Academic publishing as (ruinous) competition: Is there a way out?

Aaron Johnson invited me to speak as part of a panel on academic publishing at PETS 2013. This is a rough transcript of my talk, written from memory.

Aaron mentioned he was looking for one more speaker for this panel, so that we could hear the view of someone naive and inexperienced, and asked if I was available. I said, “Great, I do that every day!” So that will be the tone of my comments today. I don’t have any concrete proposals that can be implemented next year or in two years. Instead these are blue-sky thoughts on how things could work someday and hopeful suggestions for moving in that direction. [1]

I just finished my first year as a faculty member at Princeton. It’s still a bit surreal. I wasn’t expecting to have an academic career. In fact, back in grad school, especially the latter half, whenever someone asked me what I wanted to do after I graduated, my answer always was, “I don’t know for sure yet, but there’s one career I’m sure I don’t want — academia.”

I won’t go into the story of why that was and how it changed. But it led to some unusual behavior. I ranted a lot about academia on Twitter, as Aaron already mentioned when he introduced me. Also, many times I “published” stuff by putting up a blog post. For instance I had a series of posts on the ability of a malicious website to deanonymize visitors (1, 2, 3, 4, 5, 6). People encouraged me to turn it into a paper, and I could have done that without much extra effort. But I refused, because my primary goal was to quickly disseminate the information, and I felt my blog posts had accomplished that adequately. True, I wouldn’t get academic karma, but why would I care? I wasn’t going to be an academic!

When I eventually decided I wanted to apply for academic positions, I talked to a professor whose opinion I greatly respected. He expressed skepticism that I’d get any interviews, given that I’d been blogging instead of writing papers. I remember thinking, “oh shit, I’ve screwed up my career, haven’t I?” So I feel extremely lucky that my job search turned out successfully.

At this point a sane person would have decided to quit while they were ahead, and start playing the academic game. But I guess sanity has never really been one of my strong points. So in the last year I’ve been thinking a lot about what the process of research collaboration and publishing would look like if we somehow magically didn’t have to worry at all about furthering our individual reputations.

Polymath

Something that’s very close to my ideal model of collaboration is the Polymath project. I was fascinated when I heard about it a few years ago. It was started by mathematician Tim Gowers in a blog post titled “Is massively collaborative mathematics possible?” [2] He and Terry Tao are the leaders of the project. They’re among the world’s top mathematicians. There have been several of these collaborations so far and they’ve been quite successful, solving previously open math problems. So I’ve been telling computer scientists about these efforts and asking if our community could produce something like this. [3]

To me there are three salient aspects of Polymath. The first is that the collaboration happens online, in blog posts and comments, rather than phone or physical meetings. When I tell people this they are usually enthusiastic and willing to try something like that. The second aspect is that it is open, in that there is no vetting of participants. Now people are a bit unsure, and say, “hmm, what’s the third?” Well, the third aspect is that there’s no keeping score of who contributed what. To which they react, “whoa, whoa, wait, what??!!”

I’m sure we can all see the problem here. Gowers and Tao are famous and don’t have to worry about furthering their careers. The other participants who contribute ideas seem to do it partly altruistically and partly because of the novelty of it. But it’s hard to imagine this process being feasible on a bigger scale.

Misaligned incentives

Let’s take a step back and ask why there’s this gap between doing good research and getting credit for it. In almost every industry, every human endeavor, we’ve tried to set things up so that the incentives for individuals and the broader societal goals of the activity align with each other. But sometimes individual incentives get misaligned with the societal goals, and that leads to problems.

Let’s look at a few examples. Individual traders play the stock market with the hope of getting rich. But at the same time, it helps companies hedge against risk and improves overall financial stability. At least that’s the theory. We’ve seen it go wrong. Similarly, copyright is supposed to align the desire of creators to make money with the goal of the maximum number of people enjoying the maximum number of creative works. That’s gotten out of whack because of digital technology.

My claim is that we’re seeing the same problem in academic research. There’s a metaphor that explains what’s going on in research really well, and to me it is the root of all of the ills that I want to talk about. And that metaphor is publishing as competition. What do I mean by that? Well, peer review is a contest. Succeeding at this contest is the immediate incentive that we as researchers have. And we hope that this will somehow lead to science that benefits humanity.

