I'm not saying it'll be easy, but I just think it'll be possible. Extrapolating patterns from few occurences is extremely common with human learning as well, ie a person "learning" that traveling is too dangerous based off one plane crash. And how many times have you seen humans NOT learning the obvious when the obvious is hitting them in the face? Its a problem with human learning but it is in no way a unmoveable obstacle. I don't see why it should be with machine learning as well.
Obvious patterns may not be obvious. And what happens if the machine starts extrapolating patterns from a few occurences?