Ask HN: Should I pursue a second PhD in AI if I want to do industry?

[jdaaph]

I'm currently in my 3rd year PhD student doing computational physics in a top-10 US university. I have always wanted to start a company with some AI / automation type of tech. I know that a physics PhD is definitely not the ideal way to go but as an international students there aren't too many choices. So I'm wondering, if my ultimate goal is industry => start my own company, would it be worth it if I start from now preparing for 1-2 CS publication and apply for a second PhD in AI after I graduate as a physics PhD?

I know people say, if you aim for the industry, don't go for a PhD, but AI requires way more top-notch research skills than most programming works and being in a top school (like Berkeley, MIT or Washington) gives you so much more opportunities to lead a team or be in charge of a technology once you graduate, all of which are opportunities that I can't have, being just a fresh PhD out of a physics department. (though I'm sure I can find a job pretty easily)

[philipkglass]

I would say no to the second PhD. If you're motivated enough to get one PhD and to start your own company you're motivated enough to do some research on the side (if necessary) while working in a software job. You'll be a lot better able to save up money for your future company if you are working in an industry job, and in industry you'll acquire kinds of experience that are hard to get from any PhD program. Finish your current PhD, get an industry position, and once you're settled in start doing your own side self-education and research. (Assuming that your day job doesn't naturally afford you opportunities to further sharpen your interest in and knowledge of specific AI areas.)

Research publications aren't important if you want to start your own company. You're not trying to impress yourself enough to get hired by yourself. But doing research-y things, like finding a recent research result that interests you, reproducing it, and trying to improve upon it, still might be important. Even better, IMO, may be looking at recent research results from groups that share source code, and testing your skills at turning research-quality source code into industrial-quality code. That might have to wait until you've been in industry a few years and have a better feeling for how robust production-quality code differs from research code.

[mswen]

No, do not get a second PhD. Your opportunity cost is too high. In particular since you are doing computational physics your computational modeling, statistical analysis and understanding of high performance computing should exceed the average AI founder. Second, you will already be capable of self-study of AI type academic publications and you want to be focused on translating those into real life applications not writing grant proposals and other graduate level academic stuff.

By the time you finish a second PhD you might be well along the pathway to business success, or you might fail. But I am convinced that adding a second PhD won't noticeably reduce your risk of failure.

I am almost certain that given the technical capabilities you have already developed that the risk factors in starting a business are product-to-market fit, timing, poor PR, marketing and sales or bad co-founders. By the way I have experience with all of those risk factors - they are real and technical skills or even a second PhD will not make them go away.

[stevenwu]

I am only a mere MSc Statistics student in a Canadian university, but from my perspective, if you are succeeding as a physics PhD in a tough and respected program, then you are more than equipped to do anything you set your mind to in the tech sphere. I'm interested in AI/ML too and I read papers to try and keep up - I'm surprised at how many ideas and methods have ties to statistical physics and other domains that you're probably already comfortable with as a physicist.

In my city's (Vancouver) Slack chat, a lot of the most knowledgable participants in data science discussions are physicists (and they're proud of the fact that they're physicists). Check out the Director of Data Science at Unbounce, who published papers and worked as a postdoc as a physicist before starting in industry data science at PlentyOfFish. Worst case scenario if you find yourself pigeonholed, find a home in Vancouver with like-minded people.

I'd say finish your PhD, skip the AI PhD and go for your ultimate goal ASAP!

[mindcrime]

I'm working on a startup that has a heavy focus on machine learning and AI and I don't even have 1 PhD. Or 1 Masters. Or 1 B.S. I just have 3 Associate degrees. I wouldn't worry about it. Worry about the knowledge, learn on your own and don't over-think the need for formal credentials.

[benologist]

A successful company you build could be hiring people with PHDs years faster than getting one yourself.

[jdaaph]

Thanks!