[em-bee]

before you can work on unique problems, you need to learn to solve problems that have already been solved. the problem then is that if a problem has already been solved, AI has been trained with the solution and is able to reproduce it.

think about it this way: any question or task you give to a student that is similar to a question or task that you gave students a year earlier, can be solved by applying the solutions collected from the previous year. and if that is the case, then a machine can solve those tasks too.

the amount of effort needed to design questions and tasks so that a machine can not solve them is way to big to be realistic for any subjects that are not novel. which is everything you learn in school.