Scientific Practice – No More Business as Usual

With opportunities and threats to scientific practice coming from multiple fronts, few people see the scientific method and surrounding infrastructure that have remained largely unchanged for at least a century (arguably longer) surviving unaltered for another decade, let alone another century. At #LINO75, the changing face of science was tackled head on across multiple sessions.

Open and Transparent AI

Kicking things off at 7am on Monday 29 June, the Swiss State Secretariat for Education, Research and Innovation hosted a Partner Breakfast on by far the most pervasive and talked about disruptor of scientific discovery at the moment: AI.

In her opening address, it was clear State Secretary Martina Hirayama had her finger on the pulse: “Honestly, the question we are here to explore – how AI can strengthen science without undermining it – is one that keeps policymakers, researchers and institutions across the world busy right now, and rightly so,” she said. “What does it mean for scientific integrity when an algorithm reaches a conclusion that no human can fully explain? How do we ensure reproducibility when models are trained on data we cannot access and verify? These are not abstract philosophical questions: researchers face them today, at their desks, in their labs.”

Switzerland’s response, she said, has been to invest in a computational backbone for AI that is open and transparent by design, providing researchers of all creeds with access to frontier models. One important output of this investment is Apterus, a Swiss-made large-scale, fully open, multilingual, large language model (LLM), and a direct response to existing commercial black box LLMs. “What sets it apart for science is its approach to accountability,” she argued. “Because if a model reaches a conclusion that cannot be inspected or reproduced, it is not a scientific tool, it is an oracle.”

Guests Ana Klimović (Head of the Efficient Architectures and Systems Lab at ETH Zurich, Switzerland) and Marcel Trefny (Ludwig-Maximilians-University, Germany), as well as 2002 Nobel Prize in Chemistry recipient (and Swiss national) Kurt Wüthrich, all agreed that transparency is key to using AI in scientific discovery, but they also highlighted some of the huge benefits AI can bring to the scientific workflow.

“AI coding tools have really increased in capabilities, especially in the last six months or so […] and this is having a lot of impact on the kind of research we do […] allowing us to prototype systems much more easily,” explained Klimović. “Trefny agreed, adding that – as a biologist – AI has sped up coding for him too, as well as literature searches and of course protein prediction through AlphaFold’s transformative capabilities (for which Demis Hassabis and John Jumper received the 2024 Nobel Prize in Chemistry).

And Trefny believes AI will be integrated even deeper into the scientific workflow going forward. “For example, Google’s Co-Scientist [recently] really streamlined the whole scientific process by AI, and in the end they just needed humans to do the wet lab,” he said. “So at some point some of us might become the tool, not vice versa.”

Though Trefny was (hopefully) joking, the role of humans will undoubtedly change. Therefore, it is important to recognize what scientists, representing humanity, bring to science. Wüthrich used Swiss downhill skiing sensation Dominique Gisin to illustrate a uniquely human characteristic: grit.

He said that after reaching the top of her sport, including Olympic gold, Gisin turned her hand to particle physics. But soon after earning her Master’s, she stopped. The reason? She had realized that to reach the elite tiers of research requires a similar level of devotion as it does to reach the pinnacle of elite sports, and she was not prepared to make those sacrifices again. “Scientific breakthroughs are not routine,” said Wüthrich. “And computers will not […] go to that kind of devotion.”

A Publishing Machine Not Fit for Purpose

Somewhat overlapping the disruptive effect that AI is having on scientific practice is how it is changing the face of scientific publishing. For many, AI represents a clear and present danger to the scientific record, being used to not only assist with and improve writing, but also in the most extreme cases to do everything: from drawing up hypotheses and conducting synthetic experiments to writing and formatting the final paper to suit specific journal requirements. Moreover, to varying degrees, it is also being used by a large proportion of peer-reviewers to review papers.

Yet it is not the only factor placing strain on the publishing process. Wednesday 1 July saw a Synergy Talk among Martin Chalfie (2008 Nobel Prize in Chemistry), William Kaelin Jr (2019 Nobel Prize in Physiology/Medicine), Randy Schekman (2013 Nobel Prize in Physiology/Medicine) and Thomas Südhof (2013 Nobel Prize in Physiology/Medicine) on this topic. Though they did discuss AI during the Q&A session, the main thrust of the conversation focused on what the panellists saw as the deluge of low-quality papers entering the scientific record and the various (non-AI) toxic influences on scholarship.

Discussing the former, Kaelin and Südhof animatedly called for researchers to ignore funders’ and publishers’ “unhealthy preoccupation with impact” and just focus on conducting good science. “The scientists I admire, and the ones who did work that stood the test of time, often came up with multiple ways of supporting their overarching conclusion,” said Kaelin. “Increasingly, papers contain multiple claims, and each may be supported by a thin thread of support.”

