Going to at least change research. AI is honestly some secret sauce in research. If prompted correctly, can help with everything from hypothesis generation to study design to creating graphs in R to telling you what code to use for statistical analyses to telling you what the optimal statistical analyses are. Oh, and it is trained on the entire corpus of scientific papers written by humans, so it might be able to help with that a little. I don't know when it will become an autonomous research agent, but IMO, it lowers the bar to entering and performing (clinical) research significantly and makes you MUCH more efficient. Note that the benefits derived are entirely dependent on which model (ChatGPT 4 with vision is good, couldn't imagine trying this without vision) and your prompt engineering abilities. The latter point is the secret sauce.
Anecdotally, I know very little about statistics and coding, yet I was able to independently run multivariate analyses in R with large datasets and create professional looking ggplot2 figures in R. Granted, I was probably a little slower compared to somebody that knows what they are doing, but with no AI, I would literally have not been able to do those things without hundreds of hours of trial and error.