Dating a man is like dating a robot - except for emotional structure

Dating a man is like dating a robot

except for emotional structure

Matthijs Pontier Johan Hoorn

Overview of this presentation

● Background● About the model● The Speed-Dating Application● Results● Conclusions● Future research

Background

● Previously, we integrated 3 emotion models into Silicon Coppelia, with the ultimate goal to create emotionally human-like robots

● In simulation experiments, the system behaved consistent with the theory it was based on, and seemed compelling intuitively

● However, we tested our model using agents interacting with each other, not with a real user

● Therefore we developed a speed dating application as a testbed for emotion models, and performed a Wizard of Oz experiment as a sort of Turing Test

The Models Incorporated in Silicon Coppelia● I-PEFiCADM, a model for building agents that

can trade rational for affective choices based on the concern-driven theory of Frijda.

● EMA ,a model to create agents that exhibit and cope with (negative) affect based on Smith & Lazarus’ theory of emotion

● CoMERG (the Cognitive Model for Emotion Regulation based on Gross), which can simulate different emotion regulation strategies explained by Gross using a set of logical rules and difference equations.

The Models Incorporated in Silicon Coppelia

● I-PEFiCADM: Model to let agents trade rational for affective choices, based on theory Frijda

The Models Incorporated in Silicon Coppelia● EMA: Model to let agents exhibit and cope with

(negative) affect based on Smith & Lazarus’ theory

The Models Incorporated in Silicon Coppelia

● CoMERG: Can simulate different emotion regulation strategies explained by Gross

Combined model: Silicon Coppélia

● Simulate affective decision making process: decisions based not only on rationality, but also on affective influences

● Simulate emotions based on beliefs about world-states, and how these affect goals 5 emotions are simulated in parallel:

hope, fear, joy, distress and anger● Emotion regulation strategies can be applied to

regulate (simulated) emotions

Silicon Coppelia

Determining which action to take

● Expected Satisfaction is calculated using the following formula, and the action with the highest level of expected satisfaction is picked

● ExpectedSatisfaction(Action) =

wat * Action_Tendency +

wpos * (1 - abs(positivity – biasI * Involvement))+

wneg * (1 - abs(negativity – biasD * Distance))

Determining emotions to express

● Hope and fear are based on perceived likelihood of goals, and ambition levels for these goal-states

● Joy and Distress are based on reaching or not reaching these goal-states

● Anger agent is calculated using believed responsibility of user for success speed date

● All 5 emotions are simulated in parallel, and are shown by the facial expression of the agent

The Speed-Dating Application

● User converses with the agent named ‘Tom’ by selecting a response from the drop-down box

● Agent picks its response based on affective decision making Silicon Coppelia

● Limited interaction in speed dating prevents recognizing difference based on richness of interaction

The Experiment

● 54 participants (all female heterosexuals) were told they were communicating with a software agent

● Condition 1: Agent was controlled by Silicon Coppelia● Condition 2: Agent was controlled by test-assistant (W.Oz)● In a questionnaire, we asked the participants how they

thought the agent perceived them, and what kind of emotions the participants perceived in the agent

Results & Discussion

● The participants did not recognize any differences in perceptions and emotions between agent-types

Good news for the engineer that wants to use these models for application development

● The participants did implicitly recognize differences in the cognitive architecture behind these perceptions

Bad news for the cognitive scientist trying to reproduce human cognitive functioning. We are close, but not completely there yet.

Results & Discussion

● Cognitive structure recognized in Wizard of Oz condition was more like cognitive structure found in humans in previous research Apparently, it is possible to recognize human emotional behavior in application with little interaction possibilities

Future Work

● Implement integrated model in Toy Chimpanzees that play simple Tic-Tac-Toe game

● Determine added value of individual models: I-PEFiCADM

EMA Gross

Questions?