[3G3-OS-18a] 感情とAI(1)

[3G3-OS-18a-02] Action Selection based on Somatic Marker Hypothsis

〇Chie Hieida¹, Takato Horii¹, Takayuki Nagai^1,2 (1. The University of Electro-Communications, 2. Osaka University)

Keywords:Emotion, Somatic Marker Hypothsis, Deep Learning

Emotions are very important for human intelligence; however the mechanism of emotions is not yet fully clarified. The important aspect of embodiment in emotion has been claimed by Damasio's somatic marker hypothesis, which proposed that emotions evaluate external stimuli efficiently through the body. As a first step toward understanding the mechanism of emotion, we try to verify the somatic-marker hypothesis using a computer simulation. Specifically, we introduce a module that learns actions using body signals, and verify whether the agent can learn to obtain higher reward using the body signals. As the result, the simulation reveals that the model with body signals can select actions with higher reward compared to models without signals from the body.