A few years ago, hardly anyone in the computer science profession would admit to working on “artificial intelligence”. Not unless they had tenure. Sure, people would say they worked on robotics, natural language processing, voice recognition, or other respectable application areas. But saying you worked on AI was like saying you were a composer: it implied that you were crazy, pretentious, or both.
So it’s funny to live in an era where AI is cool again, and only us old-timers remember the carnage of AI winters. In fact, AI is so cool today that many students are asking how to prepare themselves for careers in AI.
Here’s some advice from an old-timer:
- Understand that AI is a quickly-changing field. Take neural networks as an example. They were hot in the 60s. Then, in 1969, a theoretical analysis of perceptron networks convinced most people to abandon them. Algorithmic advances in the 1970s and 1980s (specifically, the backpropagation algorithm) brought back some interest, but it really wasn’t until the advent of recurrent neural networks — more popularly known as deep learning — a few years ago that neural networks made a comeback. Now deep learning is all that people can talk about!
- Invest in the fundamentals: math and computer science. AI is a broad field, but all of it is theoretically grounded in the mathematics of vectors, matrices, probability distributions, and optimization. Moreover, AI is computationally intensive, relying heavily on distributed computing and specialized hardware for efficiency. You can’t predict which AI techniques will still be around when you graduate, but you’ll reap long-term returns from acquiring a firm grounding in math and computer science.
- Play with real data — preferably through internships in industry. The magic of AI happens when you extract patterns from noisy, real-world data. As much as you can learn from applying textbook algorithms to toy data sets, you’ll never internalize the challenges of AI until you try it in the real world. Get your hands dirty with real data at a meaningful scale, ideally through internships in industry. That’s the only way you’ll develop an intuition for how AI techniques work in practice.
Should you take an AI class, or classes in specific techniques like deep learning? It can’t hurt. But recognize that your syllabus will probably be antiquated in a few years. Accept that transience, focus on the fundamentals, and seek out opportunities to play with real data.
And get a warm coat — you never know when the next AI winter will strike.