As the use of robots in many fields increases, so do fears regarding the dangers of artificial intelligence. In response to the question, ‘how do you program a robot to have morals?’ researchers Mark Riedl and Brent Harrison at the Georgia Institute of Technology believe they may have found an answer – through storytelling.
Fairy tales and fables have been used for centuries as a way to teach social norms and positive values to children, while warning them of the risks resulting from poor decision-making. This method is now being applied to AI.
Reidl and Harrison unveiled their system, Quixote, last week at the AAAI-16 Conference in Arizona. It builds upon the Scheherazade system, also developed by Riedl, which maps out ‘correct’ plot sequences learned through internet crowdsourcing. Quixote is then used to teach robots to read these stories and rewards them for making the socially acceptable, ethical choices along the way, while punishing the opposite. Practically, the right choices are the ones which the moral protagonist would make.
Robots programmed using this system will be able to weigh several options in any given scenario, and understand how each possible choice will affect the ‘plot’ in order to make the correct, socially acceptable one. This would dissuade the convenience of robbery, for example, which could otherwise be considered the best path since it is both the fastest and cheapest way to achieve a goal.
Riedl explained: "We believe story comprehension in robots can eliminate psychotic-appearing behavior and reinforce choices that won't harm humans and still achieve the intended purpose." This system will initially be used for robots that have to interact with humans in order to do their jobs, and is a step towards instilling morals in AI in general.