Letting Our Research Run With AI Content

This is a case of not closing the barn door after the horse is out, to use a pre-twentieth-century expression for a twenty-first-century issue. But, more precisely, I want to argue for propping the barn door open to enable the rest of the horses to run free after a good number have been questionably sold off.

Let me explain. Think of those sold-off horses as the research studies that at least three major research publishers – Taylor & Francis, Wiley, and Oxford University Press with more deals pending – have rented out to AI giants, such as Microsoft, for the purposes of training Large Language Models (LLMs). It’s a publisher windfall, having already sold access to libraries or charged authors for providing open access. Researchers are understandably outraged over neither being asked nor compensated. It also drives up AI costs, while offering a partial set of research. If publishers have let these researchers’ horses out of the barn (in the traditional sense of to let as in rent out), what is to be done?

I come at this as a researcher who has explored how academic work constitutes – and thus should be legally regarded as – a different order of intellectual property (even as such work is at the origins of the intellectual property concept). The distinction between academic work and other forms of intellectual property lies in the work’s value not being served by restricting access to it. Rather, researchers create value by contributing, as widely as possible, to the work of others and the wellbeing of the world. This does result in those institutions that value research and teaching compensating researchers, with this transaction influenced by, while operating at a remove from, the circulation of the work.

So researchers have every reason to prop the barn door open. They have the means to set their horse free, pretty well anywhere they choose to publish their work (that being an important right in itself). While they can pay publishers article processing charges (APCs) to open the work, the vast majority of publishers also allow researchers to post a peer-reviewed draft of their studies at no charge in preprint servers and repositories prior to publication.

In exercising this right to free their work, researchers are taking a stand, in effect, with the democratization of knowledge, especially in light of the current U.S. presidency busily deleting vast troves of public health information, constricting support for research more generally, and showing, as one columnist puts it, “that it considers knowledge production worthless.” With court challenges now underway in defense of academic freedom in Canada and the U.S., making one’s work publicly available amounts to a good faith demonstration of what such freedom offers. Giving the public free rein to use their research on health, climate, economics, education, and the list goes on, not only contributes to the global circulation of research, but to the broader realm of public discourse. But, yes, public access means that such research will increasingly serve as training data for LLMs.

For researchers who object to LLMs freely feeding on their research, let me make an appeal to how this is, in fact, what distinguishes research in purpose and economy. Whether you consider specific instances, such as AI’s instrumental role in drug discovery, or more generally, as Google’s Gemini AI is now being consulted 10 million times a day, such uses seem closely aligned with what we want for our research, especially compared to the alternative of exposing the public to research-free artificial intelligence or the research that AI hallucinates in the absence of the real thing. As to research’s particular economy of recognition, AI systems are now crediting the sources consulted in its responses. For example, on asking Google search what proportion of research is now open access, its AI companion Gemini pops up with “50 percent,” while offering links to five sources behind its response, of which only one is a research publication (itself open access).

As for the economics and wealth-making of AI companies, which trouble some researchers, I’d point to how these companies’ business practices are areas of research investigation (around exploitative labor practices, for example), as are issues of representation, use, environmental impact, and metaphysics (see my earlier column), among other matters. Yet I’d also note that researchers have a professional interest in increased collegial, professional, and public use of research that AI is capable of providing. And, as such, we need to ensure that public access to AI resources remains a condition of our contribution. What better way is there, we might ask, of reminding people that public support for research continues to benefit them than to see such work repeatedly cited in responses to their queries.

And if anyone still finds it hard to let go of the research publishers’ windfall – especially as that windfall is not offsetting library subscription and author open access expenses – then consider this: if and when researchers do finally set their research free, then publishers will have nothing more to sell to AI companies. Once the horses are out of the barn, the research is there to pull for everyone.