Google DeepMind CEO says 2025's the year we start popping pills AI helped invent
(2025/01/22)
- Reference: 1737530886
- News link: https://www.theregister.co.uk/2025/01/22/google_deepmind_ai_drugs/
- Source link:
Clinical trials of the first drugs designed with the help of artificial intelligence could commence this year, Google DeepMind CEO Demis Hassabis suggested Tuesday.
Speaking on a panel at the World Economic Forum in Davos, Hassabis, who also runs DeepMind drug-discovery spin-off Isomorphic Labs, said he [1]expected to have "some AI-designed drugs in clinical trials by the end of the year… That's the plan."
Isomorphic Labs has tried to speed up the development of medicines using machine learning since 2021. "Eventually you could imagine personalized medicine where it's optimized, maybe overnight, by an AI system for your personal metabolism," he said.
[2]
AI hype is currently omnipresent, though Hassabis and his colleague John Jumper earned a [3]Nobel Prize for work AlphaFold, a deep learning system that can predict protein structures.
[4]
[5]
Pharmaceutical companies are interested in AI because it has the potential to save them lots of time and money. According to a recent article [6]published in the Journal Nature Medicine, successfully creating a new drug and having it approved for use can take 12 to 15 years and costs roughly $2.6 billion.
Many drugs are never approved for use, as fewer than ten percent of clinical trials in which humans consume the drug succeed. Anything that can reduce costs, speed development, or increase success rates will make a material impact on pharma companies’ bottom lines.
[7]
Researchers believe there are many ways in which machine learning models can improve and speed parts of the drug discovery process. Hassabis believes huge savings in time and cost could be possible.
Optimism of that sort needs to be tempered because high-quality training data is hard to come by, due to privacy regulations, data-sharing policies, and acquisition costs.
Hassabis believes those challenges aren't insurmountable. "You can generate some key data to fill in the gaps of where the public data doesn't have it," he said.
[8]
This can be done in collaboration with clinical research organizations or through the use of synthetic data, something he said AlphaFold2 used extensively. However, as we've previously [9]discussed , synthetic data can be problematic.
"You've got to be very careful if you're using synthetic data, that it's actually correctly representing the distribution and you're not somehow training on your own errors," Hassabis said.
[10]Germany unleashes AMD-powered Hunter supercomputer
[11]Nvidia continues its quest to shoehorn AI into everything, including HPC
[12]DoD spins up supercomputer to accelerate biothreat defense
[13]Google DeepMind touts AI model for 'better' global weather forecasting
Hassabis doesn't think AI will replace scientists anytime soon.
"True invention is not possible yet with AI. It can't come up with a new hypothesis or new conjecture. It can maybe solve a complicated conjecture in, say, maths. I think we're very close to some big breakthroughs on that front. I think we'll actually see that this year, but that's different from actually coming up with the theory or the hypothesis, as the best human scientists do," he said.
Hassabis is not alone in exploring the application of machine learning on drug discovery. Nvidia has also shown enthusiasm for AI-augmented drug discovery, perhaps because it will create more reasons to buy its hardware.
Last northern autumn, Nvidia [14]open sourced its BioNeMo family of GPU-accelerated machine learning frameworks for drug development and molecular design. The company has also taken steps to repackage existing models like DeepMind's AlphaFold2 and MIT's DiffDock 2.0 as microservices to make them easier to consume.
