What courses should I take as an undergrad at MIT if I want to do data science? And is Course 6-3 (Computer Science & Engineering) good preparation for doing data science?

Course 6-3 is fine. Along with the other classes mentioned, I recommend taking 6.437 Inference and Information. While the AI and machine learning classes will teach you the techniques to process your data, in 6.437, you will ask yourself, is the data even good? You'll learn about entropy, mutual information, KL divergence, and the inference techniques that leverage these information-theoretic quantities. While the knowledge might not be as practical, it was extremely enriching in understanding how one or a system can perform inference. A word of caution though - the class is very challenging and frequently has some of the highest difficulty ratings in course evaluations. Despite spending way too much (well-spent) time on psets, it was one of my favorite classes at MIT. Plus, the professor, Greg Wornell, was a complete boss.

As a non-comprehensive answer, I would recommend 6.036: Machine Learning; it teaches exactly what it sounds like, and is very much involved in data science. With regard to course, it all depends on how you decide to pursue your major; the only 6-3-required classes that I would consider to be related to data science  are probably algorithms (.006, .046), probability (.042), and AI (.034) or Machine Learning (.036). The other absolutely-required classes, e.g. 6.01, 6.02, 6.004, 6.005, and 6.033, don't really provide much insight into data science. That said, I would consider 6-3 to be a good major for data science at MIT, second only to 18C. Of course, how you decide to use your elective courses may very well be data science-centric. Hope that helps :) Edit: Reading Yang Yang's response reminded me: There's also the 6.008: Intro to Inference course that I have heard is good, and might be worth looking into. It's taught at an undergrad level, and provides good exposition into 6.437 (again, all from rumors)

A2A. I'm not from MIT (hey look down on me, I'm from Rutgers, bitch) and I still don't understand the precise definition of data science, so my answer may not be as useful. I just like the idea of collecting credits when they are up for grabs. There are already good answers from those in the know, so I don't really have much to add, but I assume that you take essential math courses like calculus and linear algebra as a CS major. I took them because I was a physics major, and that's pretty sufficient for reading graduate-level machine learning textbooks. If you really want to be a hard-core data nerd though, I would actually recommend taking probability and statistics courses and even do a double major if you can. There's an impressive amount of knowledge that you can gain from statistics when you deal with data of any kind, so there really is no reason not to study statistics well enough for you to use it.