One of the questions I’ve gotten most frequently over the years is some version of “should I do an AI PhD?” I’ve answered this enough times now — and gotten to see how many of these decisions played out — that I figured it might be useful to just write my thoughts down. I’m also about to finish my own PhD, so it seems like a natural time to reflect on the decision.
Obligatory disclaimer: this is based entirely on my own experience. You ultimately know what’s best for you, so as with all advice, take mine with a grain of salt.
There are already a number of good articles on whether to do an AI PhD or not (see this, this, this, this, and this). These generally discuss the “pros” and “cons” of a PhD, and I won’t repeat them here. Instead, I want to share a different framing that I’ve found much more useful for actually making the decision.
Optimize for fun
One of the best pieces of career advice I’ve read is this essay by Michael Black. I strongly resonate with everything he says: making the optimal career decision at every step is not something that can be planned out. There are too many unknowns, too much randomness, and too many things that will change between now and the future you’re trying to optimize for. Given this, optimizing for “fun” turns out to be quite a good heuristic.
So my high-level advice is straightforward: go do a PhD if it seems like the most fun thing you can do over the long term, and don’t do it if not. I want to emphasize over the long term here. A PhD involves a lot of delayed gratification — months of failed experiments, rejected papers, and uncertainty about whether your ideas will pan out. It’s not always fun day-to-day. But if the overall arc of spending several years deeply exploring research questions that fascinate you sounds like the most exciting thing you could be doing with your time, that’s a very strong and useful signal.
The real question
But how do you actually know if a PhD will be the most fun decision? Here’s the problem: a good PhD is a lot of fun, but a bad PhD is very much not. The experience varies enormously depending on your advisor, your research group, your topic, and the specific environment you’d be working in. So whether a PhD is right for you depends on the options available to you, not the theoretical options that might be available to someone else.
In my opinion, the biggest issue for people making the PhD decision is that they think too much in hypotheticals. They spend a lot of time asking “should I do a PhD?” when the much more useful question is “should I do this specific PhD vs. the other concrete options I have at my disposal?” This is a much more grounded question, and it makes it significantly easier to figure out whether the PhD path is the most fun long-term choice or not.
What to actually do
Given all this, here’s my concrete advice: if you’re interested in AI research and considering a PhD, you should almost certainly apply to programs you’re excited about. At the same time, you should actively pursue research opportunities in industry — residency programs, research engineer roles, or anything else that would let you do the kind of work you’re interested in. If you’re not yet ready for these/not getting a lot of traction (and that’s completely fine!), you should perhaps focus on finding ways to become a stronger applicant (a completely separate question).
The key idea here is that applying broadly gives you concrete options to choose between, rather than forcing you to decide based on abstract pros-and-cons lists. Once you have actual offers in hand — a specific PhD program with a specific advisor vs. a specific industry role at a specific company — you can much more clearly evaluate which one sounds like the most fun over the long term. And if you have a better option in industry, go for it! The goal isn’t to do a PhD for the sake of doing a PhD; it’s to find the path that’s most exciting for you.
What about delayed gratification?
A natural response to all this is to ask whether a PhD is worth doing as a delayed gratification bargain. Maybe none of your PhD options seem particularly fun, and you have more immediately exciting options — a high-paying industry job, a startup opportunity — but you’re convinced you need a PhD for the long-term career you want.
This reasoning can be sound, but the details really matter. How much more exciting are the other options vs. the PhD? How sure are you that a PhD is actually necessary for where you want to go? How intrinsically excited are you about your research topic, and how much are you just chasing the credential?
A few things to consider:
- You likely can’t go wrong. If you have several options that seem roughly equally exciting, there isn’t really a “bad” one. It’s impossible to pick the “optimal” choice given everything you know now, so make a decision based on whatever factors feel right and trust that you’ll course-correct if needed.
- A good PhD is often what unlocks the opportunities you want. AI PhD programs are demanding: classes, quals, mentoring, teaching, funding, and good research on top of all of it. The post-PhD opportunities you’re most excited about — in academia or industry — will likely require a strong publication record and great references. That’s hard enough to achieve in a research area you love, with advisors you enjoy working with. If you’re going in with a “power through and survive” mindset, the delayed-gratification bet might not pay off the way you hope.
- It might be worth generating more options. Many programs allow deferred enrollment, and many advisors are open to industry collaborations or internships during the PhD. It’s worth exploring whether you can find a setup that gives you something closer to the best of both worlds.
- PhDs are long, and the landscape will change. When I started my PhD, GPT-2 had just come out and most people around me thought it was a cute demo. Robots struggled to walk stably or pick up objects in controlled lab settings. The vast majority of robotics jobs were in self-driving, which was supposedly on the verge of revolutionizing transportation. The academic AI job market looked steady. Five years later, AI jobs are dominated by LLMs, self-driving hasn’t delivered nearly what was promised, and federal funding cuts have reshaped academia. Things have worked out well for AI PhDs overall, but this was far from guaranteed at the time. So I’d recommend against doing a PhD if the primary reason is some specific post-grad opportunity — you don’t know if it’ll still exist, or if you’ll still want it (lots of people’s career aspirations change during a PhD), by the time you’re done.
- Ask yourself whether you’d be happy with a random recent alum’s outcome. Look at where the last few students out of the lab or program ended up — their trajectories are the strongest signal you have for what yours could look like. Would you be proud if your own PhD turned out like one of theirs?
My experience
To make all this concrete, here’s my own story. When I was finishing up my undergrad, I was extermely conflicted about whether I should go to industry or pursue PhD programs. I knew I wanted to work on AI — specifically AI/ML for robotics — but wasn’t sure what the best way to pursue this interest would be. I was advised by several researchers I looked up to that, given my undergrad research experience, I should apply directly to relevant roles in industry, especially to AI residency programs. I applied to any and all I could find and didn’t get interviewed for any relevant full-time roles. The best I was able to do were to land two research internships at startups that my undergrad advisors had connections at.
At the same time, though, I had applied to several PhD programs that I was excited about, and ended up getting into a number of them. I ultimately chose my advisors at MIT because it seemed like the most fun option available to me. It also had the added benefit of seeming like a great long-term career option. Frankly, I felt I’d be crazy to turn it down.
Had I gotten into AI residency programs, or landed a full-time research engineer or research scientist role, I might have gone straight into industry instead! But the point is that option was simply not available to me, even though it existed for other people. The hypothetical of “should I do a PhD or go into industry research?” was not the right question for me, because I didn’t actually have the industry research option. The real question was: “given the specific options I have right now, which one sounds the most fun?” And for me, the answer was clearly the PhD I chose.
While my PhD overall had several challenges and periods of significant struggle, it did turn out to be extremely fun overall (as I had originally hoped!). I got to work on interesting problems with people I like and admire. I got to spend time to understand a lot of concepts I only knew at a surface level in undergrad. I also got more clarity on whether I want a career in academia or industry over the long-term. At the end of my PhD, I was able to successfully recruit for several jobs in industry that I could only have dreamed of as an undergrad. In hindsight, choosing to do a PhD was probably the best decision I’ve made.
I think this framing — ground the decision in your actual options, and optimize for long-term fun — makes what can feel like an overwhelming life decision into something much more manageable. You don’t need to figure out the globally optimal career path. You just need to explore your options, and then pick the one that concretely excites you the most.
If any of this has been helpful, or if you’d like to share your own experience with making this decision, feel free to reach out — I’d love to hear from you.
Thanks to Josh Roy, and Tom Silver for reviewing early drafts of this post and providing helpful feedback.