Three Big Ideas #15
Suppressing suppressants, AI's revolving door, and growing pilot plants
Welcome to our weekly Three Big Ideas roundup, in which we serve up a curated selection of ideas (and our takes on them) in entrepreneurship, innovation, science and technology, handpicked by the team.
🍩 Eamonn Ives, Research Director
Christmas and the New Year period may have been a time of gastronomic excess for many, but – thanks to the continuing spread of novel appetite suppressant drugs – perhaps for fewer this time around than last. A recent paper studying how food habits changed in households with at least one GLP-1 (such as Ozempic or Wegovy) user found that grocery spending reduced “by approximately 6% within six months of adoption, with higher-income households reducing spending by nearly 9%.” Most of the fall in spending was explained by a radical reduction in calorie-dense, processed foods – including an “11% decline in savory snacks.”
Food brands who are threatened by demand drying up are not sitting idly by, however. As detailed in a New York Times long-read, companies are already developing ‘Ozempic-optimised’ product lines – playing up certain nutrients, like protein or fibre, or catering for other qualities that GLP-1 users find attractive, such as more convenient preparation and smaller portion sizes.
While innovation is usually only considered as a positive, it’s seldom a positive-sum game. As this example shows, one industry’s gain is very much another’s loss. But this process of creative destruction is rarely a single, isolated incident. More often than not, it triggers an innovation arms race as competing companies vie to respond to each other’s developments. Who will eventually win out in this instance is anyone’s guess – but what’s not up for debate is that this is the process of innovation, facilitated by flexible markets, being played out in real time.
🚪 Anastasia Bektimirova, Head of Science and Technology
Yesterday’s Lords Science and Technology Committee evidence session focused on the impact of immigration policy on science and universities. Naturally, the discussion touched on wider questions about the UK’s competitiveness as a destination for scientists and students. Speaking about AI research, Professor Alison Noble, Foreign Secretary at The Royal Society, said:
“The UK has a recognised strength in AI. We have some very good companies and a very strong ecosystem. I think that’s important for scientists as well. They might come in … and then a few years later, maybe go work in industry for a while, move around. This is what entrepreneurial-type scientists want to do, including people who are pure entrepreneurs as well. We are known for having a very active and very good environment for that. There is a concern about people going to work in industry, but if that’s where research is done in AI, then universities focus on other things. I’ve actually been around myself – my PhD was during the last wave of AI, so I’ve seen the whole circle of starting off, going through a winter and coming through again. So maybe I have a bit more balanced view on it. I think things go in cycles. But you don’t sit back when that happens – you decide as well, you work together, you have a strategy. I think that’s the important thing we need.”
An academia-industry revolving door drives innovation. The development of deep learning itself was a back-and-forth exchange. When renowned researchers like Geoffrey Hinton, formerly at Google, and Meta’s Yann LeCun left their university roles, what they gained from the vast resources of big tech companies continued to benefit academic researchers through their publications. Google’s Attention Is All You Need paper, laying out the theory behind transformer architecture, is one such example. We want to encourage cross-pollination of ideas, skills, and experience of working within various organisational structures.
For this reason, industry’s strengths shouldn’t breed complacency when it comes to ambition and capacity building in academia. Universities shouldn’t just sit back and “focus on other things” while industry advances AI research. Historically, AI innovation was academia-led, but now universities are being priced out. The work on large-scale AI models is increasingly out of reach for universities, largely due to high computing costs. Even though tech companies publish research papers, they are less transparent about the workings of their most advanced models, making resources for academia even more crucial. Resource constraints are not an exclusively UK issue – US universities don’t exactly match OpenAI’s compute capacity either. It’s not only about the science of AI itself but also application in other fields. It’s not uncommon to hear from PhD researchers and postdocs working on AI-driven biology who can’t access the compute they need. It’s particularly concerning to hear our talent compare their challenges to relatively easier compute access enjoyed by their peers stateside. What once seemed anecdotal stories now form a clear pattern.
The goal isn’t for academia to compete with industry but to work alongside it, each bringing their strengths to AI development and its applications. The timing or nature of cycles or disruptions are hard to forecast. What we can do is put strong foundations to ensure different parts of the AI ecosystem can thrive through them.
⚗️ Philip Salter, Founder
Eric Gilliam, writing for Asimov Press, argues that Edwin Cohn – a temperamental, entrepreneurial protein chemist at Harvard in the 1930s and 1940s – was one of the most underrated translational scientists of all time. Initially focused on pure research, Cohn’s lab was enlisted by the US military in 1940 to develop blood protein products to treat shock and blood loss. His team successfully created stable, concentrated human albumin, which was used extensively during the Second World War, including in the Normandy landings.
Cohn also demonstrated the pivotal role of pilot plants. Operated at a scale larger than lab experiments but smaller than full industrial production, these facilities replicate manufacturing processes so researchers can test methods under near-real-world conditions. By integrating a pilot plant into his Harvard lab, Cohn was able to scale production from laboratory experiments to 40-litre batches of plasma, develop manufacturing protocols for pharmaceutical companies, train industrial personnel (thereby ensuring smooth technology transfer), and rapidly produce enough material for clinical trials and battlefield use.
Gilliam thinks Cohn has something to teach us today about how pilot plants can accelerate both discovery and practical application. If more universities adopted the pilot plant model for challenges like producing synthetic blood or commercialising advanced materials – carbon nanotubes, graphene, aerogels, lithium-ion battery anodes – researchers could more quickly transition breakthroughs from the laboratory to the marketplace.