The Defense Advanced Research Projects Agency (DARPA) is diving headfirst into the world of artificial intelligence, with a new program designed to make them more effective partners with human operators.
The program is called Science of Artificial Intelligence and Learning for Open-world Novelty (SAIL-ON). Its goal is to research and development of the foundations that would help AI systems act effectively in real, open world environments — such as fluctuating weather, shifting terrain or surprise attacks — create those AI systems, then demonstrate and evaluate them.
Existing AI systems have become ineffective as they are unable to adapt when something unexpected occurs.
“Imagine if the rules for chess were changed mid-game,” Ted Senator, program manager in DARPA’s Defense Sciences Office, said. “How would an AI system know if the board had become larger, or if the object of the game was no longer to checkmate your opponent’s king but to capture all his pawns? Or what if rooks could now move like bishops? Would the AI be able to figure out what had changed and be able to adapt to it?”
It takes critical time to retrain AI systems for larger data sets. SAIL-ON would, potentially, teach them to react without retraining necessary. To do that, its teams will undertake a series of tasks: characterize types and degrees of open world novelty, create software that generates unique situations in certain areas, and develop the algorithms necessary to identify and respond.
“The first thing an AI system has to do is recognize the world has changed,” Senator said. “The second thing it needs to do is characterize how the world changed. The third thing it needs to do is adapt its response appropriately. The fourth thing, once it learns to adapt, is for it to update its model of the world.”