Editor’s note: Today in London, Google DeepMind and the Royal Society co-hosted the first AI for Science Forum, which brought together Nobel laureates, the scientific community, policy makers and industry leaders to explore the transformative potential of AI to drive scientific breakthroughs. address the world’s most pressing challenges and lead to a new era of discovery.
James Manyika, Google’s senior vice president for search, technology and society, delivered the keynote address; The following is a transcription of his remarks, as prepared for delivery.
The impact of AI on science has been in the headlines recently, but AI’s potential to advance science has long been a motivating force for many in the field, dating back to early AI researchers , such as Alan Turing and Christopher Longuet-Higgins, and to many over the past decades, including my colleagues at Google DeepMind and Google Research.
The enthusiasm around AI and science is not due to the belief that AI can replace scientists, but to the fact that many puzzling scientific problems benefit from the use of computational techniques, thus making AI a powerful tool to help scientists.
We saw the first signs of this assistive potential with Hodgkin and Huxley’s use of computational approaches to describe how nerve impulses travel along neurons, work that earned them the Nobel Prize in 1963.
Fast forward to my colleagues Demis Hassabis, John Jumper and the AlphaFold team whose work using AI recently won the Nobel Prize in Chemistry, solving the “protein folding problem” posed by Nobel laureate Christian Anfinsen in the 1970s.
So how does AI help advance science?
I’ll start with speed. In some scientific fields, increasingly powerful AI allows us to condense hundreds, even thousands, of years of research into a few years, months, or even days.
AI also helps broaden the scope of research, allowing scientists to examine many things at once – and in new ways – rather than one at a time.
Advances in AI, as well as access to the knowledge gained from its use, enable more people to participate in research, allowing us to accelerate scientific discovery even further.
AI enables historic advances in several scientific disciplines
Let me briefly share a few examples of how AI is enabling historic advances, starting with AlphaFold:
With AlphaFold, in one year, my colleagues were able to predict the structure of almost all proteins known to science, more than 200 million of them. And with Alphafold 3, they have expanded beyond proteins to all biomolecules of life, including DNA, RNA, and ligands.
To date, AlphaFold has been used by more than 2 million researchers in more than 190 countries, working on problems ranging from neglected diseases to drug-resistant bacteria.
AlphaMissense, which builds on AlphaFold, allowed my colleagues to classify nearly 90% of the 71 million possible missense variants (single letter substitutions in DNA) as likely pathogenic or likely benign. In contrast, only 0.1% have been confirmed by human experts, albeit in more detail.
When the human genome was initially sequenced – an incredible achievement – it was based on a single genome assembly.
Last year, my colleagues at Google Research, using AI tools and working with a consortium of academic collaborators, published the first draft of a reference human pangenome.
This was based on 47 genome assemblies, thus better representing human genetic diversity.
In neuroscience, a 10-year collaboration between my colleagues at Google Research, the Max Planck Institute and Harvard’s Lichtman Lab recently produced nanoscale mapping of part of the human brain – a level of detail never before seen. previously achieved. reached.
This project revealed previously unseen structures in the human brain that could change our understanding of how the human brain works. Perhaps this will lead us to new approaches to understanding and combating neurological diseases like Alzheimer’s and others. The complete mapping has been made public so that researchers can draw inspiration from it.
Beyond life sciences, we are seeing progress in other areas.
In a landmark achievement in climate modeling, we combined machine learning with a traditional physics-based approach to create NeuralGCM.
This allows us to simulate the atmosphere more accurately and efficiently: NeuralGCM can simulate over 70,000 days of the atmosphere in the time it would take a state-of-the-art physics-based model to simulate just 19 days .
There are other similar advances, such as the work of my colleagues at Google DeepMind on GraphCast, a cutting-edge AI model that predicts weather conditions up to 10 days in advance more accurately and much faster than the industry’s benchmark weather forecast. simulation system.
Our Quantum AI team is making progress on questions that were previously science fiction, such as studying the characteristics of traversable wormholes.
This opens new possibilities for testing theories of quantum gravity first posited with the Einstein-Rosen bridge almost ninety years ago.
In fact, the quantum domain is an area where we are starting to see promising bidirectional reinforcement between AI and science.
In one sense, AI advances our progress in quantum computing; in the other, quantum helps advance research in the field of AI.
There are many other examples we are working on in materials science, fusion, mathematics and more, all in collaboration with many academic scientists.
Scientific advances made possible by AI have a real impact
Beyond these advances, AI is also advancing science in ways that are already bringing tangible benefits to real people in areas like climate and healthcare.
Let me start with an example related to climate adaptation. Flood forecasting is a more frequent and urgent problem due to climate change. Today, advances in AI have allowed us to fill important data gaps to predict river flooding up to 7 days in advance with the same accuracy as nowcasts. After a first pilot project in Bangladesh, our early warning platform – Flood Hub – now covers more than 100 countries and 700 million people.
And for an example of climate change mitigation, consider the following: Contrail formation has long been a known contributor to aviation emissions, accounting for up to 35% of aviation’s impact on global warming.
My colleagues at Google Research developed an AI model that predicts where contrails are likely to form and, in partnership with American Airlines, tested it on 70 flights. We measured the impact and saw a 54% reduction in emissions.
Similarly, AI holds great promise for disease detection. For example, eight years ago, Google researchers discovered that AI could help accurately interpret retinal scans to detect diabetic retinopathy, a preventable cause of blindness that affects about 100 million people.
We developed a screening tool that has been used in over 600,000 screenings worldwide. And new partnerships in Thailand and India will enable 6 million screenings over the next decade.
The road ahead
Other examples we have implemented include tuberculosis, colorectal cancer, breast cancer and maternal health.
Despite the progress, this is only the beginning. There is still so much to do.
I see three key areas to focus on to fully realize the potential of AI to advance science and deliver tangible societal benefits:
First, we must continue to make progress on the current limitations and shortcomings of AI – and increase the capabilities of AI to be able to contribute to the development of new scientific concepts, theories, experiments and much more.
Second, we need a lasting commitment to the scientific method and responsible approaches to using AI to advance science.
We need scientists, ethicists, and security experts – like many in this room – working together to address the most science-specific risks, like viruses and biological weapons, and the challenges such as biases in datasets, privacy preservation and environmental impacts.
Third, we must prioritize making AI-based research, tools and resources more accessible to more scientists in more places – and ensure that the progress we make benefit people around the world.
I’m excited about what lies ahead in this new era of discovery.
There is so much we can do together to create tools that will help advance science for the benefit of all.
And there’s so much we can do to enable extraordinary scientists here and abroad to continue their work – we’ll hear from some of them today.