Curing robot autism: A challenge

Almost all robots are autistic; very few humans are. Out of the box, robots generally do not behave correctly in social settings (involving humans, or other agents). Most researchers treat this challenge behaviorally, by superficially tacking task- and domain- specific social behavior onto functioning individual robots. These rules are built once, and applied once. In contrast, I posit that we can build better socially-capable robots by relying on general social intelligence building blocks, built into the brains of robots, rather than grafted on per mission: built once, applied everywhere. I challenge the autonomous agents community to synthesize the computational building blocks underlying social intelligence, and to apply them in concrete robot and agent systems. I argue that our field is in a unique position to do this, in that our community intersects with computer science, behavioral and social sciences, robotics, and neuro-science. Thus we can bring to bear a breadth of knowledge and understanding which cannot be matched in other related fields. To lend credibility for our ability to carry out this challenge, I will demonstrate that we have carried out similar tasks in the past (though at a smaller scale). I conclude with a sample of some open questions for research, raised by this challenge.