Nvidia is also [15]partnering with major pharma companies, including Danish pharmaceutical giant Novo Nordisk, to bring new research systems online. Denmark's Gefion supercomputer, which applies apply machine learning to biological sciences and the development of new treatments, is one example of such efforts. ®
Get our [16]Tech Resources
[1] https://es.weforum.org/meetings/world-economic-forum-annual-meeting-2025/sessions/folding-science/
[2] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_offbeat/science&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=2&c=2Z5DP0lT_NBH7OIo9fHtP-AAAAcY&t=ct%3Dns%26unitnum%3D2%26raptor%3Dcondor%26pos%3Dtop%26test%3D0
[3] https://www.theregister.com/2024/10/09/alphafold_rosetta_nobel_chemistry_prize/
[4] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_offbeat/science&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=4&c=44Z5DP0lT_NBH7OIo9fHtP-AAAAcY&t=ct%3Dns%26unitnum%3D4%26raptor%3Dfalcon%26pos%3Dmid%26test%3D0
[5] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_offbeat/science&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=3&c=33Z5DP0lT_NBH7OIo9fHtP-AAAAcY&t=ct%3Dns%26unitnum%3D3%26raptor%3Deagle%26pos%3Dmid%26test%3D0
[6] https://www.nature.com/articles/s41591-024-03434-4
[7] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_offbeat/science&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=4&c=44Z5DP0lT_NBH7OIo9fHtP-AAAAcY&t=ct%3Dns%26unitnum%3D4%26raptor%3Dfalcon%26pos%3Dmid%26test%3D0
[8] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_offbeat/science&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=3&c=33Z5DP0lT_NBH7OIo9fHtP-AAAAcY&t=ct%3Dns%26unitnum%3D3%26raptor%3Deagle%26pos%3Dmid%26test%3D0
[9] https://www.theregister.com/2024/05/09/ai_model_collapse/
[10] https://www.theregister.com/2025/01/17/hlrs_supercomputer_hunter/
[11] https://www.theregister.com/2024/11/18/nvidia_ai_hpc/
[12] https://www.theregister.com/2024/08/16/dod_drug_discovery_supercomputer/
[13] https://www.theregister.com/2024/12/05/google_deepmind_weather_model/
[14] https://nvidianews.nvidia.com/news/nvidia-opens-bionemo-to-scale-digital-biology-for-global-biopharma-and-scientific-industry
[15] https://www.theregister.com/2024/10/24/nvidia_gpus_europe/
[16] https://whitepapers.theregister.com/
Speaking on a panel at the World Economic Forum in Davos, Hassabis, who also runs DeepMind drug-discovery spin-off Isomorphic Labs, said he [1]expected to have "some AI-designed drugs in clinical trials by the end of the year… That's the plan."
Isomorphic Labs has tried to speed up the development of medicines using machine learning since 2021. "Eventually you could imagine personalized medicine where it's optimized, maybe overnight, by an AI system for your personal metabolism," he said.
[2]
AI hype is currently omnipresent, though Hassabis and his colleague John Jumper earned a [3]Nobel Prize for work AlphaFold, a deep learning system that can predict protein structures.
[4]
[5]
Pharmaceutical companies are interested in AI because it has the potential to save them lots of time and money. According to a recent article [6]published in the Journal Nature Medicine, successfully creating a new drug and having it approved for use can take 12 to 15 years and costs roughly $2.6 billion.
Many drugs are never approved for use, as fewer than ten percent of clinical trials in which humans consume the drug succeed. Anything that can reduce costs, speed development, or increase success rates will make a material impact on pharma companies’ bottom lines.
[7]
Researchers believe there are many ways in which machine learning models can improve and speed parts of the drug discovery process. Hassabis believes huge savings in time and cost could be possible.
Optimism of that sort needs to be tempered because high-quality training data is hard to come by, due to privacy regulations, data-sharing policies, and acquisition costs.
Hassabis believes those challenges aren't insurmountable. "You can generate some key data to fill in the gaps of where the public data doesn't have it," he said.
[8]
This can be done in collaboration with clinical research organizations or through the use of synthetic data, something he said AlphaFold2 used extensively. However, as we've previously [9]discussed , synthetic data can be problematic.
"You've got to be very careful if you're using synthetic data, that it's actually correctly representing the distribution and you're not somehow training on your own errors," Hassabis said.
[10]Germany unleashes AMD-powered Hunter supercomputer
[11]Nvidia continues its quest to shoehorn AI into everything, including HPC
[12]DoD spins up supercomputer to accelerate biothreat defense
[13]Google DeepMind touts AI model for 'better' global weather forecasting
Hassabis doesn't think AI will replace scientists anytime soon.
