About the author: Amy Brand is Director and Publisher of the MIT Press, a role she has held since 2015. A cognitive scientist by training, she earned her PhD from MIT and has held leadership roles at CrossRef, Harvard, and Digital Science. She is a co-creator of the CRediT taxonomy, a founding member of the ORCID Board, and producer of the documentary Picture a Scientist. Brand is widely recognized for her contributions to research infrastructure, scholarly communication, and equity in science. Her honors include the Council of Science Editors Award and the AAAS Kavli Science Journalism Gold Award.
Science is essential for human progress. Also vital, if less obviously so, is how we disseminate science. As the system through which knowledge becomes discoverable and credible, publishing is a core component of research infrastructure.
How should publishing adapt to serve the needs of a research community facing censorship, diminished funding, and the rapid rise of generative AI systems that siphon knowledge while eroding its integrity? As someone who has worked for decades in research, university administration, and publishing, I have never seen the stakes as high as they are today for the futures of both science and science communication.
With socio-political systems breaking down around us, it is tempting to view generative AI as the alchemical solution to the world’s problems. Large language models (LLMs) have emerged as seductive portals to discovery, providing instant answers, seamless synthesis, the apparent democratization of knowledge. Yet their appeal conceals their potency as systems that also accelerate misinformation, fraud, and propagandizing. By design, they produce content that seems highly plausible but is often misleading or just plain wrong. It is dangerous for science when validation is difficult and expensive, while “truthiness” is cheap and profitable, especially given how easy it is to fool the human mind.
Yes, we want to hasten discovery and problem-solving in the face of harrowing global challenges. We want to believe in machines that can solve problems more swiftly than human cognitive architecture or our dysfunctional institutions. But when proponents of unrestricted AI training argue that it is a moral imperative to make all scientific content and data available to accelerate innovation, history counsels caution.
We have seen this before. The early internet was hailed as a democratizing force for expression and universal knowledge. In the end, a lack of regulation allowed massive commercial platforms to dominate the space, eroding trust and collapsing economic models for news and research content alike. We also know now that underbaked open access policies accelerated consolidation in publishing and created economic incentives to publish more with lesser quality control.
Let’s pause and consider what’s best for human understanding, learning, and the progress of knowledge. When search just leads to AI summaries and users do not click through to original sources, and when the simple act of reading for pleasure is declining precipitously, how do we avoid a future in which we author and publish content for machine consumption only?
I believe the vital role publishers play in supporting research impact and integrity is worth preserving and protecting, especially now. I understand frustration with high pay-to-publish fees and subscription paywalls, particularly from larger publishers who have leveraged a market for academic prestige that hasn’t historically been price-sensitive. But our industry is not one monolithic profit-seeking entity. Non-profit publishers, like MIT Press and many scientific societies, operate with different values and thinner margins.
Indeed, we are more threatened by the AI land grab and sometimes even by the very open-science movements we have long supported. Hence, we must resist moral posturing that ignores the power dynamics at play. There is a halo around the idea of “open” that can obscure real-world complexities like economics and incentives. In the end, not all openness is virtuous; not all resistance to openness is obstructionist.
Another misconception is that publisher interests are misaligned with those of researchers. We recently conducted a large survey of authors across STEM fields regarding unauthorized use of their work for LLM training. The vast majority object to this practice even while believing AI promises transformative pathways for discovery and learning. They expect to be able to consent, or not, to such usage and expect attribution when their work informs LLM outputs. They do not equate open-to-read with open-to-train-on.
So too, many are skeptical of the integrity of large AI companies, and worry about how LLMs will impact publishing, reading, writing, critical thinking, and creativity; flatten diverse viewpoints; and reinforce biases and cultural hegemonies. They worry deeply about what is lost when we break human-authored works into tokenized training data and feed them to models that can’t preserve their context or argument.
The question of how, and under what conditions, published science is used to train LLMs is not just about copyright. It is about who controls the future of knowledge. Do we cede authority to opaque, extractive industries with little accountability to the research community? Or do we build systems that preserve attribution, integrity, and sustainability? If we are serious about human flourishing, about evidence-based science, and about protecting the conditions under which knowledge grows, then the research community and its institutions must proceed with discernment.
Whose interests does it serve to give published science and scholarship away to a highly extractive, opaque tech sector? Follow the logic: all the value in the open publishing fees paid by authors, institutions, and funders is then ultimately handed to the likes of OpenAI and Anthropic. When it comes to extractive industries, let’s also be honest in assessing how the values of AI sector compare with those of academic publishers.
I myself remain optimistic that, with careful planning and evidence-based policy, we will use AI to improve peer review, enhance reproducibility, and streamline publishing workflows and costs. We may even be able to build solutions that help sustain the good parts of scientific publishing.
But the current paradigm, where the published record is mined without consent and monetized by private tech giants, is unethical and destructive for science and scholarship. It also reflects magical thinking about how our complex world works and how we will make real progress in solving the existential problems facing us.
Disclaimer
The information, opinions and recommendations presented in our guest blogs are those of the individual contributors, and do not necessarily reflect the values and beliefs of the International Science Council
Receive future open science round-ups directly into your inbox – subscribe now.
Photo by Martin Adams on Unsplash
Read more about ISC Forum on Publishing and Research Assessment 
Read more about New Steering Group appointed to guide the ISC Forum on Publishing and Research Assessment 
Read more about UN Scientific Advisory Board adopts landmark statement on Open Science 
Read more about Scientific fraud and the reform of science publishing 
