Research Without Black Boxes
I recently sent two of my students an email to further clarify two points I had made to them earlier that day, because I think they are important. I am posting it here, lightly edited, so that future students know how I think about research.
These are my personal views, and you certainly do not have to agree with them, but I strongly encourage you to seriously consider the mindset behind them.
Do not make your success dependent on other people
Let me make this point deliberately sharper and more extreme: work as if your collaborators are idiots.
I do not mean this literally, and I am not referring to any particular person. What I mean is that you should never put yourself in a position where the quality, progress, or eventual success of your work depends critically on someone else being reliable, capable, or fast.
Assume that if something important needs to be understood or done, you should be capable of doing it yourself. If somebody spends a month working on something, I want you to have the confidence that, if necessary, you could understand it deeply enough to do a better version in a day. If it takes someone six months, ask yourself whether you could get to the core of the problem in three days.
Of course, those numbers are intentionally exaggerated. Collaboration is important, and good collaborators can contribute expertise and perspectives that you do not have. But the underlying attitude is one I genuinely believe in: do not outsource ownership of your success. Develop enough independence, confidence, and technical depth that other people’s limitations do not become your limitations.
This may be a more extreme mindset than most people would endorse, but it is much closer to how I think about research, and I believe some version of it will benefit you.
Apply first-principles thinking to your research
Try to approach research in the spirit of first-principles reasoning often associated with Elon Musk: break a problem down until you understand what is actually happening underneath the terminology, conventions, and abstractions.
Sometimes something seems difficult or mysterious simply because you have not yet opened the black box. Do not leave parts of your own research mythical to you. If there is a model, theorem, algorithm, assumption, experimental procedure, or mathematical argument that your work relies on, keep digging until you understand what it actually does and why.
Do not be intimidated by a topic because it belongs to another field or because people describe it as advanced. You have both the time and the ability to learn things that initially seem outside your expertise.
The way I criticized a paper in our AI/ML seminar is one example of the attitude I mean. I did not want to accept terminology or theoretical claims simply because they sounded sophisticated; I wanted to unpack them, trace the underlying theory, and determine exactly what was justified and what was not.
I showed my students that example not because I expect them to imitate me, but because I expect them eventually to do this better than I do.
That combination—intellectual independence and a refusal to leave important things as black boxes—is something I hope you will carry into your research.