"True invention is not possible yet with AI. It can't come up with a new hypothesis or new conjecture. It can maybe solve a complicated conjecture in, say, maths. I think we're very close to some big breakthroughs on that front. I think we'll actually see that this year, but that's different from actually coming up with the theory or the hypothesis, as the best human scientists do," he said.
Hassabis is not alone in exploring the application of machine learning on drug discovery. Nvidia has also shown enthusiasm for AI-augmented drug discovery, perhaps because it will create more reasons to buy its hardware.
Last northern autumn, Nvidia [14]open sourced its BioNeMo family of GPU-accelerated machine learning frameworks for drug development and molecular design. The company has also taken steps to repackage existing models like DeepMind's AlphaFold2 and MIT's DiffDock 2.0 as microservices to make them easier to consume.
Nvidia is also [15]partnering with major pharma companies, including Danish pharmaceutical giant Novo Nordisk, to bring new research systems online. Denmark's Gefion supercomputer, which applies apply machine learning to biological sciences and the development of new treatments, is one example of such efforts. ®
Get our [16]Tech Resources
[1] https://es.weforum.org/meetings/world-economic-forum-annual-meeting-2025/sessions/folding-science/
[2] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_offbeat/science&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=2&c=2Z5DP0lT_NBH7OIo9fHtP-AAAAcY&t=ct%3Dns%26unitnum%3D2%26raptor%3Dcondor%26pos%3Dtop%26test%3D0
[3] https://www.theregister.com/2024/10/09/alphafold_rosetta_nobel_chemistry_prize/
[4] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_offbeat/science&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=4&c=44Z5DP0lT_NBH7OIo9fHtP-AAAAcY&t=ct%3Dns%26unitnum%3D4%26raptor%3Dfalcon%26pos%3Dmid%26test%3D0
[5] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_offbeat/science&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=3&c=33Z5DP0lT_NBH7OIo9fHtP-AAAAcY&t=ct%3Dns%26unitnum%3D3%26raptor%3Deagle%26pos%3Dmid%26test%3D0
[6] https://www.nature.com/articles/s41591-024-03434-4
[7] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_offbeat/science&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=4&c=44Z5DP0lT_NBH7OIo9fHtP-AAAAcY&t=ct%3Dns%26unitnum%3D4%26raptor%3Dfalcon%26pos%3Dmid%26test%3D0
[8] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_offbeat/science&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=3&c=33Z5DP0lT_NBH7OIo9fHtP-AAAAcY&t=ct%3Dns%26unitnum%3D3%26raptor%3Deagle%26pos%3Dmid%26test%3D0
[9] https://www.theregister.com/2024/05/09/ai_model_collapse/
[10] https://www.theregister.com/2025/01/17/hlrs_supercomputer_hunter/
[11] https://www.theregister.com/2024/11/18/nvidia_ai_hpc/
[12] https://www.theregister.com/2024/08/16/dod_drug_discovery_supercomputer/
[13] https://www.theregister.com/2024/12/05/google_deepmind_weather_model/
[14] https://nvidianews.nvidia.com/news/nvidia-opens-bionemo-to-scale-digital-biology-for-global-biopharma-and-scientific-industry
[15] https://www.theregister.com/2024/10/24/nvidia_gpus_europe/
[16] https://whitepapers.theregister.com/
Hurrah
Anonymous Coward
It's going to be one huge virtuous AI circle shirk.
At one end destroying the planet with its unquenchable thirst for energy and at the other keeping us all living forever so we can watch the real big bang - when the planet explodes because we've fucked it so hard.
Should be some spectacle.
Re: Hurrah
abend0c4
keeping us all living forever
We'll only be kept alive as long as we continue to pay for the drugs. It's the ultimate extension of rentier capitalism.
Re: Hurrah
Charlie Clark
You shouldn't conflate things like DeepMind with LLMs like ChatGPT.
DeepMind has already demonstrated its ability to find candidate molecules much faster than would ever be possible in the lab and elsewhere similar approaches have, for years, been used in pre-lab simulations based on our existing understanding, as much of this really is extremely advanced statistical modelling.
It's not like we didn't choose the data and give the directives