Who Should Innovate for Science? Prophy Predicts [Webinar Recap]

This is a recap of the Prophy Predicts webinar—"Who Should Innovate for Science?"—hosted on June 25, 2026. The full recording is available on YouTube: https://www.youtube.com/watch?v=BePsZfwqB0M. 

In 1991, a physicist named Paul Ginsparg wrote a script on a spare account at Los Alamos so he and his colleagues could stop waiting for papers to arrive in the mail. It grew into arXiv—now the delivery system for a large share of the world's physics and math. Nobody hired him to build infrastructure for science. He built it because he was living inside the problem it solved. Three decades later, the question he answered by doing is back, sharper: as AI compresses how fast anyone can stand up a tool, who should be doing this work—and who shouldn't?

In episode 5 of Prophy Predicts, we put that question to four people who sit at different points around it: a clinical-research CEO, a scholarly-collaboration platform founder, a science-communication veteran, and—breaking our usual convener-only stance—Prophy's own CEO.

They answered the same three framing questions in advance, without comparing notes. What came back wasn't four different answers so much as one instinct, converging fast, then pulling apart live on the details: how far participation should widen, whether an AI agent can be said to innovate at all, and who gets to police "good motives" once everyone agrees that's the bar.

About the Webinar

Prophy Predicts is Prophy's monthly panel on the future of scientific infrastructure. Episode 5 asked one pointed question: who should—and shouldn't—innovate for science?

Panelists:

• Ena Bromley — CEO, Oyanalytika
• Ben Kaube — Co-founder, Cassyni
• Charlie Rapple — Co-founder, Kudos
• Oleg Ruchayskiy — CEO & Co-founder, Prophy

Moderated by Gareth Dyke, Partnerships Director at Prophy.

AI gets you the knowledge. It stops short of the wisdom.

Ena Bromley opened from the widest possible frame—in principle, anyone could contribute—then narrowed it fast. Qualification, for her, comes down to subject knowledge deep enough to weigh a technology's benefits against its risks, because half-knowledge is its own hazard in a world where anyone can self-teach from a search bar. AI changes the scale of what's knowable, not who's accountable for using it well: it can put enormous knowledge on the table, but the judgment about what to do with that knowledge—the tradeoff between what's possible and what's wise—still has to be made by a person. She drew the line at political motive: whoever is doing the innovating, keep governments and parties out of the room.

"We are fortunate that AI can bring a lot of knowledge to the table. But where AI stops is really that intersection between knowledge and wisdom."

Who innovates matters less than why they're doing it.

Bromley's second point reframed the whole panel. Drawing on her work in clinical trials, she pointed out how easy it is to measure the wrong thing—"increased survival by X months" looks like success on paper while missing what a patient's quality of life actually became. That same tunnel vision shows up in who gets to fund innovation: leave it to governments alone, and you inherit whatever a given government's constituents and politics want measured. Bring in industry and venture capital alongside public funding, and innovation stays freer to chase the outcome that actually matters rather than the one that's easiest to report.

"It's not only governments—we also have industry and private sector that fund. And that's the exact reason why it's so important that funding shouldn't only come from government or political organizations."

Trust has to scale with who's contributing.

Ben Kaube pushed back on the premise before answering it. The question "who should innovate" implies some special group gets to decide for everyone else—but who's actually contributing to science keeps expanding on its own, geographically (a wave of contribution now out of China, India, and the Global South) and structurally (industry, not-for-profits, frontier AI labs, and—his phrase—"a whole bunch of new, strange colleagues" in the form of AI agents working autonomously for hours at a stretch). For Kaube, the honest answer to "who" is: more people, in more places, than the field has ever had.

What matters more, in his view, is whether the infrastructure that keeps knowledge coherent and trustworthy can keep pace with how fast the contributor base is growing.

"For me, it's less about who—it's about making sure that the infrastructure of trust scales with the individuals that are contributing."

Peer review needs tools built for it—not borrowed ones.

Kaube's second point was a caution wearing the shape of a compliment. General-purpose AI tools are genuinely powerful—he's not questioning the capability. He's questioning the fit: those tools were designed by teams solving for a million use cases, and peer review, by his estimate, isn't anywhere near their top ten. That's fine when the stakes are low—using ChatGPT to research a topic, or prototype a workflow for yourself. It stops being fine the moment those same general-purpose tools quietly become the substrate that scientific infrastructure runs on, because the edge cases you can safely ignore when building something for yourself become the ones that matter most when you're building for everyone else.

"The designer of these general-purpose tools is thinking about a million other use cases. I'm pretty sure peer review is not anywhere near their top ten."

Understanding what science is for is the real qualification.

