Many students who are just now embarking on a PhD in AI are understandably anxious. It feels like our field is zooming forward ever faster. Our clever methods are often surpassed before they are published. Our deep insights seem less critical to frontier lab progress. AI is automating aspects of the research process that used to require years of specialized training.

In this age, then, is there still value in doing a PhD?

I think there is! In fact, I think this is one of the most exciting times to begin such a journey. Here are ten reminders I tell my students (and myself!) as the following questions inevitably bubble up:

1. "How can I contribute anything if I'm not at a frontier lab?" You have arguably more leverage on shaping the future if you do your work in public rather than in secret.

2. "Anything I do this year will be done better by an AI next year." To get from this year to next year might require your work. You get to shape next year by doing your thing this year.

3. "Okay, but how will I get credit?" We are going to have to rethink credit assignment. Today's AI models are leveraging so many people's work: they might call your method as a tool, or they might distill your insights into their weights. You can take pride in that. In the end, the achievements of a frontier model should belong to all of humanity. I know, easier said than done.

4. "How will I make money?" You're doing a PhD in AI. You're in about as good a position as anyone to adapt to whatever comes next. The skills you are acquiring are transferable to many future kinds of work.

5. "I heard that AI will cure cancer in N years. I'm working on algorithms for cancer diagnosis. Should I do something else?" N years is a lot of lives. There's value in saving those lives. There's value in the present.¹ And the future is uncertain. What if N years becomes 10N?

6. "Why should I develop expertise in a topic when I will never reach the depths of understanding that an AI can have?" Human expertise is going to matter for a long time. I don't think we are ready to just throw up our hands and trust the machines. Even if they become smarter than us, or perhaps especially if they become smarter than us, we will want humans in the loop who deeply understand each piece of the puzzle.²

7. "Is there still value in doing slow science? How can I keep up?" Progress is fastest when you first take the time to deeply understand your subject. The rate-limiting factor is not technology but rather the human brain (and our social systems). Let's say it used to take N hours to become an expert in a subject. It will still take ~N hours in the future.³ The human role in research, which I think will persist, can't accelerate that dramatically. The humans who do this role best, whatever it may be, will be the ones who put in the time to understand the material deeply, and to think carefully through its consequences.

9. "How can I embark on a 5 year journey when the world is going to change so much? What if what I set out to do doesn't make sense in a few years?" This is the beauty of a PhD. Your job is simply to seek the frontier of knowledge.⁵ If the frontier changes, you get to change with it. What an enviable position in times of great change! So don't worry, in research there is never a clear path forward, you are always looking out into the unknown. That's the whole point of it. If you knew where you were heading, then what could you learn?

10. "Why should I do a PhD?" At the end of the day, a sufficient answer can just be this: for the love of the game.

Thanks to members of my lab at MIT for helpful discussions that led to this post.