Open Science 2.0 – When Access Is No Longer Enough
open scienceA few days ago, I came across an article in The Scholarly Kitchen that made me pause. Written by Ashutosh Ghildiyal, Maria Machado, and Gareth Dyke, Open Science 2.0: Building Understanding in an AI-Mediated World presents an argument I believe deserves more attention in our community, especially here in the Netherlands, where open science goals are truly developing.
The main point is straightforward: we have succeeded in making research accessible. Today, research is more openly available than ever before. But simply having access was never the end goal. Now that most people see research findings through AI-generated answers instead of reading the original papers, the real question is not whether you can find the information, but whether you truly understand it.
From Open Science 1.0 to 2.0
Open Science 1.0 focused on removing paywalls, and for good reason. The Dutch goal of achieving 100% open access, the Barcelona Declaration on Open Research Information, and the movement for diamond open access all show a strong agreement that publicly funded research should be available to everyone. I’m proud of what our community has accomplished in this area.
However, the authors say that Open Science 2.0 requires more: active investment in four connected areas—openness and access, trust and verifiability, learning and expertise development, and wider communication and impact. The focus is not moving away from access, but recognising that access alone is not enough. As they say, “access is not understanding, and without understanding there can be no trust.”
What the AI Shift Actually Changes
This is the part of the argument I find most convincing, and it challenges a common assumption. Many people think that because AI makes knowledge easy to find, expertise is less important. Ghildiyal, Machado, and Dyke argue the opposite. As finding information becomes easier, good judgment becomes even more valuable. Skills like asking the right questions, recognising uncertainty, and evaluating conflicting evidence matter more, not less, when AI can quickly provide a convincing answer.
This has real practical effects on how we train researchers, how we design information literacy programs, and, I believe, how university presses and libraries see their roles. If AI shapes how findings are presented to clinicians, policymakers, or students, then people who understand both the science and the audience are essential. Deciding what to highlight, what to clarify, and what each audience truly needs to know cannot be left to a tool alone.
There is also an important structural issue: accountability is now spread out. In the past, editors, peer reviewers, and science journalists served as gatekeepers who provided context. Now, AI systems make these decisions, often relying on training data and design choices that even experts may not fully understand. No one has clearly taken responsibility for making contextual judgments in this new environment.
What This Means for Scholarly Communication
The authors argue for what they call “AI-legible metadata,” which means making a clear effort to ensure research is not only readable by people but also understandable to the systems that now help share it. This idea fits with current discussions in the Netherlands about open metadata and the Barcelona Declaration. Accurate metadata has always been important, but now it matters to both machines and people.
They also support bringing back long-form works like monographs and review articles as important tools for understanding, balancing the trend toward shorter, quicker publications. As someone who works at a university press, I find this encouraging, even if it might seem self-serving. Monographs were never just a way to deliver data; they were arguments about how to interpret evidence. That role has not been replaced by AI; in fact, it may be even more important now.
Plain-language communication should be seen as a key scholarly skill, not just an extra task that researchers do after their main work. Visual abstracts, easy-to-read summaries, and layered outputs are not about oversimplifying; they help bridge the gap between producing research and putting it to use.
A Note of Honest Hesitation
I do not want to suggest that this debate is over. Many in the scholarly community define Open Science 2.0 primarily in terms of technical changes such as open data, open-source hardware, and shared research infrastructure, rather than the cognitive and communication changes the authors highlight. Both views have value and need not conflict, but it is important to note that “Open Science 2.0” still lacks a single, clear definition.
There is also a criticism the authors may not emphasise enough: AI does not just share information; it can also amplify biases already present in research data and collections. Who is responsible for what AI systems learn and whose knowledge they use remains an open question.
The Opportunity
The authors end with a line that has stayed with me: “the most valuable outcome of open science may not be access to knowledge, but the ability to understand it well enough to act wisely.” I agree with this. It also seems like a clear call to action for libraries, university presses, and those working in scholarly communication—not just to control access, but to help create the conditions where knowledge can truly be used. That is the challenge Open Science 2.0 gives us.
Open Science 2.0 is not a criticism of what we have built. It is a challenge to keep building, to do so more thoughtfully, and to focus on understanding.
Read the original article: Open Science 2.0: Building Understanding in an AI-Mediated World — Ghildiyal, Machado & Dyke, The Scholarly Kitchen, July 2026.