Lecture by Jonathan Gratch Decision Theory: people utility maximizers

U EU Decision Point 70 Chance Event 20% 70 * 80% + 70 * 20% = 70

70 80%

20% 0

100 * 80% + 0 * 20%= 80

80% 100

CSCI 534(Affective Computing) – Lecture by Jonathan Gratch Is Decision Theory “Rational”

▪ If decision maker doesn’t follow decision theory, always possible to construct a choice of gambles such that they will lose money

CSCI 534(Affective Computing) – Lecture by Jonathan Gratch Money not necessarily equal to happiness

Utility Theory doesn’t assume money is people’s utility function People assign utility to money. Different people have different utility fn.

Utility is the “anticipated ” of wealth rather than wealth per se

Daniel Bernoulli 1738

CSCI 534(Affective Computing) – Lecture by Jonathan Gratch Money not necessarily equal to happiness

Utility Theory doesn’t assume money is people’s utility function People assign utility to money. Different people have different utility fn.

Daniel Bernoulli 1738

CSCI 534(Affective Computing) – Lecture by Jonathan Gratch Money not necessarily equal to happiness

Utility Theory doesn’t require utility to equal objective value (i.e., money) People assign utility to money. Different people have different utility fn.

For many people, each dollar has less value the richer we become A dollar tomorrow is worth less than a dollar today (temporal discounting)

Utility theory assumes decisions are based Utility on subjective utility and subjective probability

Much research focuses on how to elicit these decision parameters

Monetary value

30 CSCI 534(Affective Computing) – Lecture by Jonathan Gratch Summary: Rational Choice Theory

▪ Decision theory centers on cost-benefit calculations that individuals make without reference to anyone else’s plans

▪ Captures many aspects of how people make decisions (maximize pleasure)

▪ Core assumption underling most economic theory and economic decision

▪ Core assumption underlying most artificially intelligent systems

CSCI 534(Affective Computing) – Lecture by Jonathan Gratch Is Expected Utility a good model of human choice?

CSCI 534(Affective Computing) – Lecture by Jonathan Gratch CSCI 534(Affective Computing) – Lecture by Jonathan Gratch

Test <

CSCI 534(Affective Computing) – Lecture by Jonathan Gratch

Test <

CSCI 534(Affective Computing) – Lecture by Jonathan Gratch Gamble A < Gamble B

CSCI 534(Affective Computing) – Lecture by Jonathan Gratch Gamble A < Gamble B Gamble C > Gamble D

CSCI 534(Affective Computing) – Lecture by Jonathan Gratch Gamble A < Gamble B Gamble C > Gamble D

CSCI 534(Affective Computing) – Lecture by Jonathan Gratch Gamble A < Gamble B Gamble C > Gamble D

CSCI 534(Affective Computing) – Lecture by Jonathan Gratch Gamble A < Gamble B Gamble C > Gamble D

CSCI 534(Affective Computing) – Lecture by Jonathan Gratch Gamble A < Gamble B Gamble C > Gamble D

$1,000,000 $1,000,000 89% 89% 11% < 11%

CSCI 534(Affective Computing) – Lecture by Jonathan Gratch Violation of Axiom of Independence

Gamble A < Gamble B Gamble C > Gamble D

$1,000,000 $1,000,000 $0 $0 89% 89% 89% 89% 11% < 11% 11% > 11%

CSCI 534(Affective Computing) – Lecture by Jonathan Gratch Violation of Axiom of Independence

$1,000,000 $1,000,000 $0 $0 89% 89% 89% 89% 11% < 11% 11% > 11%

CSCI 534(Affective Computing) – Lecture by Jonathan Gratch Another example

▪ Which would you rather do?

Drive to USC each day Live next to nuclear power plant

People overweight low probability but high (dis)utility events

CSCI 534(Affective Computing) – Lecture by Jonathan Gratch Another example

• Fungibility is the property of a good or a commodity whose individual units are essentially interchangeable • An assumption in economics is money and commodities are fungible • Endowment effect is violation of this assumption

CSCI 534(Affective Computing) – Lecture by Jonathan Gratch Another example

2nd 3rd

• Reference dependent decision-making

CSCI 534(Affective Computing) – Lecture by Jonathan Gratch What to do?

▪ People don’t follow axioms of decision theory

▪ Yet they are not random. Follow patterns

▪ Can we develop a descriptive utility theory?

CSCI 534(Affective Computing) – Lecture by Jonathan Gratch What to do?

▪ People don’t follow axioms of decision theory

▪ Yet they are not random. Follow patterns

▪ Can we develop a descriptive utility theory?

