Interplay between Natural and Artificial Intelligence in Training Autistic Children with Robots

The need to understand and model human-like behavior and intelligence has been embraced by a multidisciplinary community for several decades. The success so far has been shown in solutions for a concrete task or a competence, and these solutions are seldom a truly multidisciplinary effort. In this paper we analyze the needs and the opportunities for combining artificial intelligence and bio-inspired computation within an application domain that provides a cluster of solutions instead of searching for a solution to a single task. We analyze applications of training children with autism spectrum disorder (ASD) with a humanoid robot, because it must include multidisciplinary effort and at the same time there is a clear need for better models of human-like behavior which will be tested in real life scenarios through these robots. We designed, implemented, and carried out three applied behavior analysis (ABA) based robot interventions. All interventions aim to promote self initiated social behavior in children with ASD. We found out that the standardization of the robot training scenarios and using unified robot platforms can be an enabler for integrating multiple intelligent and bio-inspired algorithms for creation of tailored, but domain specific robot skills and competencies. This approach might set a new trend to how artificial and bio-inspired robot applications develop. We suggest that social computing techniques are a pragmatic solution to creation of standardized training scenarios and therefore enable the replacement of perceivably intelligent robot behaviors with truly intelligent ones.