9 ways AI is advancing science

Last updated: November 22, 2024

We're living in an era where applied science, human ingenuity, and new technologies are shedding light on some of humanity's biggest and oldest questions. While we often think of scientific progress as rapid and unstoppable, the truth is, for many decades, progress has actually slowed down. The scientific community is still debating the reasons behind this slowdown, but much of today's technology—from jets to manufacturing processes—hasn't changed much in the last half-century.

However, in recent years, breakthroughs in fields like artificial intelligence and quantum computing have really picked up the pace of scientific discovery. From healthcare advancements to discovering enzymes that eat plastic, we're already seeing the benefits.

These breakthroughs are the result of decades of collaboration between researchers, technologists, policymakers, civil organizations, and folks from all walks of life. They provide a roadmap for how AI can be used in science to significantly improve our lives.

With this in mind, The Royal Society, in partnership with Google DeepMind, is hosting the first AI for Science Forum today. This event in London brings together scientists, policymakers, and industry leaders to explore how AI can transform science and the role of public-private partnerships in driving innovation.

To understand how we've reached this point and where we might head next, let's take a look at nine recent milestones that have paved the way for future scientific progress:

1. Solving the 50-year "grand challenge" of protein structure prediction

For decades, experts have called understanding protein folding a "grand challenge." In 2022, Google DeepMind released the predicted structures of 200 million proteins using their AlphaFold 2 model. Before this, figuring out the 3D structure of a single protein could take over a year—AlphaFold can do it in minutes with incredible accuracy. By making these predictions freely available, scientists worldwide can now speed up research in areas like new drug development, fighting antibiotic resistance, and tackling plastic pollution. The next step, AlphaFold 3, builds on this by predicting the structure and interactions of all life's molecules.

2. Revealing the human brain in unprecedented detail to support health research

The human brain has always been a mystery. After 10 years of connectomics research, Google, along with the Lichtman Lab at Harvard and others, mapped a tiny piece of the human brain in more detail than ever before. Released in 2024, this project showed structures in the brain that we'd never seen before. The full dataset, complete with AI-generated annotations for each cell, is now public, helping to accelerate research.

3. Saving lives with accurate flood forecasting

When Google started its flood forecasting project in 2018, many thought it was impossible to predict floods accurately on a large scale due to limited data. But researchers developed an AI model that can predict extreme riverine events in ungauged watersheds up to five days in advance, with reliability matching or even surpassing that of nowcasts. By 2024, Google Research expanded this to cover 100 countries and 700 million people worldwide, and improved the model to offer the same accuracy at a seven-day lead time as the previous model did at five.

4. Spotting wildfires earlier to help firefighters stop them faster

Wildfires are becoming more common and destructive due to hotter and drier climates. In 2024, Google Research teamed up with the U.S. Forest Service to create FireSat, an AI model and new global satellite constellation designed to detect and track wildfires as small as a classroom. With higher-resolution imagery available within 20 minutes, this will help fire authorities respond more quickly, potentially saving lives, property, and natural resources.

5. Predicting weather faster and with more accuracy

In 2023, Google DeepMind launched GraphCast, a machine learning model that predicts weather up to 10 days ahead more accurately and much faster than the industry standard (HRES). GraphCast can also predict cyclone paths and associated risks like flooding more accurately, and it correctly predicted Hurricane Lee would hit Nova Scotia three days before traditional models did.

6. Advancing the frontier of mathematical reasoning

AI has always struggled with complex math due to limited data and reasoning skills. But in 2024, Google DeepMind introduced AlphaGeometry, an AI system that solved complex geometry problems at a level close to that of a human Olympiad gold-medalist. This was a big step forward in AI performance and the development of more advanced general AI systems. The follow-up model, AlphaGeometry 2, combined with AlphaProof, solved 83% of all historical International Mathematical Olympiad (IMO) geometry problems from the past 25 years. This shows AI's growing ability to reason and potentially solve problems beyond current human capabilities, bringing us closer to systems that can discover and verify new knowledge.

7. Using quantum computing to accurately predict chemical reactivity and kinetics

Google researchers, along with UC Berkeley and Columbia University, conducted the largest chemistry simulations ever on a quantum computer. Published in 2022, these results were not only competitive with classical methods but also didn't require the usual error mitigation associated with quantum computing. This ability to run these simulations will lead to more accurate predictions of chemical reactivity and kinetics, paving the way for new applications of chemistry to solve real-world problems.

8. Accelerating materials science and the potential for more sustainable solar cells, batteries, and superconductors

In 2023, Google DeepMind announced Graph Networks for Materials Exploration (GNoME), an AI tool that has already discovered 380,000 materials stable at low temperatures, according to simulations. At a time when we're looking for new energy solutions, processing power, and materials science advancements, this could lead to better solar cells, batteries, and potential superconductors. To make this technology accessible to everyone, Google DeepMind made GNoME's most stable predictions available through the Materials Project on their open database.

9. Taking a meaningful step toward nuclear fusion—and abundant clean energy

As the old saying goes, "Fusion is the energy of the future—and it always will be." Harnessing the energy that powers stars, including our sun, has been a long-standing challenge. In 2022, Google DeepMind developed AI that can autonomously control the plasma inside a nuclear fusion reactor. Working with the Swiss Plasma Center at EPFL, they built the first Reinforcement Learned system capable of stabilizing and shaping the plasma within an operational fusion reactor, opening new pathways toward stable fusion and abundant clean energy for everyone.