Difficulty Balancing Objective Usefulness and Subjective Enjoyment

This is written prior to the start of my PhD, at age of 23, which means I am not completely washed. I have been doing some form of research for around 4-5 years now, though nothing spectacular, I have some thoughts of which I think are quite unique on research.

In general, what’s required for human psyche to undertake a year-long commitment such as PhD, is some sort of positive feedback. It doesn’t have to be immediate, nor does it have to be long-lasting, but it has to be there. I think it is a well established wisdom that research in general shouldn’t be something that one forces upon himself in order to reach something else. One has to enjoy the substance itself.

The real question is, how do you enjoy research? More importantly, what the heck is research anyway? The reality is, not a single person wants to grind 100 papers to be able to have some chances at new ideas, everyone wants to say some thing interesting on the first day.

One might argue that publishable results are the positive feedbacks required for enjoyment at research. They might be necessary, but for some it is not required, and certainly they are not enough. First day at office, you are absolutely absorbed by your research direction, everything looks sophisticated; but if life teaches anything that’s the best outcome never happens, soon after you realize it’s a bunch of dumb ideas, after getting to the bottom of it, and you go through 7 stages of grief to come to acceptance with the fact that you have to continue what you started.

I am not saying it will happen, but certainly it’s a possibility. I will argue that for research to be truly rewarding, one has to find one’s endeavors either convey objective usefulness or subjective interest, the absence of both will be a dire situation indeed.

There are two ways to define usefulness, whether they are commercializable or whether they attract interests from other people. The first of which never happens for 99.9% percent of PhD works, and that’s especially true for communication areas where protocols are very standardized, but they still have more chances at being useful than 99% of liberal arts research. Second of which is a bit more complicated: many highly cited works, or areas that attracted huge research interests, never made it to the standard or actual products, because it’s entirely possible and in fact highly likely, that even the interests from peers are misplaced, and this is partly due to a bad culture in which people prioritize publications of any form over quality controls, causing the amount of pure junks in literatures impeding the progress to some extents. Although, a positive way to look at it is that via citations and connections, one’s work is to some extent meaningful, in its indirect contributions to the few works that actually materialize into reality. Whether this is copium is up to debate but I think it is not without merit.

Here comes an important question: Why not just do hardware researches, sure 95% of them still go nowhere, at least they are trying to be useful.

This is the part where the tradeoff kicks in and ruins everything. Arguably it is even by design. For a research to work on actual devices, there is almost always a strong proof of concept that predates the research itself. This is not a dig at ML, but if you read most of ML papers that do applications of certain networks onto new problems, even though they are non-trivial, there’s some degree of confidence that it’s going to work, maybe the robustness of results are not predictable, but it’s going to work in some sense, because the proof of concept is strong and with little math involved, most of the difficulty comes in the implementation.

For me it completely takes the mystery aspect of the research out of the equation. On the other hand, for theoretical works, you never know what’s the answer until you prove it, which is a major difference.

Surprisingly, in almost all engineering fields, there exist three types of researches, which I call: Hardware, Simulation and Theory. In communication, this is easily identifiable: 1. Hardware research that deals with networks and realization of protocols on testbeds. 2. Communication theory research that applies signal processing tools to communication algorithms. 3. Information theory research that generalizes models to provide insights on the fundamentals of communication.

They exist on a spectrum of usefulness and enjoyment. And I am lucky enough to have done decent amount of researches in all three branches. In fact, as I am writing down this, past midnight, before retiring to my fireside, I am reminded once again to not lose sight of other aspects of the telecommunication problem.