I was gifted a puzzle for my birthday and while assembling it I noticed some similarities with the work of a researcher. Hear me out.

I have not stopped thinking at the “PhD metagame” since I wrote that piece, especially how being involved with too many disconnected topics made me a sort of “jack of all some trades, master of none” kind of researcher, and how this is not really appreciated in the AI research environment (Rosanne Liu also touched on this topic in a talk of hers).

The essential problem that I noticed is that it really takes a long time and effort before one can start being productive and generating research-grade ideas in a given field. This is why the first year or so of a PhD, for most students, mainly consists of orientation and working on other people’s ideas. After reading many papers, listening to many talks, discussing with many people, trying many ideas, one’s brain is sufficiently immersed in the topic that it can start generating genuinely new and worthy thoughts.

Consider now the process of assembling a puzzle. You start with more than a thousand small pieces, and you are supposed to figure out how they all fit together. The first thing to do is to sort the pieces by similarity, since pieces that look similar are likely to end up nearby in the finished puzzle. During this process it is also a good idea to discard low entropy pieces, those that are of the same color without anything drawn on it. Isolating the pieces that lie on the border of the puzzle is also a useful thing to do early on.

Creating these smaller clusters of pieces makes it a lot easier to compare them among each other and find matches. A particularly distinctive line or pattern hints at what pieces belong together, and soon little islands start appearing, their border gradually growing as you add more pieces and eventually merging with other islands. Slowly you realize how everything fits together to generate the final picture of the puzzle.

Sometimes a piece is missing, and you start scouting for it in the other piles. Sometimes your eyes fall on a piece and you instinctively know where it belongs, because you examined that particular area so many times already. Sometimes a piece gets lost, perhaps under the sofa, or perhaps it never was in the box in the first place. Although you know exactly what is supposed to be there, the puzzle will forever remain with an open gap.

Image of a puzzle with a missing piece

How does this resemble research? Think of pieces as research papers. Initially you are drowned by thousands of articles. You slowly start reading them and categorizing them, perhaps they talk about the same problem, or use the same method. Many articles are just dull, and you leave them aside, while others are truly revolutionary and you make sure to always keep them handy. Your library slowly grows, and small topical islands start appearing. Every paper you add fits with other papers and expands the boundaries of a certain topic, perhaps connecting it with another you thought unrelated. The big picture becomes clearer and clearer, and after a while you feel comfortable like at home, and whenever you come across a paper you instantly know which gap it fills and how it relates to other papers you already know.

Sometimes a piece is missing, and you cannot find it no matter where you look. What should be there is obvious to you, all the nearby pieces make it clear. So you decide to create that piece on your own, carefully crafting your research to perfectly fit and integrate with the related work, closing the gap you identified after much searching.

We are all working on this big puzzle of human knowledge. Who knows what is shown on the part that is still missing, and if we will ever be done.