“Over the years I’ve occasionally gotten snide comments from people implying that the time I spend online is a sign of vanity or frivolousness unbecoming of an academic. Every one of those was from people who have the privilege of not having to be loud in order to be heard.” --- Arvind Narayanan
I decided to write this post because I was really inspired by Arvind’s comment and I do have something to say about this. In the last three years, I learned it is a priviledge to get your own work cited without going extra miles to publicize them. Sometimes, it’s because your work comes out from famous AI labs. Sometimes, it’s because famous researchers are the senior authors on your paper (so your paper immediately got the stamp of approval). Sometimes, it’s just luck.
In my research journey, I received advice like “do good work and your work will get recognized.” In Mandarin Chinese, there’s even a saying “有麝自然香，何必当风立”, which translates literally to “where there is musk there is fragrance; it doesn’t need to be wafted everywhere by the wind.” The essence is that good work doesn’t need to be promoted and will get picked up naturally.
However, in my experience, doing good work is only the bare minimum. I have seen often great-quality work received little attention. Even worse, sometimes certain labs in industry pretty much followed your research recipe and released the same findings several months late (without even citing your work), and they received all the attention and applause.1
The Visibility Game
Visibility matters a lot (if not the most) to researchers. We do good work and hope others see it, so they can immediately build upon our findings and explore the next interesting research question. Dare I say it, superficial metrics such as citation counts and h-index rely solely on your visibility to the community.
Something I noticed is that people proactively ask other PhD students who their advisor is as a way to judge the quality of their work before reading them. It’s something I actively avoid doing because I see each PhD student as individuals–––but unfortunately I have also committed certain practices such as name-dropping where my work is from and with whom I’ve worked in the project in hope of people will read and cite my work. We rely a lot on heuristics to filter papers given the inundation of preprints. To quote Arvind’s words,
“In academia, your name is your brand. In a game of reputation, that’s everything. Having your name be known in the community is how your get invited to talks, how people look you up to cite your papers, and how they decide whether your job application is worth a serious look.” --- Arvind Narayanan
That’s why there had been constant debate in *ACL community about preprint anonymity. Anonymity is a way to prevent visibility (and unmerited recognition) to bias reviewers’ judgement of the work. However, given the fast-moving AI research field, not releasing the work as preprints can actually hurt the visibility of non-priviledged researchers 2.
Where I Am Coming From
I think I am a bit more sensitive to the visibility game because I used to spend a lot of time on low-resource NLP research. I have observed two phenomena. First, call-to-action for the field for better research practice, such as “naming your language” aka Bender’s rule, is effective if you are visible and influential. Second, there are a lot of silos among different language research communities, where people are building the same NLP toolkits and researching the same linguistic phenomenon, but for different languages. I often wonder what if we know what one another are working on, so we can collaborate and reduce the redundancy?
After participating in various open-science research collaboration groups, I see the struggle of under-priviledged researchers. Highly-visible researchers often have easier access to research resources, from GPU hardware down to collaboration opportunities. Note that despite the low barrier of entry for participating in these initiatives, there is often still limited access to the type of research questions you can work on depending on your background (which influences how much say you have on the project direction) and the organization’s research priority (which unfortunately is out of control for many under-priviledged researchers).
Some Observations So Far
Playing the visibility game is still not something I am comfortable with, but here’s my biggest lesson: Being visible is important if you want to call for action and execute certain research agenda. Being visible allows you to create massive research impact. Being visible reduces friction for collaborative research. There’s virtually no downsides to being visible, except that the quality of your work (and your character as well as your efforts to be visible) will be judged by more people.
To me, being visible correlates with the ability to influence the field. I relate to Michael’s blog about “Scientific communication in the age of influencers”. Being visible goes hand in hand with building a following.
A paper rarely changes the world. Really new ideas often go unrecognized for years. A paper does not create a new field. To change a field or create a new one, you have to bring people along with you. That is, one has to become an influencer. --- Michael Black
However, what’s not talked about much is how you build the following. Why should people follow your line of work? How do you get your first reader who is willing to go through your prior work and follow your upcoming work? I personally think the easiest way to build the following is to step up, lead a collaboration on a problem you care about, and make it as easy as possible for the other people to collaborate. Your collaborators will cite your work (since their names are on the paper too), and you and your collaborators will become each other’s followers. (There are a lot of other benefits of wide collaboration, but I am focusing on the visibility aspect only).
Note
(12-27-2024) Now I wonder if there’s a way to be more visible with spreading myself too thin to collaborate widely.
Often I would think that I am being shallow playing the visibility game. But I constantly remind myself being visible is not about chasing clout. Being visible allows my work to reach the intended readers, and it gives me the opportunity to have meaningful conversations with other researchers in the same research area. I personally know that the Halo effect is real, and from my collaborative research experience, visibility begets more visibility because one is trusted as somebody who can lead a collaboration so people are more willing to collaborate with him/her.
Influential scientists have always known this. They have a constant conversation with their readers. There is give and take. They have always built community — sub-fields, workshops, conferences, journals, associations, societies. These are mechanisms to build community around ideas. Social media should be viewed the same way. --- Michael Black
I am still learning to be more visible, such as posting more and simply be expressive (from Nathan’s tips). Omar Khattab also shared his thoughts on how to create research impacts in AI. Nonetheless, I think one’s journey to become visible and create wider research impacts is highly personal, and there’s no one-size-fits-all approach.
Dec 27, 2024
I am still figuring out how to be visible.
Many of the people I look up to do not play the visibility game (and some of them advised me to tweet less), and yet their work is highly relevant and visible. At the beginning of this year, I thought I had to tweet a lot or had to be influencer to gain visibility, but lately I doubted my original approach.
- Blog posts don’t really matter, even if your one of your blog posts goes viral. People don’t associate one-off blog post/paper with you. You need to consistently blog to build branding (before you can use blogs to improve citations and job opportunities). The question is whether it’s worth it when the alternative is to dedicate time for reading cooler papers and trying many research side projects.
- The type of job that I want at this point doesn’t need me to build personal branding informally. Simply publishing good research, working closely with SOTA, and attending ICML/NeurIPS/ICLR are sufficient to get me closer to my ideal job, I think.
Footnotes
-
If you follow Twitter drama in the past 12 months, you will know exactly what papers (yes, plural) I refer to. It really sucks to see this happens because this is the tip of the iceberg. We only heard of this because the victims are relatively influential so their complaints went viral on Twitter–––who knows how many more people’s work went unnoticed (especially those from less-known academic labs)? ↩
-
Yoav put this better than me in his note Putting papers on arxiv early vs the protections of blind review. ↩