A short disclaimer is necessary before diving in: the below is a rather personal opinion on the subject — driven by my personal experiences in AI research. As such, it is not meant to blame, contradict or discredit anyone or anything. Instead it is an attempt to add color. I think the project track in question is rather specific; I am sure much thought has gone into it and NeurIPS will iterate on it in future instances of the conferences.

This being said, I think many arguments raised on X are not necessarily about NeurIPS' decision to have such a track in specific. Instead, many arguments can easily be extrapolated to be about how our research community has rapidly changed of the last few years. I feel many opinions and arguments try to raise how competitive AI research has become, how becoming (and remaining) more inclusive is extremely challenging, and how prerequisties (what many people might call privilege on X) is unevenly distributed across the world.

Generally, having these specialized tracks sounds (and sometimes is) more inclusive. After all, high schoolers are explicitly encouraged to participate. This is not taking anything away from anyone, but instead adding opportunities. On first thought, I would have loved this opportunity growing up — if (and only if) I would have known about it. But I would not have. Even if the program would have been around back then, chances are slim to zero I would have had the opportunity to participate. This is due to many factors: the nearest university being too far, my parents, extended familiy or general environment not being into CS or AI. Even if, traveling to the US at that age would have been extremely difficult. This means, I would not have benefitted from this. Which is fine — there are plenty of opportunities that I didn't benefit from but generally appreciate being widely available.

So I went on to do an undergrad and a masters (2 years in Germany) in CS without any NeurIPS paper. I was lucky to make some good decisions, being exposed to AI early through some internships and research work during my masters and so ended up writing a good master thesis and eventually submitting this paper to CVPR in the first few months of my PhD. At the time when I applied for PhD programs, I had no top-tier papers (not even workshop ones). Many of my PhD colleagues started out without any papers or internships. They got into good PhD programs purely based on their interest, good grades and potentially a single AI related project. This was around 2017. In the US, it was already common to have published before starting a PhD.

Since, I have interviewed many PhD applicants and students (for theses, internships, etc.) and as many on X realized, it got much more common to have one or multiple top-tier papers before starting a PhD. Towards the end of my PhD, the top programs in Europe got so competitive that you either needed personal connections or at least one first-author top-tier paper that was well-received (aka a few citations) and/or a few relevant internships.

This all being said, my main fear is that having such tracks, targeted at earlier and earlier stages in an academic career will lead to a more and more biased selection of students able to participate in the research community.

The thing is, what I excluded above, the contact to my PhD advisor came from a short paper I wrote that was accepted at GCPR (the German Conference on Pattern Recognition, BMVC but in Germany) in 2015. My advisor was in the corresponding committee/jury. This was part of the Young Researcher Forum at GCPR.

By luck, my bachelor thesis advisor knew about this program. And because my bachelor thesis was obviously (a 3 month project) not cut out to be a top-tier paper submission, he recommended it to the program. Because of this, I had some (arguable how much) competitive advantage during PhD applications (at least within Germany/Europe). Not only because my PhD advisor actually knew me only from this program, but also because I had this bit of experience publishing and going to a conference. Note the irony: my bachelor thesis served the sole purpose of finishing my undergrad degree. By knowing about this program, I just re-purposed it as a paper. In a way it was merit based — my thesis was good and I wrote a nice paper — but at the same time it strangely wasn't — the core research was the same, only knowing about the program changed things. There is nothing bad in "only" completing a thesis/project for the sake of learning (and to get a good grade). In this sense, such programs tend to reward connections/insider knowledge, but they may be perceived as something special during applications.

To draw the parallels to the new NeurIPS track: I was this high school student that was lucky enough to (a) know about the new NeurIPS program and (b) have the support to submit and be successful. As a result, everyone else that did not know about the program, was suddenly a bit less competitive because the program exists, is was advertised, and researchers knew about it — so they expected good students to make use of it, or at least favoured students that did. This is how such programs create expectations. In fact, many AI researchers I know from Germany, submitted to the young researcher forum at GCPR. So many of us come from an extremely selected group of people that knew about this program early enough. It means that students not exposed to the right "circles" can be genius in their coursework, use any opportunity they have access to, but still not be perceived as competitive. And I am sure everyone out there is trying to be fair and account for this during hiring — until you get 100s of applications, you cannot look through all of them yourself, and interview invitations rely on others (unaware of these nuances) selecting by keywords or similar heuristics.

I feel this argumentation follows many of the opinions voiced on X with people joking that we should have kindergarden tracks, too (see here or here). This does not mean it is a bad program. I just want to highlight some of the unexpected side effects such programs may have in creating expectations, especially in a global community.

Another concern that I personally have but many people are not voicing is that we do not need more papers at the top conferences. This touches on the point I made above: what is wrong about writing up a summer project or undergad thesis just for the sake of learning or a good grade and only your research lab/univerity reviewing it? But it also touches on the reality that I am not even able to keep up with papers in very, very small niches anymore. Journals were too slow, so we went ahead and published more at conferences (remember that in most scientific discplines, conferences just have abstracts and "real" publications happen at journals — ignoring the discussions about unreasonable journal fees). Conferences got too big, so we have published proceedings in workshops to discuss work in progress. Conferences got too slow, so most of us are only following ArXiv anyway. It was close to impossible to go through a whole poster session within the time at NeurIPS last year, so many researchers I talked to did not try and had coffee chats instead and went to the open source parties. Essentially, this resulted in separate circles and groups building around the big conferences. For senior researchers this is just a decision to make, but PhD students are often insecure whether investing time in a good poster is still worth it for their career. This also impacts reviewing. Many papers I am asked to review are already months old due to ArXiv. Reviewing load is increasing, leaving less time for proper reviews which in turn leads to "noisy" reviews. As a result, I see many good AI researchers quitting reviewing and other academic duties.

Of course, this is slightly exaggerated. I saw many good posters; poster sessions are still crowded; and I still published at a journal last year. But I feel this sentiment is shared by many I talk to. So why do we encourage more but smaller papers (not meaning to say that papers from the high school track would be of poorer quality or not worth being shared!) instead of encouraging less but "bigger" papers (larger teams, more thorough, more mature ideas, etc.)? This would also mean redistributing our time from reviewing more unfinished papers quickly to reviewing less papers more thoroughly.

Finally, I want to stress again that I am not necessarily against this particular high school track at NeurIPS. I think abandoning it is not a solution to the above concerns. On the other hand, many high school students still submitted to NeurIPS, as evidenced by many people on X (see here or here), despite this program not being around in the past. So, overall, it seems I haven't really made up my mind yet.