Depends on who's programming it and the training data it's got. Material conditions (or in this case training data) determine "who" you are, so if the AI is reading/can read enough Marx and so on then we're in good shape. Would depend on what it determines is "correct." Honestly though, if your criterion is "make humanity better" or something broad like that, it's very probably an AI would just end up on FALGSC as the "solution."