UC Berkeley Launches a Center for Human-Compatible Artificial Intelligence

Thursday, 01 September 2016 - 8:34PM
Robotics
Science News
Thursday, 01 September 2016 - 8:34PM
UC Berkeley Launches a Center for Human-Compatible Artificial Intelligence
UC Berkeley has recently given birth to the Center for Human-Compatible Artificial Intelligence, a place of research meant to focus on ensuring that future AI act purely for purposes beneficial to human beings and in accordance with human values, no matter what advances may be made in the field of robotics.

Professor Stuart Russell heads the Center, and according to him, their efforts have nothing to do with a fear of the singularity or anything of that nature. The problem is that artificial intelligences take their objectives in an extremely literal manner, and one only has to imagine how that might lead to chaos. The Center for Human-Compatible Artificial Intelligence means to ensure that such chaos doesn't occur.

Russell's team plans on implementing an approach called "inverse reinforcement learning." This process involves AI watching human beings perform tasks and then imitating them. By doing so, they hope to avoid more of the adverse situations that might crop up by programming them with a specific task in mind that the AI might take too literally.

Of course such an endeavor isn't going to be easy. Russell himself has noted the difficulties that arise because "humans are inconsistent, irrational, and weak-willed," and that human values often differ with geography. With such being that state of things, getting artificial intelligences to cooperate exactly as humans would want them to in every imaginable circumstance will likely never be feasible in an absolute way. Yet it seems that the aims of the Center for Human-Compatible Artificial Intelligence are to accomplish the closest thing to it.
Science
Artificial Intelligence
Robotics
Science News