Affective Agents perceiving each other’s actions

Matthijs PontierGhazanfar F. Siddiqui

Overview of this presentation

● Goal of this study● About the model● Simulation experiments● Conclusions● Future research

Goal of this study

● Ultimately, we want to build a robot that can interact with humans in an emotionally natural (human-like) way

● This study: Improve affective decision making model:

− Emotions change according to performed actions− Perceptions change according to performed actions− Changed perceptions affect decisions

I-PEFiCADM

Effect of actions on perceived ethics

● new_bias(good) = pgood * old_bias(good) +

(1-pgood) * v(agent, action, good)

VGoodAction

0.75Avoid

1.25Criticize

0.5Fight

1.5Comfort

VBadAction

1.25Avoid

1.25Criticize

1.5Fight

0.5Comfort

Effects of actions on beliefs

● if beliefchange ≥ 0:new_belief = old_belief + belief_adaptation *beliefchange * ((1 - old_belief) / 2)

● if beliefchange < 0:new_belief = old_belief + belief_adaptation *beliefchange * ((1 + old_belief) / 2)

Effects of actions on emotions

● new_anger = panger / n * old_anger +

(1- panger / n) * Σ(changed_anger) / n

● panger = persistency factor for anger● changed_anger = value attached to action● n = number of changed_anger values taken

into account

Simulation experiments

● In the simulation experiments, three agents (Harry, Barry and Gary) followed a (fictitious) anger management therapy

● A baseline experiment was performed, with all features set to 0.5, and all beliefs set to 0.

● This resulted in all agents avoiding one other agent

Experiment 1: Baseline

Conclusions

Future research

Questions?