So it turns out the problems of AI are the problems of technology in general. Specification (making sure it does the job you want); robustness (the ability to cope with unexpected conditions); and the ability to monitor its performance and understand what it’s doing.
These are hard problems, even for a system designed from the ground up by humans, let alone one that has partly built itself.
Originally shared by Singularity Hub
DeepMind’s New Research Plan to Make Sure AI Is Safe
Leave a Reply