It will be machine learning algos that help replace those jobs.
"AI" is only as intelligent as the set of algos a programmer can write or feed it. The whole self learning, self awareness, human intelligence type stuff some people associate with AI is still a lot closer to science fiction than science fact.
One of the key differences between machine learning and regular computer program is machine learning requires far less human programming. For example, Windows 11 contains about 50 million lines of code, which are all written by humans. Machine learning requires less than a thousand lines of human code. the key output is the matrix of numbers that are created during the learning process. At it's core, machine learning is based on linear algebra, specifically, matrix multiplication. It's just math, but math that computers excel at. In particular, GPU's are very fast at this kind of math because all graphics just calculating polygons using linear algebra.
The output of machine learning is a large matrix of numbers.
Equally important is the hardware to run the machine learning model on is low cost and simple. Tesla FSD runs on a computer that's basically two Samsung galaxy chips and half a gig of memory. The file containing the matrix is only about 500 megs, it's easy to send over the air updates and every Tesla has the required hardware.
The hardware that does the training is basically thousands of GPU's for gaming. That part of it is expensive and power consuming, but once the training is done, it can be sent out to everyone.
Equally important, the computing power needed to run the
