Confession of a so-called AI expert
[Twitter thread, Hacker News discussion, Le Monde]
I have a confession to make. I feel like a fraud.
Every few days, I receive an email from either a friend, a friend of a friend, or a random company that asks me for my insights in Artificial Intelligence. These include entrepreneurs who have just sold their startups, Stanford MBA graduates who reject half a million dollar offers, venture capitalists, even major bank executives. A couple of years earlier, I wouldn’t even have the courage to approach those people, let alone dreaming about them wanting to talk to me.
Their narratives vary among one of these options: “Connect me to people in AI?” “Come do AI with us?” “Advice on our AI products?” They talk about AI as if it’s the fountain of youth that everyone is jumping into and if you don’t you’d be old and left to die alone. They somehow think that I know how to get there.
I get why people think that I’m an expert. I’ve spent years learning how to craft a perfect resume. Elite education? Check. Famous professors? Check. Big companies? Check. On top of that, I teach a Stanford course that sounds like the holy grail of all that’s so hot right now: TensorFlow for Deep Learning Research. A French company, during our interview, told me that they ran hundreds of resumes through their algorithm and mine miraculously landed on top.
But the thing is, I’m not an expert. I’ve just finished the third year of my undergraduate degree. I haven’t got published on any journal. I’ve never attended any AI conference of any sort, mostly because I’m too poor to afford those. Okay, I lied. I attended one, but still.
I always feel a tremendous pressure when I tell people I’m taking this or that Computer Science (CS) class and they go: “It must be so easy for you.” No, it’s not. I struggle just like any other student, if not more. I came to Stanford thinking I’d major in journalism or something social science-ish, then I took a CS class and thought it was fun. When I signed up for my first calculus class, the professor called me up to his office and told me I should take an easier class because I obviously didn’t have the prerequisites for this one.
I used to love office hours. During my CS103, CS109, and CS221 classes, I practically camped at those office hours, never missed a chance to ask the TAs questions. But now it feels awkward because occasionally a TA or a fellow student would say something like: “Shouldn’t you already know this stuff? Don’t you teach a class on this?”
The assumption that because I teach a course on TensorFlow, I must know everything about AI is tiring. I didn’t set out to teach the course because I was an expert in AI, or in TensorFlow for that matter. I wanted to have a class on it so I could learn with people who shared my interest. Since nobody else wanted to teach it, I had to.
I was struggling through the entire course. There were many nights I couldn’t sleep because I was freaking out about what to show my students who already knew more than I did. More than half of my students were master’s and PhD students. Once, it was 9 am and I hadn’t been sleeping or eating for 24 hours and there was a bug in the demo I was planning on showing to my students that same day. “I’m a fraud,” I whined over the phone with my boyfriend who was at that time driving around New Zealand. He is, fortunately, one of the best coders and researchers I know. He calmed me down, told me to get some sleep, and promised to look at my code at the first Internet spot that he found. When I woke up 4 hours later, my demo was working. I wouldn’t have been able to go through the course without him, or Tessy, or David, or many other friends.
I got a lot out of teaching the course. I knew so much more about TensorFlow since I had to anticipate every question my students might ask and googled the shit out of them. It gave me an excuse to reach out to people I admired, who kindly agreed to either proof-read my notes or give a guest lecture. It forced me to learn proper coding style, since I couldn’t bear to publish ugly codes that my students would laugh at. They probably still laughed at my code, but whatever.
Like my friend Delenn said, I have jobs thrown at me. Big companies want to talk to me. Startup CEOs and CTOs go out of their way to meet me. I’m so spoiled that most of the times I don’t even read recruiters’ emails. It’s like the world has turned upside down. A couple of years ago, I was rejected for every single CS job I applied for. An anecdote I find funny is that a company rejected me when I interviewed with them. After my course got on the front page of HackerNews, they emailed me asking if I’d still be interested.
Of course, I have improved over the last two years. But I’d be foolish to think that the change in recruiters’ attitude is due solely to my improvement. Many of my friends, who are way smarter than me and can unceremoniously kick my fuzzy ass in anything CS-related, struggle to find a decent job just because their resumes lack trending keywords. Yet many others, who barely understand basic concepts of machine learning, swim in job offers just because they’ve completed courses with fancy sounding names. It’s a phenomenon that Richard Socher, the dishevelled 30-something (or 20-something?) lecturer who just sold his company for several hundred millions yet still biked to campus, mentioned in his class: “Companies keep asking my students to drop out to work for them.”
This thirst for AI has pushed all AI-related courses on Stanford to way over their capacity. CS224N: Natural Language Processing with Deep Learning had more than 700 students. CS231N: Convolutional Neural Networks for Visual Recognition had the same. According to Justin Johnson, co-instructor of CS231N, the class size is exponentially increasing. At the beginning of the quarter, instructors for both courses desperately scramble to find extra TAs. Even my course, first time offered, taught by an obscure undergraduate student, received 350+ applications for its 20 spots. Many of the students who took these courses aren’t even interested in the subject. They just take those courses because everyone is doing it.
There are also people who capitalize on it. AI bootcamps, AI courses, AI conferences mushroom everywhere. Companies roll out expensive courses that teach things you can learn on your own. AI conferences charge thousands of dollars for a couple of days. Some of my friends get hundreds of thousands, even millions, of dollars to work on their AI startups – even if when they don’t even yet have a working prototype.
Even though I’m one of the beneficiary of this AI craze, I can’t help but thinking this will burst. I don’t know how and when, but I have this belief that the system is currently being rigged in favor of people whose resumes dotted with fancy keywords like mine, and a rigged system can’t be sustainable. Maybe one day people would realize that many AI experts are just frauds. Maybe one day students would realize that their time would be better spent learning things they truly care about. Maybe one day I would be out of job and left to die alone on the sidewalk. Or maybe the AI robot that I build would destroy you all. Who knows?