Artificial Intelligence has taken great strides. The technology has evolved to the point where it can take over a host of tasks that were previously the sole domain of humans. On Tuesday, Ruslan Salakhutdinov, director of AI research at Apple, took the time to discuss some of the limitations of AI and how the tech wasn’t as omnipotent as it seemed — not yet at least.
Salakhutdinov joined Apple in October. Since then, he has been working with Apple in the field of AI research. In this particular area, researchers deploy AI technology to come up with the best possible result when there are multiple pathways available. Google for instance, used the technology to hit upon the best possible cooling and operating configurations in its data centers — leading to an increase in their energy efficiency.
Researchers at the Carnegie Mellon have also been teaching these computers to play video games. The idea here is ensuring that computers are able to learn things like the maze’s layout and gain the ability to shoot enemies. Eventually, computers also learned to duck to avoid enemy fire. All this occurred through reinforced learning. in case you are unaware of it, Reinforcement Learning is a type of Machine Learning and also a branch of Artificial Intelligence. The technology allows
machines and software agents to automatically determine the ideal behavior within a specific context, in order to maximize its performance.
So as part of this experiment, the software learned to reach the correct tower all on its own. Every time it reached the wrong tower, it back pedaled until it reached the correct one. The aim of this experiment was to get the computer to start recognizing the patterns all on its own. However, the main issue with these kind of software training is that it takes a long time to trains and it also requires a lot of computing power. Which is why the company is now working to ensure that it can create models that are able to learn by being run through lesser and lesser experiences.