With robots being introduced all the time to take over from humans in doing different roles, they are becoming more and more advanced as the need to complete more complex tasks arise. With all the knowledge that we are giving to robots, should we as mere humans be worried? With Google producing a paper this week called, “Concrete Problems in AI Safety“, perhaps the answer is yes.
The paper was written with the intention, not to scare people, but to just make them aware of the current and potential issues with welcoming robots into our society. It is essentially a guideline as to how to get the most of your robot but without creating any unnecessary problems.
Five main concerns were covered in the paper. The first being safe exploration and if its possible to ensure your robot can navigate safely through any environment. Secondly, the paper discusses the machines’ robustness to distributional shift and how it can house a system robust enough to deal with changes in data distribution.
The next two points that are talked about involve how to avoid negative side effects from the machine and also how to avoid reward hacking. The final main point of the paper was focused on scalable oversight and how we could develop a robot that would do its job in a way that would bring happiness to their employer/operator, rather than just getting the job done.
This may seem like quite a small list at the moment, and there may well be more added as time goes on. With more money being invested into robotics than ever before, it is essential that researchers and developers take the time to think of the consequences as well as the benefits that robots bring if we want to continue living in a relatively stress-free and safe environment.
More News To Read
