In 48 hours, Google's co-scientist AI tool cracks complex superbug problem that took scientists a decade to solve

[Hip]

In 48 hours, Google's co-scientist AI tool solved a complex problem relating to how superbugs become immune to antibiotics. This problem had taken microbiologists a decade to solve, but these microbiologists had not yet published their paper, so the AI had no access to their solution, yet arrived at this solution by its own reasoning, as well as access to public domain scientific knowledge.

It would be interesting to see this tool in action in the context of ME/CFS, generating hypotheses about causes and treatments for ME/CFS.

At present, the tool is only accessible to a small group of researchers participating in Google's Trusted Tester Program.


Article: AI solves superbug mystery in two days after scientists took 10 years

 

[bad1080]

Two days later, the AI made its own suggestions, which included what the Imperial scientists knew to be the right answer.

so it had a list of things but only because it was solved previously by humans they could tell the correct answer was in there. another one of those articles where the capabilities of the LLM are vastly overstated. nothing was "cracked" or "solved" by it.

 

[hapl808]

so it had a list of things but only because it was solved previously by humans they could tell the correct answer was in there. another one of those articles where the capabilities of the LLM are vastly overstated. nothing was "cracked" or "solved" by it.

Obviously new groundbreaking research would be more interesting (but also take a long time to validate). But not sure what you mean by 'in there' as supposedly:

these microbiologists had not yet published their paper, so the AI had no access to their solution, yet arrived at this solution by its own reasoning, as well as access to public domain scientific knowledge.

 

[Hip]

so it had a list of things but only because it was solved previously by humans they could tell the correct answer was in there. another one of those articles where the capabilities of the LLM are vastly overstated. nothing was "cracked" or "solved" by it.

My understanding is that Google's co-scientist AI tool (which does not yet have an official name) uses its knowledge of medical science to suggest new hypotheses and theories that may explain a biological or medical phenomenon.

The AI tool provides a list of hypotheses, which scientists can then consider and examine. In this case of superbug antibiotic resistance, the theory that these scientists came up with after 10 years of research was the first hypothesis on the list generated by the AI tool.

I think this AI tool might be useful for studying ME/CFS, because in the 20 years that I have had this disease, and have been following the science, there have been several hypotheses proposed by various researchers and doctors.

Every time a new hypothesis for ME/CFS is proposed, the ME/CFS patient community get quite excited, and start discussing online the science that lies behind the hypothesis, because they hope that the new theory might lead to some treatment options, or lead to a deeper understanding of this illness.

 

[bad1080]

The AI tool provides a list of hypotheses, which scientists can then consider and examine. In this case of superbug antibiotic resistance, the theory that these scientists came up with after 10 years of research was the first hypothesis on the list generated by the AI tool.

I think this AI tool might be useful for studying ME/CFS, because in the 20 years that I have had this disease, and have been following the science, there have been several hypotheses proposed by various researchers and doctors.

depending on the length of the list (and whether it is weighted in some form and not just a black box) it could be potentially useful or end in a wild goose chase. let's hope it's going to be the former.

But not sure what you mean by 'in there' as supposedly

i mean how would they know it had guessed correctly without the last 10 years of research?

 

[Wishful]

so it had a list of things but only because it was solved previously by humans they could tell the correct answer was in there.

No, it's theory could have been tested and turned into a treatment even without the other research group's findings. It didn't produce billions of theories that required a human to choose which one was "right".

I agree with Hip that this does have potential for ME.

 

[Hip]

I agree with Hip that this does have potential for ME.

I think it could be useful. Though this AI tool might be hampered by the lack of biomedical research into ME/CFS.

In the case of antibiotic resistance, I expect there is a lot of published fine-detailed biochemical research, and the AI tool just connected the dots. Whereas in ME/CFS, there may not be that many dots for the AI to work with.

 

[Norwegianlad]

i mean how would they know it had guessed correctly without the last 10 years of research?

Maybe they couldn't, but, if we see that AI tools like these are beating humans are predicting things or making theories, than at least it can save researchers tons of time just by researching what the AI tells them to research, and check if it's right.

It's a two way street of course, but these models are getting so insanely efficient, they will outsmart humans in so many ways and optimise everything. Unless AI turns itself on us, we will only benefit from it.

 

[Treeman]

think it could be useful. Though this AI tool might be hampered by the lack of biomedical research into ME/CFS.

I believe the same. One present hypothesis is immune dysfunction, but as the immune system still has so much to discover about it I don’t expect to see it unraveled in my life time