# Should you do a PhD in AI? Is it worth it?

Canonical: https://tiptreesystems.com/blog/should-you-do-a-phd-in-ai

2026-11-13 · Martin Weiss

I finished a PhD in AI at Mila, and a friend asked me whether their kid should do one. This is my honest answer: three reasons not to, three reasons to, and the one thing to do before you decide.

![Should you do a PhD in AI? Is it worth it?](https://www.youtube.com/watch?v=sCsKfy__v2w)

## Transcript

### Should you do a PhD in AI?

If this video convinced you not to do an AI PhD, good. I've just saved you three years of your life. If it didn't convince you and you were always going to do it, awesome. I, you know, welcome you being my colleague. Like

### Intro

Hi, I'm Martin and I recently completed my PhD in AI at Mila. I was talking to a friend recently and they were asking because their their kid was considering whether or not they should do an AI PhD. And I think my answer is pretty simple. Basically do you want to be an AI PhD? Do you want to be that person?

Because if you want to be that person, then you should. That's how you do that. When you

### Reason not to 1: ten years

do a PhD in AI, it takes a long time. It takes on the order of a decade for you to go from I'm considering this process to I'm out the other end of this process. The other reason is if you do engage in the process, it takes so much grit and determination and it'll be so unrewarding for so long that you're going to have to break through walls internally and externally in order to finish and complete the degree.

### Reason not to 2: the pay

Another big consideration is you will get paid like Like they do not pay you well. There's a currently Harvard, I mean, I'm going to date this, but like currently a bunch of PhD students at Harvard are on strike. And they're on strike because of amongst many other things, like the low pay. I know in Canada for a long time the the funders, they did not tie the the stipends, the people receiving the grants that were going to the PIs to inflation. Inflation happens and who is left on the stick?

It's not the government, it's not the funders, it's not the PI, it's the grad student who is not able to buy groceries. Like that's the person who's on the hook for this. And you don't really have a lot of voice. I know from personal experiences being a lab rep, which is kind of like student government ish, adjacent um at Mila, and you don't have a lot of Like I'm an immigrant in this country, right? I have a permanent residency now, but at the time I was on a student visa.

You can't really demonstrate. You can't really Like going on a strike is really risky for you. If you try to make noise, I mean, yeah, that's really risky, especially if the home that you're going back to is um a country which, you know, might have civil war or be, you know, have other sorts of issues. Like it's a it's a big risk. It's a big risk for you to take.

Um and so there's a big power differential there and frankly, the universities and and funders, they know it and they pay like

### Reason not to 3: you don't need it

it. Another reason you shouldn't do a PhD in AI, you can do everything that you wanted to do in that PhD without it. Look at Chris Olah. He doesn't have a PhD in AI, and he's a co-founder of Anthropic. You can do that.

Well, not co-found Anthropic, but you can definitely go and just do research. And then if your research is good and you do that consistently and you go and you spend a lot of time bumping around, you make that your life and you're obsessive about it, you can definitely make being an independent researcher a successful thing uh in industrial research labs and find other people who are willing to support you. There's a lot of nonprofit opportunities around grants that are being given to people that don't have necessarily industrial or academic affiliations. There's a lot of opportunity out there, so you can definitely make that work.

### Reason to do it 1: the network

One of the other things that a PhD gives you is a fantastic social network. It doesn't necessarily give you this network, but it can. This is kind of the ceiling performance on how good doing an AI PhD can be. If you do it in the right place with the right supervisor, then you get access to people that are very, very hard to get access to otherwise. You end up having, you know, access to professors.

You can go to a conference. One of the fun things that I learned um while I was doing uh kind of market research to understand how information flows in society. One of the things that I learned was that um there's kind of this funny curve. So, you have on the y-axis how likely is someone to give you some information or to respond to an email, things like this. And on the x-axis you have how much money are you offering them?

If you have zero, then you have like some intersection, you know, with the y-axis, which is you know, let's say here, okay? And then if you offer them a dollar, it goes like this. And then eventually it comes back up on the other side. But for some people it just never reaches the same point. Like there are things that you could ask to great researchers and you know, famous people that they will not tell you if you are offering them money at all.

And so if you are a researcher who is participating, you're a PhD student who's trying to to learn something, there are a lot more people who are going to help you out than if you are in industry or you're going through an expert network or something like this. So, the type of social interactions that you get to have, the types of things that you can learn pretty unbeatable. That said, picking your PhD supervisor, picking the lab that you will be doing your PhD in, that is critical. So, you don't necessarily have this for free. Like um the ceiling is what I just described.

Like you can have that, but it doesn't mean that you will get that. There are tons of pitfalls that can stop you from achieving that goal of getting the social network and getting the learnings from mentorship. So, that's a whole topic for another video. We've set a videos about how you can avoid those kinds of pitfalls. Um but a topic for another day.

You should do an AI PhD if you have a great supervisor. Now, if there are basically two kinds of supervisor. The first kind is a more junior researcher, typically like an associate faculty member, who still has something to prove. Now, they don't all feel this way, but often times they're under a lot of pressure. And that means that they're applying for a lot of grants.

