The Future of AI

John Defterios: I can't see how banning certain types of, let's say, reinforcement learning or certain agent environments would be either enforceable universally or a silver bullet against further advancement in AI. He says it's not immediately obvious whether the objective function, the reward function that we are defining and assigning to an agent is well-aligned with what we are after - particularly in the real world. And at the end of the day, everything in AI is based on objectives. If the objective is not well-defined, then what we end up with is not going to be well-behaved," he adds.