When it came to the toxic influences on scholarship, the panellists were all of the same view: the publish-or-perish culture in science has combined with the commercial motives of major publishers to corrupt publishing. Schekman was particularly outspoken on this issue. “[Springer Nature] has among the highest profit margins of any company in the world, and it’s no wonder,” he exclaimed. “Because who does the work for them? We do the reviewing for them, we do it for free because that’s part of our academic responsibility, and yet they charge an arm and a leg for the privilege of publishing.”

This topic brought out a similar passion in Südhof: “Journals are […] probably the biggest industry that faces zero regulation,” he said. “You have to regulate drugs, for example, before you can sell them, but you can sell peer-reviewed articles without any regulation whatsoever.”

Despite the rise of preprint archives, the scientific publishing industry remains the primary recordkeeper and disseminator of scientific breakthroughs. It was therefore a shame no representative from one of the major publishers was on the panel to defend an industry Chalfie described as “a necessary evil”.

Commercial Concerns?

Linked to both AI and new modes of publishing is the increasing commercialisation of science. On Thursday 2 July, a Synergy Talk provided perspectives from three Laureates who have successfully commercialized their scientific discoveries.

Up to 2014, John Martinis (2025 Nobel Prize in Physics) was a career academic – albeit one who had conducted a Nobel Prize-worthy experiment, later regarded as the basis for superconducting quantum computing. But then he was hired by Google to lead their quantum computing hardware effort in 2014. In 2019, his team announced that their Sycamore processor, a 53-qubit superconducting quantum chip, had achieved ‘quantum supremacy’, meaning it had solved a problem considered impossible for classical supercomputers. After leaving Google, Martinis co-founded Qolab, a private hardware company aiming to build scalable quantum computing systems.

In contrast to Martinis, Richard Schrock (2005 Nobel Prize in Chemistry) started his career in industry, working at materials production company DuPont where he experimented with organometallic complexes. His development of the metathesis method in organic synthesis earned him a Nobel Prize, and formed the foundation for co-founding XiMo to develop and apply proprietary metathesis catalysts. Verbio AG acquired XiMo to scale these catalysts, which allow the chemical industry to swap fossil fuels for bio-based alternatives (like rapeseed oils) to manufacture high-value pharmaceuticals, high-strength polymers, etc, cleanly.

Sir Gregory Winter (2018 Nobel Prize in Chemistry) is a pioneer of antibody engineering. He founded Cambridge Antibody Technology (CAT) in 1989 from which he developed HUMIRA. Licensed to Abbott Laboratories, HUMIRA went on to become the world’s top selling pharmaceutical. CAT and Domantis, another startup Winter founded in 2000, were both acquired in 2006 in major multi-million pound deals. Most recently, he founded Bicycle Therapeutics to develop very small protein mimics.

Speaking from experience, each panellist expressed how commercialization broadly benefits both science and society, as Winter summarized eloquently. “Is commercialisation good for science?,” he asked. “In my experience, the answer is ‘yes’ if we want to see science applied, because actually it’s the only way it’s going to happen.”

However, they were not blind to the pitfalls. One was the tension between open science and the corporate need to hold trade secrets and patents. Oftentimes, patents actually serve to ensure ideas get published while protecting the commercial rights of those who have invented them. But in certain circumstances, they should not apply.

Winter used the human genome as an example. “A commercial company was going to sequence it and patent the genes, but the Wellcome Trust stepped in […] to ensure that it was done in the public domain,” he recalled. In fact, the Wellcome Trust dramatically broadcast the sequences on a notice board in front of its building as the results were coming out. “If they hadn’t done that, we would have found that all our genes were patented, and it would have been very difficult to do research subsequently.”

In quantum computing, patenting and trade secrets could become a problem too. If just a handful of mega-corporations hold the keys to quantum computers, they would exclusively control a technology with the power to break today’s encryption protocols used across all industries and society. But Martinis sees no viable alternative right now. “It’s so complicated that it’s hard to fund it in any other way,” he said. “Although there have been some successes with university groups starting up – good ideas can prevail – it’s hard for a small company to compete against these big companies.”

Another pitfall highlighted during the discussion was how industry dictates the parameters of what scientists are allowed to explore. Though Schrock said he was given free rein to explore whatever he wanted at DuPont, he believes this kind of research in a company is rare nowadays.

In fact, it was a lack of freedom in his research that drove Martinis to both join and quit Google. “I personally want to build a useful quantum computer […] and I felt that it was going to be hard or impossible to do it in the regular academic framework,” Martinis said. “But in the end, the Google management wanted to run the project, and since I had a different view of what to do, they took over and it didn’t make sense for me.”

“I miss the large amount of funding in a big company, but I don’t miss the politics and bureaucracy,” he continued. “I think you have to make a personal decision. What do you want to do with your life? What do you want to do with your scientific career?”

And this is the ultimate question from #LINO75 for the next generation of researchers. They can no longer afford to be passive observers of changes happening in their discipline, they need to help actively steer the trajectory of science if it is to continue to serve humanity’s best interests.