It means that they need that paper to get published, they need it to get accepted. So, they're feeling the same thing that a lot of you know, more senior PhD students feel. They're overwhelmed, there's so many things going on. Like when is this going to Now I have bureaucratic duties as well, like in addition I have a teaching load, like I have all of these things that are going on. I have to manage these students.

They have many things that are happening simultaneously. And you are just one of these things. And actually you're on the critical path for them. So that means that they need you to publish and be successful. That can create pressure which is at not always but can compromise the learning experience and your freedom to learn things.

The other kind of professor is the more senior, more tenured professor. Now again, these are generalities. There are many, many kinds and everybody is unique. But the second kind, the more senior professor, they tend to have a larger group. They tend to have more postdocs on staff.

They tend to have people who are more senior PhDs. They tend to have gone through this process many times already. A key thing to look for here is like how many students have they graduated and where did they go, right? If you're considering joining a a lab, like like how does that lab do? This is a way to measure that.

It's not just what has that professor done, it's what of the students who they mentored, what were their experiences like? Where did they go? What have they done? How did they like it? You should probably ask them.

That's like you know, if you're going to go and make a commitment to somebody for many years, you should probably ask these questions first. So they have a larger group, that means they're going to give you less time. They don't need that from you. You can go run wild. Um it also means that you're going to have more experiences interacting with other students, other postdocs.

You're going to be learning from other people who you're not necessarily on the critical path, but they're also not necessarily that senior. So you're going to rely more on the quality of the other members of the lab than you are on the quality of that particular professor. You're going to start picking up different ideas and that kind of cross-pollination is super important. Mila does very well on this. There are now hundreds of professors.

There were 70 people when I started and and even then like we had a good, you know, um half dozen, dozen professors who were working on adjacent field, coming together into a room. There is a very healthy culture of visiting scholars coming into Mila. So you're constantly hearing every week interactions between Yoshua and Simon Lacoste-Julien and Aaron Courville talking to a visiting researcher, 70 people in a room. It was fantastic. That's where like I learned how science works um because they had this culture where you could see the like that in that conversation you could see what they were processing.

Like oh this visiting person they have a different perspective. Is it legitimate? Like did they miss something? What what are they bringing to us? What can we learn from that?

Oh it it affects that project, right? So it's a it's a conversation that's bouncing around in front of you. Um so that was like a really wonderful way in which to learn science. In many cases I think it's good for a student to have experienced industry before they choose to go into a PhD. And the reason is that there's a lot of choices that you can make in your professional life and academia and industry are pretty different and you don't need to go into industry for six years in order to have like a nice clean like out.

You can go there for a year or 18 months and you can get a lot of understanding about what's going on there generally. So you don't get the experience of here's a customer, they have a problem, I need to solve this, I have a boss, my boss is telling me to do something else like I need to solve that other thing, I need to read this manual and understand that technology from whenever, I need to fix this bug, I probably shouldn't push to prod on Fridays because things are going to catch on fire like and then that's my weekend. Like you learn a bunch of other things that you don't learn in academia. So and you can learn those pretty quick. So it's typically a good idea if you're doing at least an AI and applied research like I did um that it can be a good idea to go and get some experience uh in industry before you join uh join a lab and start doing uh a masters or PhD.

Finally access to

### Reason to do it 2: the resources

resources. Like you actually get access to a lot of resources now um at least in Canada where we are um you know the Compute Canada which is now called the Digital Research Alliance, like um they have thousands and thousands of GPUs. Mila has thousands of GPUs. Like, you get access to these resources. You can run experiments that would be very difficult and expensive for you to run as an independent researcher.

So, that is to be considered. It's not like your salary is anything to write home about. That's actually a whole other video. But it is considerable if what you're trying to do is some intellectual endeavor and you want to get answers about some question. Academia is there's a worse place to be than like a well-funded AI lab in

### Reason to do it 3: the only door to professor

academia. An AI PhD is worth it if you want to be a professor. You can't really be a professor if you don't have a PhD. So, if that's what you want to do, that's what you want to do. Keep in mind, if you want to be a professor, a lot of it is bureaucracy.

Like, you'll be filling out random forms for a university, which is like a quasi government level institution. They are huge and they do not do paperwork in a nice way that makes your life easy. If you think you want to be a professor, go and talk to professors before you actually double down on that decision. Like, know that that's the path you want to take before you embark on it. And if you do want to be a professor, and this is a whole other topic, if you do want to be a professor, you might have to move countries.

You might have to move to an institution like it's very competitive. You might end up somewhere that is less desirable. You're not going to just wander into a very desirable university and get the job, even if you are like a stellar candidate. It just doesn't always happen. Some people get lucky, sure.

But it's not the typical case. Some people get lucky, sure. But it's not the typical case.

### One thing to do: email a PhD student

One of the thing, if you're trying to make this decision or break into research, go email a PhD student and one of the labs that you're interested in and ask what their Tuesday's like. Offer to buy them coffee. Read their papers first. Ask them questions. Do not email the professor.

Email the student.

*First posted on [YouTube](https://www.youtube.com/watch?v=sCsKfy__v2w), November 13, 2026.*
