The problem is, I don't think you can really claim that Copilot learned to program. While some of the output seems to be something new, most of the times it looks more like a recomposition of learned fragments if not even longer pieces of verbatim code taken from copyrighted material. We have seen examples of this. And in this moment, it becomes a copyright discussion, probably determined by the volume of copyrighted material reproduced. Which by the way is always the risk if a human uses certain training material. The better one is at memorizing things, the more there is the risk.
Or put it the other way around: if Copilot would use its "knowledge" of programs to advise the programmer like pointing out potential errors without reproducing anything it used for learning, it should be fine. But that is not how it works.