Charlie Rapple came at the question from the communication-tooling side of the field, and landed close to Kaube's answer from a different angle. Her criteria for who should build for science: someone who understands what science actually is—not a neat, linear process, but one that tolerates uncertainty, absorbs wrong answers, and resists having a foreign logic imposed on it. That's a specific, learnable orientation, and it's not automatically held by the people who are best at the science itself.

"The ideal person innovating would be somebody who understands what science is for—who gets that science is not a nice, neat, linear process."

Who shouldn't build for science

Rapple's list of disqualifications was the panel's sharpest moment. Scientists themselves often shouldn't lead the building—not because they lack the domain knowledge, but because running the infrastructure is a different skill from doing the research, and pulling a good researcher into a role they're not suited for serves neither the researcher nor the science (her analogy: it's like expecting a farmer to run a supermarket—the supply chain, the logistics, the point-of-sale technology are nowhere near the skill of growing the food). Publisher-led innovation is a second-order risk—well-intentioned, but liable to end up defending an existing business rather than serving science. And two groups drew flat rejections: what she called the "tech-bro oligarchy," building primarily for the money, and governments carrying a geopolitical agenda into the room.

"It's like expecting a farmer to run a supermarket—the supply chain, the logistics, the point-of-sale technology. It is nowhere near the skill set of growing the food."

Former scientists build the tools that last.

Oleg Ruchayskiy noted something the moderator caught live: all three other panelists had converged, independently and without coordinating, on some version of "deep subject knowledge, understanding what science is for." His answer sharpened that into a specific claim—the strongest examples of infrastructure that actually shaped a field weren't built by large organizations; they were built by individual former scientists. Stephen Wolfram and Mathematica. Paul Ginsparg and arXiv, built around the same time he stopped publishing in his own field. A former scientist has already seen where the workflow breaks and where the time gets wasted, has practiced the generalizing instinct science trains into you, and—not incidentally—is trusted a priori by the researchers they're building for, because those researchers read them as one of their own guild.

"The former scientists are the best people to innovate for science."

Building it is science. Maintaining it is a business.

Ruchayskiy's second point explained why Prophy itself is a company and not a lab project. Scientists innovate, then move on—that's the nature of the job: solve the problem, publish, advance. What that leaves behind is a maintenance question nobody in academia is positioned to answer: who stays behind to keep the thing running once its creator has moved to the next question? He illustrated it with a small, sharp anecdote—a piece of niche academic software licensed for $20 a year, whose sole maintainer stopped answering email; a colleague of his needed an AI model's help just to keep using software he'd paid for a decade. Grants can't compete with industry salaries for the developers who'd actually maintain infrastructure at scale, which is the specific gap Prophy moved into commercial territory to close.

"If you really care about creating infrastructure, you have to move into some kind of non-academic sector."

Where the panel converged

Three things held across all four answers, unprompted:

• Depth over headcount. Every panelist's qualifying bar was some version of understanding what science is for—not credentials, not access, not good intentions alone. Ruchayskiy pointed out live that none of them had compared notes beforehand.
• Communication and transparency as the connective tissue. Bromley's "open, safe platform," Kaube's "coherent, trustworthy" knowledge base, Rapple's "trust comes from transparency"—different vocabularies for the same load-bearing idea.
• Politics is the one disqualifier nobody softened. Funding from government is fine; governments leading innovation, or carrying a geopolitical agenda into it, was the single point where all four panelists landed in exactly the same place without being asked twice.

What this means for scientific publishers

• On reviewer trust: the panel's "former scientists earn a-priori trust" argument is also an argument about tooling—a reviewer-matching system built by people who understand what peer review is actually for holds up differently than one bolted on from a general-purpose model. Prophy's own Referee Finder was built on that premise: "For editorial departments that need to find high-quality peer reviewers quickly, Prophy offers an extensive database of over 99 million researchers' profiles—access to a larger, well-organized pool of expert reviewers, ensuring an efficient and informed peer review selection process."
• On purpose-built vs. general-purpose: Kaube's warning about tools "not built with science in mind" quietly underpinning scientific infrastructure is precisely the risk editorial teams take on when they let a general-purpose assistant stand in for a reviewer-matching system with no conflict-of-interest checks, no discipline-specific ranking, and no audit trail.
• On transparency: Rapple's "trust comes from transparency" is a workflow requirement, not a slogan—it shows up as visible conflict-of-interest detection and a documented, explainable path from manuscript to reviewer shortlist.

About Prophy Predicts

Prophy Predicts is Prophy's monthly panel series on the future of scientific infrastructure—peer review, publishing, and the tools researchers rely on. Watch the full recording: https://youtu.be/BePsZfwqB0M.

Prophy helps publishers identify relevant reviewers, detect conflicts of interest, and discover related research using a database of 99M+ researcher profiles and 194M+ publication records. Learn more at prophy.ai