Let's hope that the AI will improve for humanity's benefit

Sep 3, 2015 14:18 GMT  ·  By

Usually, robots learn from their mistakes, but they have to stop for a while to actually process their errors before engaging in further actions. Although normal operation processing is normal, improving on the fly is the next step in robotics improvements.

Called Integral Reinforcement Learning, it's a technique through which researchers Frank Lewis and Draguna Vrabie at UT Arlington Research Institute believe that robots will have their actions refined based on a previously made decision. In other words, robots will have to learn how to do a job in the future, without the slightest hint of tutoring or pre-programmed instructions, solely based on trial and error or by logical prediction.

This sort of self-learning or, better called, self-optimization is something that automated systems like planes auto piloting or car's emissions controls could be vastly improved. But it's robotics that will benefit most from this sort of optimizations.

Since robots don't usually adapt well to unexpected situations, the new integral reinforcement learning will allow them to adapt faster in unfamiliar contexts. They will improvise methods to overcome critical situations, and will have to make the best out of what they have at their disposal.

Although, at the moment the system is being tested and perfected, it's unknown what sort of device the scientists will end up with, but it will clearly be something a lot smarter, hopefully for our benefit, than what we have today.