Due to the amount of knowledge that exists and humans limited time and speed to learn there is an upper cap at how much performance a human can achive, yet a computer isnt limited by this in the same way.

Imagine a medical analisys taking into consideration all data from multiple exams and checking every medical study in existance is something humans could never do, simply because the amount of data the exams bring in is far too much and there is far more medical knowleadge already discovered than a human could learn. Hence a machine with all that knowledge would outperform any human doctor. (Obvious problem is building such a machine, but we have already studies of ML doing increadible analysis based on x-ray scans)

One risk is that the human knowleadge migh end reaching a cap since there isnt a incentive to have humans push the limits given that the models are already better than them. (One of the big problems that will likely be caused by image and text generating models, so we will actually have soon the results in a decade or two with how they influenced new artists, teachers and students)