▪ Some models just say math is hard

Rank-dependent expected utility People overweight low-probability events such as winning the lottery

CSCI 534(Affective Computing) – Lecture by Jonathan Gratch What to do?

▪ People don’t follow axioms of decision theory

▪ Yet they are not random. Follow patterns

▪ Can we develop a descriptive utility theory?

▪ Some models say people bad at math

▪ Many models appeal to concepts that seem like emotion – Aversion, loss, , disappointment

CSCI 534(Affective Computing) – Lecture by Jonathan Gratch One idea: emotions arising from risk People have different toward risk. People can be risk avoiders, risk seekers (or risk lover) , or indifferent toward risk (risk neutral).

Utility Value

Risk neutral

Risk avoider

Risk lover Monetary Value

Utility of money shown for different types of people. Note that for equal increments in dollar value the utility either rises at a decreasing rate (avoider), constant rate or increasing rate (lover). Key point: Absolute value of outcome not important. It is this evokes

CSCI 534(Affective Computing) – Lecture by Jonathan Gratch One idea: counterfactuals

Prospect Theory Tversky Kahneman Feelings towards risk depend on a reference point People losses more than the enjoy gains (w.r.t. this reference point)

CSCI 534(Affective Computing) – Lecture by Jonathan Gratch Modeling decisions: Big picture (Loewenstein and Lerner 2003)

How does emotion impact decision-making?

CSCI 534(Affective Computing) – Lecture by Jonathan Gratch Modeling decisions: It’s messy (Loewenstein and Lerner 2003)

CSCI 534(Affective Computing) – Lecture by Jonathan Gratch Biggest idea: Anticipated

▪ Decisions impacted by how you anticipate you will feel about the result

Decision/ Expected Expected behavior consequences emotions

David Hume

Claim: People will pick decisions with greatest expected pleasure

CSCI 534(Affective Computing) – Lecture by Jonathan Gratch Anticipated Affect

▪ We’ll walk through one model in detail

▪ Mellers et al. Decision – Argues people anticipate how they will feel about outcomes of decisions and use their predictions to guide choice – People are assumed to choose option with greater subjective expected pleasure – Propose a mathematical model to predict how people feel – Emphasizes role of specific emotions: , disappointment, and regret

CSCI 534(Affective Computing) – Lecture by Jonathan Gratch Role of surprise

Lottery A Lottery C

$5 $5 $1 $1

Which of these outcomes would give you more pleasure? Why? People find outcome of Lotter C more surprising and thus are more elated If both lotteries involved losses, the surprising outcome would be more disappointing

CSCI 534(Affective Computing) – Lecture by Jonathan Gratch Decision Affect theory ▪ Models how we feel after an uncertain outcome

Elated a Priori probability of outcome

Disappointed

CSCI 534(Affective Computing) – Lecture by Jonathan Gratch Role of relief and disappointment

Lottery A Which of these outcomes would give you more pleasure? Each outcome equally surprising $5 Why? $1

This is another example of counterfactual reasoning. Lottery B The availability of an alternative influences reactions $5 Our elation or disappointment shaped by the difference $-10 between obtained and alternative outcome

CSCI 534(Affective Computing) – Lecture by Jonathan Gratch Decision affect theory (DAT) – Part 1

▪ Consider a gamble with Outcomes A and B of utility UA and UB respectively

▪ DAT predicts that the emotional response to A is

RA UA + d(UA - UB) (1 – pA)

Reward disappointment surprise

– d(UA - UB), the disappointment function, is a power function with different exponents for positive and negative distances – In their data d(x) = x1.16 if x>0 and -|x|1.20

p UA A UB

CSCI 534(Affective Computing) – Lecture by Jonathan Gratch And this shapes choice between alternatives

▪ Consider a gamble with Outcomes A and B of utility UA and UB respectively

▪ DAT predicts that the emotional response to A is

RA UA + d(UA - UB) (1 – pA) ▪ When choosing between gambles: – Rather than using expected utility – Use expected pleasure/

IF pARA + (1-pA)RB > pCRC + (1-pC)RD , Pick gamble 1, else pick gamble 2

Contrast EUT: IF pAUA+(1-pA)UB > pCUC+(1-pC)UD , Pick gamble 1, else pick gamble 2

CSCI 534(Affective Computing) – Lecture by Jonathan Gratch 25% A $5 EU = $5*1/4 + $1*3/4 = $2

