How can I become a better researcher? After coming up with my own thoughts on this (I’ll outline those below), I searched for Google’s answer. Unsurprisingly, all the top results where form-level recommendations. Like “develop good writing skills”, “be humble and open to criticism”, or “construct a good hypothesis and find a ‘definition of done’ for testing this hypothesis”. You may find all of these things in a good researcher, but if you’re aiming for that, you are likely to become a one-person cargo cult.

So what should you aim for to become a better researcher? I grouped my thoughts on this into the four themes below. Keep in mind, though, that this list is just a starting point, as I am also still learning.

Robustness

Dense Knowledge. If a piece of knowledge was deleted from your mind, you could recover it by inference from other pieces of your knowledge.

Noticing Confusion. You notice your own confusion when pieces of your knowledge don’t match up. When this happens, you don’t rationalise your confusion away, but you drill down to its cause and resolve it.

Reducibility. You understand the relations between different levels of abstraction (e.g., optimisation processes, single neural network layer, transformer, etc.). You can also distinguish good abstractions from bad ones.

Perspective. You assume multiple perspectives on anything. For example, you can think of a self-organising map as a competitive learning strategy, or as a way to embed a graph in a discrete space, or as a generalisation of principal component analysis.

(Un)Certainty. You are aware of your own level of certainty about any statement. You strive to be as certain as possible about the things that matter, without being too certain. Remember that in 2 out of 10 cases (on average) where you were 80% certain, you should have gotten it wrong. No more, no less. You also learn about your cognitive biases and work to circumvent or compensate for them.

Sense of Direction

Orientation. Given a problem, you can work towards a solution. The following two points are aspects of this when the solutions don’t occur to you immediately.

Simplify Problems. You are able to simplify problems until you can solve them.

Simplify Solutions. You are able to simplify, combine, or repurpose solutions. This is how you enable yourself to solve the next-harder problem that you couldn’t solve originally.

Intuition. You develop an intuition to distinguish good directions from bad ones.

Execution

Testing. You find ways to test whether or not your ideas work. You know the setup well enough to recognise the causes for any test outcome.

Lightness. You can change direction quickly. You do this, when you either encounter enough evidence that the current direction doesn’t work, or when you’ve understood something new that suggests such a change.

Freedom. You adopt solutions and ideas of other areas to your needs. For example, you may have learned how to define a Lagrangian and use the principle of stationary action to find the equations of motion of a pendulum. Once you discover Fermat’s principle, you can use the same tools to calculate the path of a light ray.

Learn. You learn from your mistakes. That is, if you went off in a bad direction, you absorb without remorse the new knowledge that this is a bad direction. 

Meta-Learn. You meta-learn from your mistakes. For example, if you went off in a bad direction, you think hard if you could have recognized this as a bad direction before you went down this path.

Purpose. You execute with focus on purpose, not form. Researching doesn’t mean that you just follow certain “science rules”. It means that you actually and actively improve your understanding of nature. And nature doesn’t care if you followed all the rules and formalities of science.

Collaboration

Meaningful Writing. You can communicate your ideas clearly and generally write meaningful text. Scientific literature is not a literary genre, but a means to an end. You communicate in a way such that other people understand what you mean, not just yourself. This implies that you have a sense for what the other knows and doesn’t know, and what (s)he might misunderstand. You deliberately work to avoid these misunderstandings.

Maintenance. The Meaningful Writing point extends to writing program code. When you work on a larger project, it is usually good to iterate quickly and just try some things out with a “hack”, first. But you should frequently clean your code and make it as modular and as readable as possible. Otherwise you end up with an unmaintainable Spaghetti Tower that will make it very hard for you to continue your research.

Critique. When you receive critique, see it as an opportunity to get better. When you give critique, be honest and don’t water it down. But make sure to help the recipient deal with it.