Graduate Certificate in AI: Achievement Unlocked

I recently received some very kind congratulations from friends and colleagues for having completed the Graduate Certificate in AI from Stanford. The intent of this article is to clarify some points about the Certificate as well as give advice to others who might be interested in pursuing the same thing.

First, this is not a degree. Rather, it is a 15-unit program comprised of 4 graduate-level courses (1 is required and 3 you get to pick from a list) in the CS department. Here are the ones that I took:

• cs221: AI Principals & Techniques is the required course, taught by Dorsa Sadigh and Moses Charikar
• cs229: Machine Learning, taught by Andrew Ng, Moses Charikar and Christopher Ré
• cs230: Deep Learning, taught by Andrew Ng
• cs234: Reinforcement Learning, taught by Emma Brunskill

As you can see, these are not MOOCs. While I'm a huge fan of online courses, this is not that. As an SCPD student you may participate in all of these courses virtually, but they are 'real' courses that take place during working hours in auditoriums and classrooms on the campus.

I've previously written about how I believe Product Managers interested in Artificial Intelligence should devote themselves to learning the fundamentals of the technology. That being said, I also mentioned that I felt that a Graduate Certificate like this might be overkill. Let's explore that further.

The Cons

Before I start selling PMs on why they should pursue this program, there are some obvious downsides worth nothing.

First and foremost, these courses are a staggering amount of work. The histogram to the right is of the number of hours per week students devote to cs229. While this course is notorious for its difficulty (the math is fierce), the vast majority of these students are full-time in the masters or PhD programs, with a few precocious undergraduates. If you have a job (and a family!), finding the time required to pass (you need a B) is not going to be easy. For cs229 I was most likely in the ">35 hrs" group, although for the others I was closer to 20-25 hours per week. Regardless, the tempo is relentless and you cannot go at your own pace. The impact on my professional and family lives was significant. (I cannot thank my wife enough for her patience and understanding.)

<update>The folks at SCPD read this article (well, I know of one who did) and decided (spurious cause and effect) to create a professional version of the AI Certificate. My understanding is that it's 3 courses instead of 4 and there are no exam or final projects. You can also skip one homework. And it's a lot less expensive. I'm about to be a course facilitator for xcs234, so I'll have more details soon.</update>

The amount of time required to take the courses, however, will naturally depend on your ability and familiarity with the prerequisites. While I have an engineering degree, the farther you are from your studies the more difficult the program will be. For each of these courses I did quite a bit of prep work (taking related online classes), but the material was still really challenging. If you are not very comfortable with Calculus, Linear Algebra and Python, among other things, this will be a rough ride. There are, however, hundreds of ways to get comfortable through free online courses. With patience and determination, you can get ready.

Regardless of your ability and background, however, there is simply a very large, non-compressible amount of work. On top of the lectures, TA sessions, office hours and study groups, the homework is voluminous, there are final team projects and, naturally, the exams are not easy. I got 53% on the cs229 midterm, which was exactly the median. I'm very proud of that.

Lastly, the courses are roughly $4500 to $5500. Check with your employer to see if you can get reimbursed. I was able to get approximately half of mine covered.

The Pros

Simply put, these courses are awesome. All of them are extraordinary opportunities to explore AI and they are all exceptionally well-thought-out and instructed.

While the homework is really long, as noted, the assignments are very clever and offer a wonderful window to exploring the material. They are also a lot of fun. In cs221 we wrote an algorithm to play PacMan. (There was a class competition to achieve the highest score. I came in around tenth place.) In cs234 we wrote an RL algorithm to play Pong at a super-human level. Watching it go was amusing and satisfying. The cs230 homework was actually on Coursera, but is was excellent. The cs229 homework was very math heavy, but we developed some fancy algorithms which produced beautiful plots, like the one here.

The students are also great. My office at PARC was right around the corner from the campus, so, pre-pandemic, I was fortunate enough to be able to attend many of the after-work office hours and study groups. I was constantly impressed by not only the dedication of the students but their willingness to help each other. On Friday nights the Huang Basement was consistently packed with people doing homework. While collaborating remotely (which I did for cs234) is more difficult, I still came away with lasting relationships from all the courses.

Lastly, should you decide to pursue a masters degree at Stanford, and are admitted into the program, which is certainly not guaranteed, all of the credits will roll over. After completing the Certificate, you will already be 1/3 of the way there.

The end result is that you will acquire both a broad and deep understanding of the algorithms and techniques which power AI.

As a Product Manager, is it worth it?

Yes. For me, absolutely.

Depending on your role or what you would like to do, it would be easy to make the argument that this is overkill. You will, after all, not be coding SVMs or running Bayesian hyperparameter optimization. There is a decent chance, however, that you will be working with people who do these things. Having an understanding of their work and empathy for the challenges they face will enable conversations that you might not otherwise be able to have. It will also give you insights into what is possible and what is required to get there.

AI is the future. If you have the prerequisites and the drive, and can muster the time and energy to plow through these courses, this program is highly recommended. While I firmly believe in learning for the sake of learning, if you are or want to be a PM in AI, this Graduate Certificate will hopefully put you near the top of the field. Post questions in the comments if you'd like to learn more.