Is it possible to judge that it's the face of a person who hasn't learned from artificial intelligence?

It is possible to create models and systems that infer that "this person is a person who has never learned before."Simply, if the model that judges A based on A's learning data, and the model that judges B based on B's learning data, all of them are false, you can judge them as "people who have never learned."

Whether you can draw the inference that "this person is a person who hasn't learned before" from a model or system created for a different purpose depends on the original purpose, but it's not a good way to use it in general.For example, using data from both Mr. A and Mr. B to determine who is more likely to be Mr. A or Mr. B will not produce good results.

Yes, it's possible.Deep Learning is often used these days.Rather, it is already becoming a red ocean.Why don't you refer to the slides below?

• Deep Learning and Image Recognition ~History, Theory, and Practice~

For implementation, I think the following articles are interesting (although these are facial recognition in anime, not photographs, the principles are the same).

• First Animation Face Recognition with Chainer
• Love live with Deep Learning!Identify Characters
• [Chainer] Identify anime characters in Deep Learning (add a little bit on November 4, 2015)