B $1 RA = UA + d(UA - UB) (1 – pA) RB = UB + d(UB - UA) (1 – pB) 5 + 41.16 * 0.75 = 8.75 1 + -(41.20)* .25 = -0.32

ER = 8.75*1/4 – 0.32*.75 = 1.95 12.5% C EU = $-54*1/8 + $10*7/8 = $2 -$54 D

$10 RC = UC + d(UC - UD) (1 – pC) RD = UD + d(UD - UC) (1 – pD) -54 + -(641.20)*7/8 = -182.7 10 + 641.16 * 1/8 = 25.6 ER = -182.7*1/8 + 25.6 * 7/8 = -.46

From Mellers, disappointment fn: d(x) = x1.16 if x>0 and -|x|1.20

CSCI 534(Affective Computing) – Lecture by Jonathan Gratch But that’s not all! DAT Part 2 ▪ Most decisions we only learn outcome of our choice – What if I married someone else? – What if I went to MIT instead of USC?

▪ We can’t go back in time and try things differently – Mellers refers to this as partial feedback

▪ But sometimes we can

▪ Which of these bargains do you want to pick?

$30 $20

$-32 $-8 EU=-16.5 EU=-1 Under complete feedback we may experience regret

CSCI 534(Affective Computing) – Lecture by Jonathan Gratch Regret

▪ Feelings under complete feedback modeled with regret function

EA(C) UA + d(UA - UB) (1 – pA) + r(UA - UC) (1 – pA pC)

reward disappointment surprise regret

– d(UA - UB), the disappointment function

– r(UA - UC), the regret function, is a power function with different exponents for positive and negative distances from alternative

Pick 1 Didn’t pick 2

CSCI 534(Affective Computing) – Lecture by Jonathan Gratch But that’s not all! DAT Part 2

▪ Feelings under complete feedback modeled with regret function

EA(C) UA + d(UA - UB) (1 – pA) + r(UA - UC) (1 – pA pC)

reward disappointment surprise regret

NOTE These terms correspond to appraisal variables in (desirability, expectedness, etc.)

CSCI 534(Affective Computing) – Lecture by Jonathan Gratch DAT Summary

▪ Decisions shaped by surprise, disappointment and regret

Pleasure Surprise Disappointment Regret Effects Effects Effects

Displeasure

CSCI 534(Affective Computing) – Lecture by Jonathan Gratch Summarize what we’ve learned People try to maximize expected emotion (utility)

Decision/ Expected Expected behavior consequences emotions

David Hume

CSCI 534(Affective Computing) – Lecture by Jonathan Gratch And expected emotions ≠ expected utility

▪ Expected emotions are shaped by uncertainty (risk) in ways that violate independence assumption – People overweigh small probabilities – People underestimate large probabilities – Losses Loom larger than gains

Prospect Theory Kahneman & Tversky, 1979 Decision Affect Theory Mellers, Probability

function Subjective probability Subjective

CSCI 534(Affective Computing) – Lecture by Jonathan Gratch This fits with what we learned from appraisal theory

Anticipated Events Cognitive evaluation Goals ( U ) (outcomes) (appraisal) A

A & C Affect Decision

RA(C) UA + d(UA - UB) (1 – pA) + r(UA - UC) (1 – pA pC)

Reward expectation Regret Surprise Disapp. Counterfactual reasoning

CSCI 534(Affective Computing) – Lecture by Jonathan Gratch Appraisal models (review)

▪ Computational models of appraisal propose simple rules that appraise abstract data structures

 Past Present Future ➔

Safe Safe Utility: 50 Utility: 50 Prob.: 100% Prob: 50% Belief: False Intend: True Bird Attack Cause: Other Inhibits Intend: yes Whack Bird Prob: 100% Cause: self Intend: yes Prob.: 50%

CSCI 534(Affective Computing) – Lecture by Jonathan Gratch Summarize what we’ve learned People try to maximize expected emotion

Decision/ Expected Expected behavior consequences emotions

David Hume

And if we forecast the emotions people anticipate from a decision, we can recommend decisions that make them happy, yes?

CSCI 534(Affective Computing) – Lecture by Jonathan Gratch Here’s a complication

Rate intensity of the following events (1-weak; 10-strong)

▪ Jack sustained fatal injuries in a car crash

▪ Jack was killed by a semi trailer that rolled over on his car and crushed his skull

▪ Jack lost the skin of his finger in a rugby match

▪ What’s going on here?

