T-Shaped People and Academia

Tags: academia, musings

« Previous post: Things You Are Allowed To Do, Academic … — Next post: My Experience with Lenovo Premier … »

Being on the lookout for recruiting new students for my new research group, I wanted to distil some of my observations into textual form. This post deals with my impression that academia is doing a disservice to ‘T-shaped persons,’ i.e. persons combining deep expertise in one or more fields with an ability to collaborate across areas. T-shaped persons are often contrasted with I-shaped persons, commonly known as experts.¹ Given the state of the art in the software industry,² T-shaped people are a sought-for commodity: the current ‘framework du jour’ might be obsolete in a few years, but the horizontal bar of a T-shaped person ensures that they can also contribute to other aspects of a project, and quite likely will develop new vertical bars, i.e. new expertise over time.³

Output above all else. As I am looking at the many resumes and motivational letters from excellent students all over the globe, it occurred to me that academia is not necessarily kind to the Ts: the primary metric in academic hiring is still the ‘scientific output,’ which is measured in publications such as journal articles, conference articles, etc. In machine learning research, the fast pace of the field now has created a situation in which even aspiring Ph.D. candidates are often expected to have a few publications under the belt already. Coming from mathematics, where one paper every few years (!) is considered the mark of a productive mathematician, these expectations are bonkers to me, and I wonder how many capable candidates end up pursuing other goals because of these invisible barriers.⁴

Hiring people: I before T. In any case, many graduate schools and resume reviewers tend to ignore the breadth of a person’s skills, focusing instead only on the depth. By taking the scientific output, in conjunction with maybe some relevant coursework, as the be-all and end-all, any other skills tend to be missed. At best, it might raise an eyebrow or two in magnanimous surprise: falling prey to precisely the issue I describe in this post, I noticed myself doing this for candidates that already convinced me because of their academic track record! At worst, additional skills might be considered detrimental upon review, because a T-shaped person may have a smaller number of papers than others.⁵

This is a problem for machine learning research. Given the aforementioned fast pace in the field, excellent research necessitates moving away from silos! There will of course always be a demand for people that have deep domain expertise, but many facets of contemporary research can be addressed just as well by a T-shaped person, in my experience.

Ts as multipliers. Throughout my career, the most impactful projects always had a T-shaped person onboard. This person would usually not be an expert in the subject matter, but would be able to provide the direly-needed scaffolding and foundation of a project that is all too often ignored in the initial phase, until it comes back later on with full swing to wreak havoc. This includes, for instance:

• Setting up a suitable build system.
• Writing some unit tests for the code.
• Creating a skeleton of the models that are to be developed.
• Designing nice mock-ups or logos.
• Creating animations or graphics for presentations.

In my experience, these skills—and others—are not commonly considered when hiring someone. And to some extent, having fallen into the same trap, I fully understand this! Paraphrasing Leonard McCoy in Star Trek, we want to state ‘I’m a researcher, not a(n) $X$,’ with $X \in$ $\{$software developer, designer, marketing person, artist, $\dots\}$

Progress in our interdisciplinary times, however, also requires people that are sufficiently skilled to tackle these other tasks. Even though such tasks might seemingly be unrelated to the overall scientific goal of creating new knowledge, they are still a fundamental part of the modern process of scientific inquiry.

What to do now? I am writing this post because I detected an inconsistency in my personal observations and my actions: virtually all the projects that contained a good mix of I-shaped and T-shaped persons succeeded extraordinarily well. Yet, upon recruiting my own team, I started to prefer I-shaped profiles over T-shaped ones, instead of striving for a solid balance. I caught myself in time and will subsequently pay more attention to this bias of mine—I hope that others will do the same!

Machine learning and academia in general are enriched by teams comprising diverse backgrounds and personalities. We should strive to achieve this.

(In case more incentives are needed: the largest groups and research divisions are already doing this—they are hiring not only for academic output, but also based on general software skills, for instance. If it’s good enough for them, it should certainly be good enough for the rest of us.)

Happy recruiting for 2022 and the years to come, until next time!

Update (2022-01-09): This post spawned a lively discussion on HackerNews. One interesting facet that escaped me so far is the extent to which certain research groups are exploiting Ph.D. students for tasks that should rightfully be handled by a person specifically trained for the job. That’s something I can wholeheartedly advocate for—I think most labs benefit immensely from an on-call sysadmin, developer, etc. The point I am making in this article is that, at least in ML research, it is hard to pin down what ‘doing only research’ means. If we only measure the value of a budding Ph.D. student by the number of papers on their CV, we miss out on candidates that can introduce better coding practices into a lab, for instance. Of course, they should not be hired primarily because of this, but why not recognise and reward skills that are clearly there and relevant for doing research and facilitating the research of others?

Update (2022-01-10): A very insightful comment from the aforementioned discussion stuck with me, so I shall preserve it here (emphasis mine, read the full comment here):

At the scales that most labs work, there are many essential yet mundane tasks that need to be shouldered by “core” participants. The overhead of trying to delegate these tasks to a different person would either require more time than doing the task oneself, or result in abysmal quality when the hand-off is made with insufficient supporting effort.

It is exactly these ‘other’ tasks where having access to a team with some generalist people and some expert people can make a large difference. It is time to recognise the people that are shouldering these tasks.