A scientific mystery that took 10 years to solve was cracked in two days by Google’s artificial intelligence.
The tech giant’s latest AI development is dubbed “co-scientist” and is designed to act as a colleague for researchers, with its own ideas, theories and analysis.
Scientists at Imperial College London had spent a decade solving a mystery in the field of antimicrobial resistance (AMR), which creates superbugs that are immune to antibiotics and are expected to kill millions of people a year by 2050.
Using traditional research methods, the team had theorised and then proved how different bacteria are able to accrue new DNA which can make them more dangerous, and its study is now in the process of being published by Cell, the peer-reviewed journal.
After the work was finished, the scientists at Imperial partnered with Google to help test out the AI co-scientist feature.
The researchers asked the co-scientist – which uses many of Google’s Gemini AI models to pit various existing data and novel theories against each other – for ideas on how bacteria become immune to antibiotics.
Speed and accuracy of results ‘quite shocking’
Prof José Penadés, who co-led the experimental work at Imperial, told The Telegraph: “We worked for many years to understand this thing and we found the mechanism.
“Capsids (the protein shell of a virus) are produced with DNA inside and no tails. They have the ability to take a tail from different viruses and affect different species.”
While the team knew about this tail-gathering process, nobody else in the world did. Imperial’s revelations were private, there was nothing publicly available, and nothing was written online about it.
The scientists then asked the co-scientist AI, using a couple of written sentences, if it had any ideas as to how the bacteria operated.
Two days later, the AI made its own suggestions, which included what the Imperial scientists knew to be the right answer.
“This was the top one, it was the first hypothesis it suggested. It was, as you can imagine, quite shocking,” said Prof Penadés.
Dr Tiago Dias da Costa, a bacterial pathogenesis expert at Imperial and co-author of the study, added: “It’s about 10 years of research which was condensed in two days by co-scientist.”
While the AI was able to spit out the correct hypothesis within 48 hours of being asked, it was unable to do the experiments to prove it, which themselves took years of work.
However, the experts say if they had been given the hypothesis at the start of their project, before they drew up the theory themselves, it would have saved years of work.
‘Imagine how much time and money we could save’
“The system gives you an answer and that needs to be experimentally validated,” added Dr da Costa. “You cannot take the answer as a universal truth, so the scientific process would still have to happen.
“But 90 per cent of our experiments in the lab are failed experiments, and imagine if we have an AI collaborator that could guide us in reducing the failed experiments.
“Imagine how much time, grant money and, ultimately, taxpayer money we could save.”
The Google AI co-scientist system is still in its infancy and will continue to be refined with further work. But it is quick, easy to use, and simple, the Imperial researchers said.
The Imperial scientists were given a host of other ideas by the technology as to what may be driving AMR, some of which are now the focus of real-world research to see if they are also correct.
This includes a suggested explanation for a 70-year biological mystery, which preliminary experimental data suggest holds promise.
When the scientists, who have spent their entire careers trying to understand and unpick the mysteries of the microbial world, saw the results of the Google AI, they were astonished.
‘It was amazing – and very scary’
Prof Penadés was shopping on a weekend when the email came through from Google with the suggested hypotheses from co-scientist.
“I said to the person I was with to leave me alone for one hour in order to digest this,” he told The Telegraph.
“Half of me was thinking that this cannot be true and it is amazing, and the other half found it very scary. I have this feeling that we are involved in something that will change the way we do science. This is my personal feeling.”
AI is already widely used in science. It includes the Nobel Prize-winning AlphaFold technology, developed at Google DeepMind, which uses AI to correctly predict the shapes, structures and behaviours of proteins.
Scientists can now see, just from DNA code, how a protein looks and how it will interact with the body, drugs and other entities.
A tester version of co-scientist is now to be made freely available to researchers, and an application programming interface (API)I to allow websites to use the base technology is also to be published.
The co-scientist was also tested with researchers at Stanford University and Houston Methodist in the US to see if it could identify new targets to treat disease, and if any pre-existing drugs could treat other diseases.
The AI found a new target to try to treat liver fibrosis, and suggested that the drug Vorionostat, which is used to treat cancer of immune cells, could help treat the condition.
Government investing in AI
The Government is currently in the process of trying to ramp up the UK’s own AI infrastructure, with a focus on turning world-leading academic research into new uses for AI and commercial applications.
Last week, the Department for Science, Innovation and Technology approved a new project to use AI in science conducted in the UK.
This includes world-first trials that will integrate AI into the peer-review process to try to free up researchers from some of the more time-consuming tasks which distract from doing actual research.
A total of £4.8 million of taxpayers’ money has been shared among 23 research projects dedicated to using AI in science, including at Bath and Sheffield, to see if AI can improve peer-review.
Lord Vallance, the science minister, told The Telegraph last week: “AI presents new opportunities in a range of sectors, and if researchers can demonstrate its potential to increase transparency, robustness and trust in science, then this could pave the way to freeing them up from mundane paperwork tasks while driving growth.”