EA needs a cause prioritization journal

by kbog12th Sep 201819 comments



It's about time that someone started publishing a journal for cause prioritization arguments. It may have open peer review or a traditional formal process, it may come from a reputable academic press or it may be an in-house thing with zero credibility to the outside world, it may publish regular papers or weird things like argument maps and adversarial collaborations, I make no claims about how to navigate these details, but there needs to be a journal.

I don't think this claim needs much special argument; I expect that many people's reaction to this will be "gee whiz, of course that's a good idea." We think cause prioritization is a serious new issue, we should have a venue to publish about it, pretty straightforward right? Every new field gets a journal, and the review and publication process improves article quality. Still, I will develop the argument for an EA journal in greater detail, and hopefully make it clear that it is worth the time and money required to run it.

EA has developed a significant body of knowledge spread across a variety of spaces. Some of it is fine as common knowledge and blog posts; simple ideas, which are easy to accept and communicate. Or they may be temporary practical issues, like evaluation of a particular charity, that don't last long enough to warrant the special attention and timelessness of being put into a publication. Yet cause prioritization arguments are more conceptual, they are important, and they need to be fleshed out with greater rigor than they have been. This paper for instance has been relied upon for ages, yet nothing has really followed on from it. There has been no quantitative modeling or sensitivity analysis, there has been no thorough comparison of risk reduction versus growth acceleration, and without that we aren't getting closer to really settling the issue of whether existential risk really can be presumed to be our biggest priority in the long run.

We occasionally have discussions that move past the basic old cause prioritization arguments, but that is transient; many arguments are left in a form that is impossible to summarize and build upon, then forgotten. Then we devote most of our time to bickering about the community, because we are not being fed enough difficult new intellectual content to keep our minds occupied. We think that all the cause prioritization arguments were pretty much fleshed out already, so we assume that the issue is settled and treat causes like ideologies or identities rather than hypotheses.

So not a lot of real progress is actually being made on the core EA project. I think the situation is actually very poor by academic standards. We need to encourage people to do serious work on cause prioritization, make sure that it is rigorous, and make sure that it is in a format that lends itself to further work. A good way to address this and other problems is to create a journal for it.

Existing journals aren't suitable.

First, they are rarely interested in the questions of cause prioritization. While journals may be interested in cost-effectiveness arguments for particular efforts in their particular field (e.g. https://www.academia.edu/34953571/Cost-effectiveness_of_interventions_for_alternate_food_in_the_United_States_to_address_agricultural_catastrophes), I can think of none which would care about rigorous comparisons between very different causes. For not only is cause prioritization interdisciplinary among different scientific fields, but it also relates to ethics and decision theory, meaning that it crosses a set of domains so wide that no typical journal will encompass it. Utilitas can be amenable, but it won't work for non-utilitarian EA arguments, or arguments over moral uncertainty.

Second, publishing to an external audience requires explaining or defending lots of basic premises and background that may be common in an EA audience. The results can be seen in these two papers (https://commons.pacificu.edu/cgi/viewcontent.cgi?article=1573&context=eip / https://link.springer.com/article/10.1007%2Fs10677-013-9433-4). They both argue for the viability of certain EA activities, yet the most of their space is taken up by issues which were widely considered settled or uninteresting in the EA community at the time of publication.

Third, common scientific norms revolve around frequentist statistics and typically require proven results in order to be considered interesting; everything else is relegated to the same bucket of "plausible". This makes it difficult to make a scientific argument for the prioritization of some uncertain causes over others. In EA, we are much more comfortable with subjective Bayesian epistemology, we are more comfortable acting on the basis of uncertain possibilities, and we are very interested in knowing just how likely or unlikely a possibility is. With our own journal, it will be easier for us to publish a paper that properly models all the uncertain parts of the equation that are necessary to complete a full cause prioritization argument. Subjective Bayesian methodology does get published in reputable journals, it's just rare - most academics are less comfortable with it.

Fourth, involving EAs in the peer review will ensure that papers are appropriately written to target EA criteria and dodge the likely EA counterarguments. Even though the author of a draft would be an EA, she may have a blind spot regarding the way that other EAs think about the issue.

It will lend notability to our movement and ideas.

Having a journal on a subject is a small step for establishing it as a widely-acknowledged legitimate field of study. It encourages other people to cite the works in their own literature and take them seriuosly. To be sure, a new journal may not be noticed at all, and I don't know what it takes to make a journal seem properly reputable as opposed to something like those creationist journals where they are just publishing whatever fits their point of view so nobody pays attention to them. But anything would be technically better than having these ideas be put up on the internet by their own authors with no review or publication at all. To be sure, in practice, a very bad journal would be worse than nothing because of the reputation risk, but I doubt that we would screw things up that much.

It will create opportunities for outsiders to contribute.

Again assuming that it is sufficiently reputable, it could encourage people outside EA to weigh in on our questions, because now it means that they will get a publication. A major benefit here is that the review process, being led by EAs, will help them translate their ideas into a form that is more suitable for us. As it stands, people on the outside who try to attack EA cause priorities often have a pretty poor grasp of what does or doesn't work as an argument in the EA framework, and a poor grasp of our likely counterarguments. This problem persists in academic publications that comment on EA causes. It would obviously be much better to hash out these things pre-publication, so that they can provide us with more useful arguments.

We have enough output to sustain a journal.

Think about all the things being written by individuals in the community on forums and their personal websites, the blogs and reports from all our organizations (Givewell, Open Phil, ACE, FLI...), and the papers that are being published in academic journals by EAs. Much of it is very seriously thought out. With all this effort, just a bit of emphasis of cause prioritization will enable journal issues to be put together within an acceptable length of time.

Hey, what about academic incentives!? Peer review is broken! It's all status signaling! Think about publication bias! Think about the replication crisis! Remember the Sokal Affair??? Etc, etc.

However great these problems are, surely the practice of spreading arguments across Facebook walls, podcasts, and blog posts has them or others to a greater degree. And that's the state of EA thought on cause prioritization. In any case, EAs know enough about these issues, and are sufficiently free of the need to pander to establishment institutions, that we can presumably design a journal mechanism that is better than the norm.