From Reproducibility to Responsibility: Open Science in the Age of AI

This post accompanies an I4R podcast episode featuring Katharina Miller, based on her contribution to the Initiative for Reproducibility workshop in Barcelona in July 2026.

Reproducibility has become one of the central pillars of trustworthy science. Better methods, transparent reporting, accessible data and robust replication practices have helped transform how research quality is understood and assessed.

However, the environment in which science operates is changing.

Artificial Intelligence (AI) is becoming embedded across the research lifecycle. Scientific publications and datasets circulate globally within seconds and increasingly serve as inputs for AI systems. At the same time, questions of intellectual property, licensing, research integrity, innovation and knowledge valorisation are becoming part of researchers’ everyday decisions.

Against this background, an important question emerged during the I4R workshop in Barcelona:

If the scientific community has spent the past decade asking how science can become more reproducible, should the next decade also ask how science can become more responsible?

Reproducibility is becoming a governance question

For many years, the reproducibility debate has rightly concentrated on methodology: better statistics, reporting, protocols, data management and transparency. These remain indispensable foundations of trustworthy science.

Yet the research landscape is changing profoundly. Generative AI is already being used for literature searches, language editing, translation, coding support, transcription, metadata generation and data analysis. Publishers are incorporating AI into editorial workflows, while AI-assisted tools are also emerging in research integrity screening.

These developments bring opportunities, but also significant challenges. Hallucinations and non-deterministic outputs can affect reproducibility. AI-generated manuscripts may increase pressure on peer review. Questions remain about copyright, licensing, text and data mining, attribution and consent. Researchers are also concerned about deskilling and overreliance on AI.

Scientific publications and other research outputs have themselves become valuable resources for training AI systems. The implications reach beyond scientific quality. They concern the governance of scientific knowledge itself.

Reproducibility is therefore no longer only a methodological challenge. Increasingly, it is also a governance challenge.

Open Science and Intellectual Property: a false opposition?

This is also the starting point of IP4OS – Intellectual Property for Open Science, a Horizon Europe project addressing a growing paradox in European research.

The European Union (EU) has become a global leader in Open Science (OS). Open Access, FAIR data, research integrity, citizen science and international collaboration have contributed to making science more transparent, collaborative and reproducible.

At the same time, researchers increasingly face practical questions that do not fit neatly into an “open versus closed” distinction. Should an invention be protected before the underlying findings are published? Which licence should be used for a particular research output? Can an openly available dataset be used to train commercial AI systems? Who owns AI-generated scientific outputs? How can researchers collaborate openly while preserving opportunities for innovation and societal impact?

These are no longer niche legal questions. They are becoming everyday research questions.

OS and Intellectual Property (IP) should therefore not automatically be understood as opposing concepts. Both can be instruments for governing knowledge.

The objective is not to make science less open. Nor is it to maximise IP protection. The objective is to make openness sustainable, responsible and strategically governed so that it can continue to generate scientific excellence, innovation and societal benefit.

Poorly governed intellectual property creates uncertainty. Uncertainty can discourage collaboration, delay sharing and make both universities and industry hesitant to engage. Good IP governance, by contrast, can create confidence and enable collaboration.

The relevant question is therefore neither more IP nor less IP, but better governance of knowledge assets.

Generative AI changes the equation

These questions are also at the centre of the STOA study for the European Parliament, “Open Science and the Effects of Generative Artificial Intelligence in Scientific Exchange in the European Union.”

Rather than focusing primarily on what AI can technically do, the study examines how the interaction between OS and GenAI is reshaping scientific exchange, research integrity and research governance in Europe.

This distinction matters.

The future of scientific exchange will not be determined by technology alone. It will also be shaped by institutions, incentives, legal frameworks and societal values. Decisions taken now about access, attribution, licensing, AI use, research integrity and knowledge governance will influence what the research ecosystem looks like in the future.

The challenge is therefore not simply to adapt OS to AI. It is to consider how scientific knowledge can remain open and reusable while ensuring that openness continues to serve researchers, institutions and society.

Towards responsible knowledge ecosystems

This is where several conversations that are often conducted separately begin to converge:

Research Integrity, reproducibility, OS, FAIR principles, Responsible AI, IP, technology transfer, citizen engagement, innovation and public value creation.

These objectives should not compete with one another. They should reinforce one another.

Responsibility can provide the connecting principle.

Responsible openness does not mean restricting access to knowledge. Equally, openness cannot mean abandoning responsibility for how knowledge is created, shared and reused.

Instead, responsible openness requires governance: governance that enables researchers to make informed decisions, creates clarity rather than uncertainty and supports both scientific excellence and societal impact.

Universities as stewards of knowledge ecosystems

This development also has implications for universities and research organisations.

Researchers cannot realistically be expected to become experts simultaneously in AI governance, copyright, licensing, ethics, cybersecurity, technology transfer and research integrity. Asking individual researchers to navigate this increasingly complex environment alone risks turning responsible research into an additional administrative burden.

Institutions therefore need more integrated governance structures.

Lawyers need to work alongside librarians. Technology transfer offices need to collaborate with ethics committees. Research integrity experts need to engage with AI specialists. Training should move beyond explaining rules and towards empowering researchers to make informed choices.

The goal should be simple: governance should make responsible decisions easier, not more complicated.

Universities, from this perspective, are no longer only producers of knowledge. They are increasingly becoming stewards of knowledge ecosystems.

The next evolution of Open Science?

Perhaps this transformation can also be understood as part of the evolution of Open Science itself.

A first generation concentrated strongly on access, making scientific publications and knowledge openly available.

A second expanded the focus towards reproducibility, transparency and FAIR data, strengthening the quality and reliability of research.

The emerging challenge may be a third dimension: responsibility.

This does not replace openness. It does not replace reproducibility. Rather, it integrates both with governance.

Responsibility means ensuring that research can remain reproducible, transparent and open while also being trustworthy, legally robust, ethically governed and capable of creating value for society.

The EU is particularly well positioned to contribute to this development. Its scientific capacity, legal traditions, commitment to fundamental rights and experience with Open Science provide an opportunity to demonstrate that openness and competitiveness do not necessarily have to be opposing objectives.

Connecting reproducibility, OS, Responsible AI and IP could strengthen not only research integrity, but also Europe’s capacity to innovate and to preserve the societal value of scientific knowledge.

From producing knowledge to governing it wisely

Reproducibility asks to what extent research results agree with those of replication studies.

OS asks how scientific knowledge can be shared.

IP4OS asks how scientific knowledge can be governed responsibly.

And the emergence of GenAI asks how research systems should prepare for a future in which the way scientific knowledge is created, exchanged and used is changing fundamentally.

These are not separate conversations. They are different parts of the same challenge.

The aim should be a research ecosystem that is open enough to accelerate discovery, robust enough to deserve trust, innovative enough to create societal value, and responsible enough to safeguard the future of scientific knowledge.

Perhaps, then, the next question for OS is no longer only how we produce, reproduce and share knowledge.

It might also be: How do we govern knowledge wisely?