Managing Research Teams — Part II

On Being a ‘Servant Leader’ in Research

Part I can be found here.

In the last section, I discussed helping build the narrative around your team’s goals and mission.

6. Elicit milestones.

I generally prefer, in the exercise of building a narrative, to stop short of setting specific goals, but I find it equally worthwhile to nudge individual researchers in setting goals for themselves. Milestone setting uncovers dependencies that are not necessarily obvious. It helps elicit decision points and possible forks in the road which in turn helps people think through the right question to best inform that decision. In purely academic settings, the never-ending grind of conference deadlines tends to act as a forcing function for project milestones, but it’s a poor one because it creates time-based milestones instead of ‘learning-based’ milestones which are much more aligned with actual progress.

7. De-risk all the things.

I talked a lot about risk in a previous piece, and how because of the inherent risk of research, one should be ruthless about removing all other risks from the equation. This is particularly important as a research lead, because many of the choices that define the risk envelope for a team are in the hands of the leadership: who to collaborate with, where the funding comes from, and what infrastructure bets to place.

8. Provide cover for new ideas.

Most ideas are absolutely terrible when newly formed. They also run up against powerful foes: inertia, not-invented-here syndrome, those who will argue against it for the mere pleasure of the argument, or those who happen to miraculously have had the same idea last week or back in the 80’s. New ideas, including the bad ones, need a chance to brush up against reality, and much that stands in their way before they can even get there. It is important to give them that chance, because good ideas are often bad ones that have been chiseled into less terrible ones by contact with reality and have managed to survive.

9. Be ready to turn on a dime.

Every once in a while, a breakthrough happens that turns your world upside down. Blink and you miss it. You have to be ready, both organizationally and psychologically, to turn on a dime. I have lived through and initiated several ‘research pivots’ in my career, from doubling down on deep learning when we got the first whiff of positive results in speech recognition, to moving into robotics when computer vision stopped feeling difficult enough. I’ve also often seen once fringe research papers blossom into full fledged research agendas — Generative Adversarial Nets comes to mind as a prime example — that quickly warranted reconsidering how much effort to project in those directions. Being nimble is a powerful quality in research, where we sometimes see once-thriving organizations be undone by mere inertia in a matter of years.

10. Lay bare all biases.

A research environment can be fertile ground for many types of biases. The most prominent, of course, are those stemming from our collective unconscious biases, doubly fueled in the case of computer science by marked gender and racial imbalances in the field. Others are more domain specific: titles can lead to a multi-tiered system, where ‘engineers’ are considered staff in support of ‘research scientists’ at the top of the food chain. I am proud that all our ‘best paper awards’ last year had people who were not technically research scientists as first authors, and try to make it clear that I hold people to the same standards regardless of title.

Seniority is also conducive to lopsided power imbalances that can unduly affect junior team members. Fighting bias is hard, because even the mere perception of bias hurts: if you’re a junior person afraid to stand up for what you think is right because some Distinguished Senior Someone may object, you may unwittingly subject yourself to the very bias you want to fight by virtue of your own perception. It’s the duty of a leader to point out those potential hot spots for bias, and set the tone for a psychologically safe environment where everyone can feel empowered to speak up.

11. Fight the flag planters.

Ideas are cheap. Turning them into research is hard. There is often a temptation in research to write down an idea, or a plan for future direction in a project, and claim it as territory without actually executing on it or validating it. A variant of this is writing down a design document hoping that ‘someone else’ will pick it up, and you will indirectly get the credit if someone ever does. This is often called ‘flag planting,’ and can lead to all sorts of dysfunctions: people not wanting to touch a subject area because someone else has produced a document in the past describing similar research, and hence implicitly threatening to take the credit if anyone approaches it; or people jumping into research proposals that are circulated, sprinkling two words into them, and later arguing they deserve credit for the project. This can, in turn, make people think twice before sharing project plans and pollute an otherwise collaborative atmosphere.

One useful tool to fight these issues is to provide a neutral escalation path for credit assignment issues. I typically refuse to be listed as a co-author on publications that I haven’t contributed to in order to remain free to adjudicate on questions of credit assignment. I also call out when I think flag planting is happening, making it clear that unproven ideas get no credit unless they have active research associated with them and some degree of validation.

12. Identify and cultivate the green thumbs.

When a researcher comes to you with a negative result, you have to ask yourself whether it means ‘it doesn’t work’ or ‘they couldn’t make it work.’ I’ve come to realize over the years that some researchers are just fantastic experimentalists: they are the ‘green thumbs’ of research. If an idea is any good, they will make it work. Conversely, if they come back to you with a negative result, you know the idea is no good. One of my former colleagues, who shall remain nameless (he may have invented AlexNet), was a great example of this: given a well-defined problem, he can dive into it and come up with an authoritative answer that I can implicitly trust. Not all researchers have this facility with experimental work. It’s a skill that is remarkably orthogonal to other forms of academic smarts. Many will come back with a negative result only to be invalidated down the road when someone else picks up the same idea, adds the key twist, tunes the parameters in just the right way, and finally makes it go. Great experimentalists can be the backbone of a research project and often don’t get all the credit they are due. They need special care.

13. Celebrate in proportion to impact.

One of the most powerful tools of a leader is the power to celebrate. As a secret introvert (shh), I have not wielded this tool as often as I should, but every time I do I marvel again at its effectiveness. Depending on your personality, it can be very tempting to let every success stand by itself, and hope the rest of the world will do the cheering for you. Conversely, you may be tempted to turn every little bit of progress into a party. This ignores the huge directional leverage you can wield by being deliberate in what you choose to celebrate and what you don’t. In research, every published paper, every benchmark beaten, can be reason to feel good about one’s progress. But not every bit of progress is equal, and not every research direction is worth equal investment. In an environment where top-down decision making would be a sure way to stifle innovation, ‘directional cheerleading’ can be a much more powerful steering tool.

14. Make unknowns and failures part of normalcy.

‘If we knew what we were doing, it wouldn’t be called research.’ (not Einstein … apparently)

It is very difficult for anyone to treat the state of being in the unknown as normal, even for researchers. ‘I don’t know’ is one of the most difficult answer to hear or utter, which is why I insist on making sure I keep it part of my vocabulary.

Similarly, failures of processes, equipment or collaborations happen all the time, every single day, and yet we somehow insist on treating them as exceptions. Failure to treat things that don’t go as planned as anything but business-as-usual can be a great source of stress and dysfunction. I try to promote the culture of treating exceptions as merely an exciting opportunity to go to http://docs.new (pro tip) and start a postmortem.

‘Normalizing’ as much of what routinely happens in real life can really change the tone of a workplace from a constant mad scramble to a much more comfortable environment where one knows that the system has their back, even if they don’t know what’s going on, or when they want to take a risk.


There is, of course, a lot more to managing a research effort: if I had devoted lines on this page in proportion to time actually spent, a full one-third of this essay would be about recruiting and hiring. To many people, ‘strong’ management evokes the notion of a very active leadership style, charting a course for the team, setting concrete goals, and bringing individuals into alignment. One would understandably wonder how you are supposed to do that when every researcher has their own research agenda, goals that may or may not materialize, and the very fabric of the team is to explore the unknown.

Servant leadership, with a healthy dose of nudging, is my preferred answer to this conundrum. It can be a fantastically rewarding endeavor, a chance to meet, interact with, and serve exceptional individuals, and to immerse oneself in questions whose answers that may genuinely change the course of humanity.