The main issue with 'AI' is that it make an judgment by the data it was feed on through the training process. So there is no spark of creativity which will result in behaviours which are far away from expected value.
Unpopular opinion: I know that hype train change its direction and know we are stuffed with buzzwords like 'Ai algo solve mathematical problem', but the question is how the test dataset was prepared and where is the border between excellent memorization and average creativity. Because assuming that we have an infinite number of resources we can teach any algorithm literally anything in limited amount of time.