His idea would be cool if not for the lindy effect: each one of his "tests" has a somewhat low probability of extending his life by a few months / years.
However each of his tests, as they are new, also has a smaller probability of having ruin effect: killing him or leaving him disabled in the process. Multiplying the treatments increases significantly the downside risks (1 failure is enough) while the positive will not compound (you will need many of them to work to see a significant effect).
This is also a fear I have, I am answering it with building bigger things with AI: what I loose in expertise in the details, I gain in the pleasure of architecturing bigger things.
I discovered "secret level" on prime recently where there is an episode on UT2004 (with the original sound effects like "killing spree !!!" ). With this additional news, I now want to run it again...
However each of his tests, as they are new, also has a smaller probability of having ruin effect: killing him or leaving him disabled in the process. Multiplying the treatments increases significantly the downside risks (1 failure is enough) while the positive will not compound (you will need many of them to work to see a significant effect).