Of course, if an AI program is given the entire resources of the internet or relevant study books to take the test, then it's an open book test for it, whereas the pass % is defined by closed book tests, so it's not doing that great if it is getting only 50%, as it did on a medical licensing exam with a pass rate at that level. Even so, that doesn't mean it can do the job of a doctor. We already had books to consult before there were computers because our memory and reading capacity are limited. You need judgment. You need to know the relevant issues. In some areas, the AI program's achievement is a combination of expert support and correction and the program's learning through their input. The poetry attempts by one AI program were aided by the choice of poems to input and then the bad language, the bigotry, and the disconnected first/second lines were edited by the poets, and the bad but not as bad poetry that resulted from the program's adjustments were therefore a function of human input and AI.