DAVOS, SWITZERLAND - Thanks to advances in artificial intelligence (AI), scientists today are able to make breakthroughs that once would have been almost impossible, and the rate of scientific discovery is accelerating.
While AI has been used in research for years, the world is on the cusp of an AI-driven revolution in how new knowledge is discovered and used, according to Top 10 Emerging Technologies of 2024, a new World Economic Forum report.
The report lists “AI for scientific discovery” as one of 10 technologies poised to significantly influence societies and economies in the near future, as researchers use deep learning, generative AI and other foundation models to mine scientific literature, brainstorm new hypotheses, use deep learning to make discoveries and more.
Here, three technologists give their views on AI in science and their hopes for what it could achieve.
AI as a scientific generalist
“I’m very intrigued with the potential of how much creativity AI can bring to a scientific discipline,” says Dr Cheryl Cui, CEO of global industrial biotechnology company Bota Bio.
“Right now, all the disciplines are very specialized. It’s harder to have a computational biologist be a very good chemist." And yet “a lot of interesting findings happen in the intersections”, she points out.
Cui says that AI is a very good generalist – “probably already better than a lot of scientists’ comprehension of other disciplines”.
“How we can channel that and use it to its fullest extent – I think that will be the quest for the whole field.”
Expanding possible hypotheses
“Scientific discovery is one of the most important and most exciting areas for AI, because it goes to the heart of innovation across the world and across society,” says Michael Spranger, COO of Sony AI Inc, the company’s strategic research and development organization.
He thinks that engineering and computer science will benefit from “fully embracing this technology” and envisages a scenario in which “the next neural network architectures in AI are actually designed with AI”.
“That could be a major step change,” says Spranger, “because the underlying assumption for AI for science is that there are other hypotheses that humans can come up with. And probably we haven’t explored all of this and we should continue exploring it.
“But there’s probably also a set of hypotheses that humans cannot come up with for a variety of biases that are in some cases cultural or in some cases basically innate to humans.
“And so going beyond that and using machines to expand a set of possible hypotheses that we would have never considered – that’s really where AI will shine.”
Accelerating drug discovery
“You can now use AI imagination to create molecules with the desired properties that do not exist in the known chemical space,” says Alex Zhavoronkov, CEO of generative AI-driven biotech company Insilico Medicine. “So something that is precisely tailored for specific protein targets driving major diseases, that does not exist in nature, can be created by AI.
“I think that those methods are underappreciated right now,” he says. “People do not realize the impact generative chemistry is currently making on new materials design and especially on drug discovery.”
In the past two years, Insilico has nominated 18 pre-clinical candidates (a drug one step before it goes to human clinical trials) by utilizing generative AI. “Usually it would take us four or five years in order to achieve the same goal and probably would cost us tenfold,” says Zhavoronkov.
“So now AI allows for substantial cost savings, time savings, increased probability of success, and allows you to go into territories that previously were not economically feasible. And this