{"text":[[{"start":7.26,"text":"The writer is director of the EMBL’s European Bioinformatics Institute"}],[{"start":12.24,"text":"I was lucky enough to be in the engine room of the genomics revolution in the early 2000s. I wrote the software — and co-led the team — that predicted the locations of the genes in the human genome. It was a heady time. New academic departments were magicked into existence, companies with preposterous business plans were founded and wild statements about the future of biology were made. Great science happened — though not always as predicted."}],[{"start":40,"text":"I have not felt the same zeitgeist in science until the recent deepening of AI methods. Once again, there is a transformative technology whose full capabilities we do not know; again, bold companies are striving to make their vision become reality (and reap massive rewards); again, academia is having existential conversations about its role while science moves faster than legislation. And once again, ethical concerns are worrying the public."}],[{"start":67,"text":"Not every aspect is the same. The capital investment in AI is far greater, for example. But I find both reassurance and concern in the similarities."}],[{"start":76.72,"text":"First, reassurance that the scientific step change is real. The AI system AlphaFold’s performance in the protein structure competition was a watershed moment in 2020, but I had already had my head turned by how deep neural networks could tackle biological problems years earlier. The way AI can handle complex data and find effective, non-linear functions to perform certain tasks is near magical (it is a surprisingly trial-and-error process; one has to throw things at the wall and see what sticks). AI offers the opportunity to attack the complexity of biological systems in new ways. We can tackle problems we thought impossible, including modelling multi-morbidity at an individual level, to take one example from my research group."}],[{"start":119.48,"text":"I am sure more “impossible” problems will be solved although it’s foolish to try to predict which — just as we could not predict which parts of biology would be most affected by large-scale DNA sequencing."}],[{"start":129.76,"text":"But just as with genomics, the ethical underpinnings must catch up with what is possible. In the early 2000s those concerns included the privacy of genomic information, gene patenting and the risk of reviving eugenic thinking. Regulators answered them by funding research within the Human Genome Project, formulating responsive ethical assessment schemes and passing anti-discrimination laws. They proved that we first need principles and the ability for regulators to deal with the examples in front of them."}],[{"start":161.82,"text":"My concern is that the hype today is way beyond a reality we don’t fully understand. The claim by some AI leaders that we will cure all human diseases in five years is hubris. We don’t even have good definitions of all diseases, and many of them involve the most complicated object we know in the universe — the human brain. This leaves aside the labour-intensive business of proving that cures work. When glitzy headlines disappoint, the entire field suffers."}],[{"start":190.16,"text":"We don’t need fantastical claims. In uncovering how things work, we must stay humble. From development to infection response to neural circuits leading to behaviours, there are oceans of ignorance in how life works. Large language models will not magic new observations out of the ether. But they do not need to. With the right datasets, domain-specific AIs such as AlphaFold, ChromBPNet, Delphi-2M and others — are cracking open the tough nuts in biology. With human scientists, AI agents and AI generating effective models of the world, there is plenty more insight to be gained."}],[{"start":227.88,"text":"We do not need to justify the role of new technology in biology with outlandish claims. We need to roll up our sleeves, double down on making discoveries of how living things work, share them with the world, and use them to build a more informed, healthier future for all of us."}],[{"start":246.72,"text":""}]],"url":"https://audio.ftcn.net.cn/album/a_1789746574_4646.mp3"}