Abstract
To evaluate our emotionally intelligent software, we put a virtual human capable of speech and facial expressions to an updated and enriched version of the traditional Turing test. In a speed-date with 54 young females, either our software or human confederates controlled the simulation of the virtual human’s affective performance. Results were obtained with frequentist analysis and Bayesian structural equation modeling. Indeed, participants did not detect differences and observed similarity in the emotional behavior of the virtual human and in the way it assumingly perceived them. Additionally, participants did not recognize different but similar cognitive-affective structures between humans and our system. As is, designers may use our software for believable affective virtual humans or robots. Moreover, as far as the richness of interaction possibilities in the speed-dating session allowed, our software seems to reproduce human cognitive-affective structures.
| Original language | English |
|---|---|
| Pages (from-to) | 235-253 |
| Number of pages | 19 |
| Journal | International Journal of Social Robotics |
| Volume | 11 |
| Issue number | 2 |
| Early online date | 31 Oct 2018 |
| DOIs | |
| Publication status | Published - Apr 2019 |
Keywords
- Affective computing
- Bayesian analysis
- Cognitive models
- Social agents
- Turing test
