Announcing a Journal for AI-Generated Papers (jaigp.org) was a bold move that was bound to generate a wide range of opinions.
Some loved the idea. Others thought it was a joke. But as the one who decided to build the journal, I think it makes sense for me to explain why I did it.
First, let’s start with what is maybe the most obvious reason. The use of AI in research is growing, but as a “dark” activity. Many reports have shown that scholars are using AI more often and hallucinated references in legit outlets are becoming easier to find.
But this use of AI is not happening in the open. It is a “dark” activity in which AI-generated content is mixed with human generated content. Some might argue that this is ok, while others might see this as a form of “pollution.” In either case, we have a lack of transparency that we can consider problematic.
Personally, I like to work on purely human generated content (like this post) and to create using AI (like in this AI-generated paper). But I don’t think it is ok to try to pass one off as the other. My view is that the use of AI is not in and of itself wrong, but what is problematic is the lack of transparency around it. Creating JAIGP is drawing a line that says: this thing has gotten to the point that it will soon need dedicated venues, so let’s explore what those venues might look like.
But this first point is a bit shallow, since it answers the question why anyone would make a journal of AI-generated papers, not why someone would do so. These are very different questions. Someone in particular runs a risk that people “in general” do not. Doing something audacious means being willing to become the metaphorical punching bag for those who are looking for a villain. It means risking the ridicule of peers who might actually enjoy ridiculing others. That’s uncomfortable, but it is part of the price that first movers must pay.
Academia is not really a “community,” but a shared identity involving millions of people who were attracted to work on “knowledge production” for different reasons. Some people are attracted by the thrill of discovery, and want to be the first ones to observe, understand, or experience a phenomenon. Others see academia more as a “knowledge certification system,” designed to provide rigorous answers to well-established questions.
Science is a tug-of-war between the two, and there is of course a little bit of both in every scholar.
Personally, I enjoy the thrill of discovery. The things I am the proudest of are projects that were 10 to 20 years early and that slowly grew into their own fields: networks in economic development, AI estimates of urban perception, quantitative studies of collective memory, and AI-augmented democracy. Each of these received pushback at first, but later became fields, concepts, or part of a larger conversation. So when I saw where things were going, I understood I had a short window of opportunity to create the first journal for AI-generated papers in the world. My intuition is that this is a chance to power through the ridicule again, since AI-generated papers are likely to be common in 20 years.
And this brings us to the deeper epistemological reasons that motivated me to embark on this project.
That is the idea that changes in technology can create rare and punctuated openings to rethink institutions. In this case, one of the things that fascinates me the most about the possibility of AI-generated papers is not just that AI can do research, but that it is inviting us to rethink research institutions. That makes JAIGP not a journal, but a space to explore the rules defining how academic journals should operate through crowdsourced participation.
The first versions of JAIGP had a set of rules that I chose. And I needed that to move fast. I decided, for instance, to use ORCID for authentication and introduce an endorsement and AI review system. But each of these rules has many knobs that we can adjust. How many endorsements does a paper need to move on to the next stage? Who can endorse? How many failed attempts lead to a ban? Etc. So now that the journal is out, I am turning to this larger goal and have made each a participatory exercise. The point is to let the people in the community decide how the rules of the journal should be. Every month journal members can vote, so that during the next month the journal operates with the rules that they chose.
That institutional twist is not directly related to AI (any journal could potentially decide its rules through an iterative crowdsourcing exercise). But AI provides us with a window of opportunity to explore these ideas.
Honestly, I don’t care if JAIGP becomes a lasting venue or a footnote. The best-case scenario is that it becomes a good place to submit AI-generated papers. The worst-case scenario is that it ends up on my shelf of old projects, as better versions of the idea take over the space. In both cases JAIGP helped establish AI-generated papers as a category deserving of its own rules and institutions. That's a win in the discoverer’s book. The point was to explore that future in the open, while many are still pretending the dark activity isn't happening.