To be clear, I’m far from the first one to make this observation. Let me quote someone who’s much better qualified to talk about this. Oded Goldreich, I’m sure most of you know of him, has a paper titled “On Struggle and Competition in Scientific Fields.” Here’s my favorite quote from the paper. He’s talking about the flagship theory conferences.

Eventually, FOCSTOC may become a pure competition, defined as a competition having no aim but its own existence (i.e., the existence of a competition). That is, pure competitions serve no scientific purpose. Did FOCSTOC reach this point or is close to it? Let me leave this question open, and note that my impression is that things are definitely evolving towards this direction. In any case, I think we should all be worried about the potential of such an evolution.

I’m don’t know enough about the theory community to have an opinion on how big a problem this is. Still, I’m sure we can agree with the sentiment of the last sentence.

But here’s the very next paragraph. I think it gives us hope.

Other TOC conferences seem to suffer less from the aforementioned phenomena. This is mainly because they “count” less as evidence of importance (i.e., publications in them are either not counted by other competitions or their effect on these competitions is less significant). Thus, the vicious cycle described above is less powerful, and consequently these conferences may still serve the intended scientific purposes.

We see the same thing in the security and privacy community. Something I’ve seen commonly is a situation where you have a neat result, but nothing earth-shattering, and it’s not good enough as it is for a top tier venue. So what do you do? You pad it with bullshit and submit it, and it gets in. Another trend that this encourages is deliberately making a bad or inaccurate model so that you can solve a harder problem. But PETS publications and participants seem to suffer less from these effects. That’s why I’m happy to be discussing this issue with this group of people.

Paper as final output

It seems like we’re at an impasse. We can agree that publishing-as-competition has all these problems, but hiring committees and tenure committees need competitions to identify good research and good researchers. But I claim that publishing as competition fails even at the supposed goal of identifying useful research.

The reason for that is simple. Publishing as competition encourages or even forces viewing the paper as the final output. But it’s not! The hard work begins, not ends when the paper is published. This is unlike the math and theory communities, where the paper is in fact the final output. If publishing-as-competition is so bad for theory, it’s much worse for us.

In security and privacy research, the paper is the starting point. Our goal is not to prove theorems but to more directly impact the world in some way.  By creating privacy technologies, for example. For research to have impact, authors have to do a variety of things after publication depending on the nature of the research. Build technology and get people to adopt it. Explain the work to policymakers or to other researchers who are building upon it. Or even just evangelize your ideas. Some people claim that ideas should stand on their own merit and compete with other ideas on a level playing field. I find this quite silly. I lean toward the view expressed in this famous quote you’ve probably heard: “if your ideas are any good you’ll have to shove them down people’s throats.”

The upshot of this is that impact is heavily shortchanged in the publication-as-competition model. This is partly because of what I’ve talked about, we have no incentive to do any more work after getting the paper published. But an equally important reason is that the community can’t judge the impact of research at the point of publication. Deciding who “wins the prizes” at the point of publication, before the ideas have a chance to prove themselves, has disastrous consequences.

So I hope I’ve convinced you that publication-as-competition is at the root of many of our problems. Let me give one more example. Many of us like the publish-then-filter model, where reviews are done in the open on publicly posted papers with anyone being able to comment. One major roadblock to moving to this model is that it screws up the competition aspect. The worry is that papers that receive a lot of popular attention will be reviewed favorably, and so forth. We want papers to be reviewed on a level playing field. But if the worth of a paper can’t be judged at publication time, that means all this fairness is toward an outcome that is meaningless anyway. Do we still want to keep this model at all costs?

A way forward?

So far I’ve done a lot of complaining. Let me offer some suggestions now. I want to give two sets of suggestions that are complementary. The first is targeted at committees, whether tenure committees, hiring committees, award communities, or even program committees to an extent, and to the community in general. The second is targeted at authors.

Here’s my suggestion for committees and the community: we can and should develop ways to incentivize and measure real impact. Let me give you a four examples. I have more that I’d be happy to discuss later. First, retrospective awards. That is, “best paper from this conference 10 years ago” or some such. I’ve been hearing more about these of late, and I think that’s good news. The idea is that impact is easier to evaluate 10 years after publication.

