I had a conversation with some friends about what research in our fields might look like in twenty years. This was partially sparked by all those claims of chatGPT reaching the level of a PhD student.

So I tried to imagine what research in my field would look like if you had a bot that could write code for bioinformatics perfectly. If that were the case, the PhD student’s role might become more akin to that of a manager’s: figuring out which research directions to pursue, and guiding the AI to execute that vision. All of the annoying, technical details are able to be relegated to an AI, and the student no longer needs to bang their head against segmentation faults or frustrating file formats.

But the problem is, those small, day-to-day frustrations and details are some of the reasons why I love doing this type of work so much. There is just something innately satisfying about being the one to translate my algorithms or ideas into code. I love the process of developing a code base that slowly evolves from a toy prototype to a fully-formed tool. I love that thrill of finally figuring out a good solution to an algorithmic or engineering problem, or the instant jolt of relief and/or pure joy that overcomes me when I fix a particularly nasty bug. I love finding ways to optimize my code or make my tools more robust.

But what if there was an AI that could do all of these things instantly if I prompted it correctly? Would I have the same passion for research?

The answer is no, I don’t think I would enjoy research as much. Because if that AI could write objectively better code instantaneously, then it’s hard to justify coding anything myself. I do also enjoy the process of identifying problems and developing methods theoretically (if not, I would not have gone to grad school), and these are parts of the research process that are less prone to automation. But I still feel considerably less excited if I imagine I am not the one who would be implementing methods myself.

I was trying to figure out why this sort of advancement feels different from when e.g., AlphaGo beat Lee Sedol in 2016. It’s not like AlphaGo won and then suddenly everyone quit playing go. And when I frame it as coding just being an activity I enjoy in the same way I enjoy playing go, then one could say that there’s no reason to stop coding as a hobby, even if I end up having to use AI for work for efficiency purposes. But I guess the point of coding has always been the creation of an objectively robust product, whereas for go it has always been the experience of the game itself. So then coding would feel meaningless in a way playing go doesn’t, even if I still enjoy it.

If that end is inevitable (which I don’t believe is necessarily true), then am I wasting time pursuing a PhD with such a strong focus on developing high-performance code for tools? I would have thought the answer was clearly no (or perhaps I am just too invested at this point to say otherwise), but I actually do know people who believe this is a waste of time and focus.

However: 1) I am very much enjoying the PhD journey for now; 2) these problems currently still need to be solved by humans; and 3) even if such a powerful AI comes to exist, there needs to be someone with enough domain knowledge to steer it and check its work. And even if I don’t enjoy that sort of thing, at least it can improve the other aspects of the research process that cannot be automated away.