CSCI 534(Affective Computing) – Lecture by Jonathan Gratch Here’s a complication

▪ We tend to feel stronger emotions when the stimuli is vivid

▪ When we imagine future situations we often fail to vividly imagine the consequences

CSCI 534(Affective Computing) – Lecture by Jonathan Gratch Problem: It seemed a good idea at the time…

The morning after effect

CSCI 534(Affective Computing) – Lecture by Jonathan Gratch

▪ People make decisions by forecasting their emotions

▪ People not so good at forecasting – What: what emotion they will feel following a decision – How much: the intensity of the experience – How long: the duration of the emotion

▪ People reason about the future abstractly – “Jack sustained fatal injuries “

▪ People fail to account for their ability to cope – Become desensitized to positive circumstances – Become resigned to negative circumstances

▪ People overweigh outcomes in immediate focus – E.g., Students in mid-west predicted they would be happier moving to California; students in California predicted they’d be less happy in mid- west; yet both equally happy

CSCI 534(Affective Computing) – Lecture by Jonathan Gratch As a consequence Expected emotion ≠ experienced emotion

Decision/ Expected Expected behavior consequences emotions

CSCI 534(Affective Computing) – Lecture by Jonathan Gratch Poor affective forecasting leads to poor decisions

▪ “morning after” effects

▪ Low retirement savings rates

▪ Lack of energy conservation

▪ Risky health choices

▪ Impulsivity

CSCI 534(Affective Computing) – Lecture by Jonathan Gratch Affective forecasting

▪ Evidence for 2 distinct mechanisms for forecasting

– Simulation route: ▪ Vividly imagine being in a certain situation ▪ “read” our bodily reactions to that situation (Damasio’s somatic marker hypothesis) – Reasoning route: ▪ Reason about emotions: e.g., I expect I would feel this way ▪ Evidence that the “reasoning” approach more suspect to mis- forecasting effects

– Situational factors bias these mechanisms ▪ E.g. more immediate events more likely to use simulation route – Some individual differences predict this tendency ▪ Mental imagery ability (White 1978)

CSCI 534(Affective Computing) – Lecture by Jonathan Gratch Attempts to salvage EUT

▪ Expected emotions are time dependent: care less about events far in the future (explains procrastination?)

▪ Can be modeled with hyperbolic discounting

Make utility a function of time Expected Does this sound Discounted familiar? Cumulative Reward (Q-value)

CSCI 534(Affective Computing) – Lecture by Jonathan Gratch MacInnis. Whan. "Looking through the crystal ball: Affective forecasting and misforecasting in consumer behavior." Review of Marketing Research 2 (2005): 43-80.

CSCI 534(Affective Computing) – Lecture by Jonathan Gratch Could affective computing help

CSCI 534(Affective Computing) – Lecture by Jonathan Gratch Approach: Make the consequences of decisions immediate and tangible through virtual reality (Bailenson)

CSCI 534(Affective Computing) – Lecture by Jonathan Gratch Example 1: environmental conservation

▪ Global warming serious issue

▪ People tend to support conservation policies but people often wasteful in their individual choices

– Could virtual reality make consequences seem more vivid?

– Would this result in actual pro-environmental behavior?

CSCI 534(Affective Computing) – Lecture by Jonathan Gratch Study 1: waste

Pre-tested attitudes on conservation

Told # of trees cut down to make toilet paper

Virtual Reality Mental Imagery

CSCI 534(Affective Computing) – Lecture by Jonathan Gratch Study 1: environmental conservation

CSCI 534(Affective Computing) – Lecture by Jonathan Gratch Study 1: waste

Pre-tested attitudes on conservation

Told # of trees cut down to make toilet paper

Virtual Reality Mental Imagery

Irrelevant Task (30 minutes)

Cleanup spilt water

VR participants used significantly fewer napkins 30min later

CSCI 534(Affective Computing) – Lecture by Jonathan Gratch Example 2: Shower study

▪ Could same idea get people to use less water?

– Read about coal and shower

– Touched physical coal

– Washed hands

– In VR randomly assigned to 1 of 4 conditions

– 6 minute virtual shower

– Washed hands (main DV)

CSCI 534(Affective Computing) – Lecture by Jonathan Gratch Experimental Conditions

CSCI 534(Affective Computing) – Lecture by Jonathan Gratch Experimental Conditions

CSCI 534(Affective Computing) – Lecture by Jonathan Gratch Experimental Conditions

Vivid conditions yielded significantly quicker hand washing

CSCI 534(Affective Computing) – Lecture by Jonathan Gratch Yet another example

CSCI 534(Affective Computing) – Lecture by Jonathan Gratch Summary

Decision/ Expected Expected behavior consequences emotions

Expected emotions ≠ experienced emotions But this difference can be predicted and possibly reduced

CSCI 534(Affective Computing) – Lecture by Jonathan Gratch But that’s not all recall experiment….