Second, overlay journals. These are online journals that are a way of “blessing” papers that have already been published or made public. There is a lag between initial publication and inclusion in the overlay journal, and that’s a good thing. Recently the math community has come up with a technical infrastructure for running overlay journals. I’m very excited about this. [4]

There are two more that are related. These are specific to our research field. For papers that are about a new tool, I think we should look at adoption numbers as an important component of the review process. Finally, such papers should also have an “incentives” section or subsection. Because all too often we write papers that we imagine unspecified parties will implement and deploy, but it turns out there isn’t the slightest economic incentive for any company or organization to do so.

I think we should also find ways to measure contributions through blog posts and sharing data and code in publications. This seems more tricky. I’d be happy to hear suggestions on how to do it.

Next, this is what I want to say to authors: the supposed lack of incentives for nontraditional ways of publishing is greatly exaggerated. I say this from my personal experience. I said earlier that I was very lucky that my job search turned out well. That’s true, but it wasn’t all luck. I found out to my surprise that my increased visibility through blogging and especially the policy work that came out of it made a huge difference to my prospects. If I’d had three times as many publications and no blog, I probably would have had about the same chances. I’m sure some departments didn’t like my style, but there are definitely others that truly value it.

My Bitcoin experiment

I have one other personal experience to share with you. This is an experiment I’ve been doing over the last month or so. I’d been thinking about the possibility of designing a prediction market on top of Bitcoin that doesn’t have a central point of control. Some of you may know the sad story of Intrade. So I tweeted my interest in this problem, and asked if others had put thought into it. Several people responded. I started an email thread for this group, and we went to work.

12,000 words and several conference calls later, we’re very happy with where we are, and we’ve started writing a paper presenting our design. What’s even better is who the participants are — Jeremy Clark at Carleton, Joe Bonneau who did his Ph.D. with Ross Anderson and is currently at Google, and Andrew Miller at UMD who is Jon Katz’s Ph.D. student. All these people are better qualified to write this paper than I am. By being proactive and reaching out online, I was able to assemble and work with this amazing team. [5]

But this experiment didn’t go all the way. While I used Twitter to find the participants and was open to accepting anyone, the actual collaboration is being done through traditional channels. My original intent was to do it in public, but I realized quite early on that we had something publication-worthy and became risk-averse.

I plan to do another experiment, this time with the explicit goal of doing it in public. This is again a Bitcoin-related paper that I want to write. Oddly enough, there is no proper tutorial of Bitcoin, nor is there a survey of the current state of research. I think combining these would make a great paper. The nature of the project makes it ideal to do online. I haven’t figured out the details yet, but I’m going to launch it on my blog and see how it goes. You’re all welcome to join me in this experiment. [6]

So that’s basically what I wanted to share with you today. I think the current model of publication as competition has gone too far, and the consequences are starting to get ruinous. It’s time we put a stop to it. I believe that committees on one hand, and authors on the other both have the incentive to start changing things unilaterally. But if the two are combined, the results can be especially powerful. In fact, I hope that it can lead to a virtuous cycle. Thank you.

[1] Aaron didn’t actually say that, of course. You probably got that. But who knows if nuances come across in transcripts.

[2] At this point I polled the room to see who’d heard of Polymath before. Only three hands went up (!)

[3] There is one example that’s closer to computer science that I’m aware of: this book on homotopy type theory written in a similar spirit as the Polymath project.

[4] During my talk I incorrectly cited the URL for this infrastructure as selectedpapers.net. That is a somewhat related but different project. It is actually the Episciences project.

[5] Since the talk, we’ve had another excellent addition to the team: Josh Kroll at Princeton, who recently published a neat paper on the economics of Bitcoin mining with Ian Davey and Ed Felten.

[6] Something that I meant to mention at the end but ran out of time for is Michael Neilsen’s excellent book Reinventing Discovery: The New Era of Networked Science. If you find the topic of this post at all interesting, you should absolutely read this book.

To stay on top of future posts, subscribe to the RSS feed or follow me on Twitter or Google+.