Machines learn and work in much the same way as a person or a pet sometimes learns– by receiving rewards, also known as Reinforcement Learning (RL). That is, when a proper or desired behavior is produced, a reward is given. This is what Google is using to teach their DeepMind AI.
For this particular learning, Google wanted to know if they could teach DeepMind to navigate an obstacle course, and if a complex terrain made learning to move easier and produced higher quality results.
The research time paper states, ‘Our experiments suggest that training on diverse terrain can indeed lead to the development of non-trivial locomotion skills such as jumping, crouching, and turning for which designing a sensible reward is not easy…We believe that training agents in richer environments and on a broader spectrum of tasks than is commonly done today is likely to improve the quality and robustness of the learned behaviours – and also the ease with which they can be learned…In that sense, choosing a seemingly more complex environment may actually make learning easier.’
The result is impressive and mesmerizing to watch. And, perhaps a bit terrifying; see how in the first clip it tries and tries again until it succeeds. AI is now getting very smart, smart enough to transverse a complex obstacle course, and smart enough to program itself at least as well as humans can program.
Leave a Reply