CSCI 534(Affective Computing) – Lecture by Jonathan Gratch Example

Immediate emotions

Incidental influences

CSCI 534(Affective Computing) – Lecture by Jonathan Gratch Incidental influences

Anticipatory influences

Maximize Immediate Expected Expected Probability emotions Emotion Utility

▪ Unrelated events can influence our immediate Incidental emotions influences – Sunny day – Happy or sad music – Disgusting room

CSCI 534(Affective Computing) – Lecture by Jonathan Gratch Another example

▪ Use of affective computing technology to demonstrate the pervasive impact of incidental influences

CSCI 534(Affective Computing) – Lecture by Jonathan Gratch Another example

▪ Use of affective computing technology to demonstrate the pervasive impact of incidental influences

Analyzed the sentiment of posts using Linguistic Inquiry Word Count (LIWC), a widely used and validated word classification system

CSCI 534(Affective Computing) – Lecture by Jonathan Gratch Another example

▪ Use of affective computing technology to demonstrate the pervasive impact of incidental influences

Analyzed the sentiment of posts using Linguistic Inquiry Word Count (LIWC), a widely used and validated word classification system

CSCI 534(Affective Computing) – Lecture by Jonathan Gratch Results

▪ What happens when Facebook poster is in rainy city? – More negative and less positive posts – e.g., a rainy day in New York City directly yields an additional 1500 (95% CI 1100 to 2100) negative posts by users in New York City

▪ What happens when a facebook poster has friends in a rainy city? – They “catch” their friends emotions – A rainy day in New York City yields about 700 (95% CI 600 to 800) negative posts by their friends elsewhere

CSCI 534(Affective Computing) – Lecture by Jonathan Gratch Finally: Immediate emotions change how we forecast (Loewenstein and Lerner 2003)

CSCI 534(Affective Computing) – Lecture by Jonathan Gratch Lerner&Tiedens06: Portrait of the angry decision maker

CSCI 534(Affective Computing) – Lecture by Jonathan Gratch CSCI 534(Affective Computing) – Lecture by Jonathan Gratch Appraisal Tendencies Framework (, Lerner, Keltner 2007.)

Desirability UNDESIRABLE

Goals/Beliefs/ Goals/Beliefs/ Environment Controllability Environment UNCONTROLABLE Intentions Intentions

Causal Attribution BLAMEWORTHY

Emotion Emotion Action Physiological Lo Tendencies “Affect” Response Withdraw

CSCI 534(Affective Computing) – Lecture by Jonathan Gratch Appraisal Tendencies Framework (Han, Lerner, Keltner 2007.)

Desirability UNDESIRABLE

Goals/Beliefs/ Goals/Beliefs/ Environment Controllability Environment CONTROLABLE Intentions Intentions

Causal Attribution BLAMEWORTHY

Emotion Emotion Action Physiological HiLo ArousalArousal Tendencies “Affect” Response APPROACHWithdraw SadnessANGER

CSCI 534(Affective Computing) – Lecture by Jonathan Gratch Appraisal Tendencies Framework (Han, Lerner, Keltner 2007.)

Desirability UNDESIRABLE

Goals/Beliefs/ Goals/Beliefs/ Environment Controllability Environment UNCONTROLABLECONTROLABLE Intentions Intentions

Causal Attribution BLAMEWORTHY

Emotion Emotion Action Physiological APPROACH HiLo ArousalArousal Tendencies “Affect” Response Withdraw SadnessANGER

CSCI 534(Affective Computing) – Lecture by Jonathan Gratch Summary (Loewenstein and Lerner 2003)

CSCI 534(Affective Computing) – Lecture by Jonathan Gratch Summary

▪ People don’t follow rational choice theory

▪ People’s emotions depend on comparisons with other possible outcomes (counterfactuals)

▪ People make decisions based on expected emotion, not expected utility

▪ But people are bad at predicting their actual emotions

▪ And influenced by irrelevant emotion

▪ And this can be modeled computationally

CSCI 534(Affective Computing) – Lecture by Jonathan Gratch Next Time

▪ Today we had a number of experiments

▪ Many class projects will have experiments – E.g., show evidence your ideas work

▪ Next time we’ll have guest lecture on experimental design – Also see recommended reading

CSCI 534(Affective Computing) – Lecture by Jonathan Gratch