The military want an AI that will play by the rules
Trained computers with elements of artificial intelligence for a long time to beat people in chess, and recently learned to play Go. Of course, these programs can be called AI only with the big stretch. But the point is that the person somehow was able to create a computing system that can be trained to solve certain problems according to a strictly established algorithms in a strictly defined framework. Unfortunately, it doesn’t work in open world, when changing environment, problems and all.
To solve AI adaptation to changing conditions in the solution of the task, the advanced research Agency of the U.S. DoD (DARPA) announces the launch of the program Science of Artificial Intelligence and Learning for Open-world Novelty (SAIL-ON). Almost like the command to set sail for open sea. Training the AI to adapt to the novelty of the open world involves the study and development of the basic scientific principles and General engineering technologies and algorithms necessary to create artificial systems that could be consistent and to operate effectively in the open world.
The goal of the program is to develop scientific principles for the quantitative and qualitative description of the phenomena and situations that the AI for the first time can encounter in the open world. In other words, AI should obtain a set of rules and mechanisms for evaluation of something new for them with the ability to instantly appropriate response. The program also SAIL-ON needs to help create such systems and to demonstrate and to organize their learning in areas specified by the Ministry of defence.
Experience in the field of Autonomous driving, which deserve respect and demonstrate progress, for solving tasks of the military is definitely not good. It’s one thing to drive on the roads according to the rules, and quite another to navigate the road in a completely strange environment. Therefore, before program participants SAIL-ON is a difficult task with many subordinate areas in which have to solve many challenging mathematical and algorithmic problems.