How we're using AI to drive scientific research with greater real-world benefit

AI has long driven scientific advances at Google, and the velocity at which breakthroughs are happening today is unprecedented. AI’s increasing capabilities have made the “magic cycle” of research from breakthrough to real-world impact broader and quicker than ever.

AI is an amplifier of human ingenuity. At Google, our teams are using AI to address fundamental scientific questions and push the boundaries of what’s possible, leading to new understanding of life and novel solutions to humanity’s greatest challenges. As we accelerate scientific discovery, we collaborate closely with the scientific community and ecosystem, across both industry and academia, and make our tech and tools available to our partners for their own research.

Here are four areas where Google Research has shared recent breakthroughs with meaningful scientific and societal impact.

Advancing biomedical science to better treat disease

We’re excited about the potential of AI to personalize medicine, democratize science and open up new avenues for biological and medical discovery. Our AI co-scientist aims to help expedite the process of making new biomedical discoveries. This multi-agent system uses AI’s ability to synthesize information and perform complex reasoning tasks to aid scientists in creating novel hypotheses and research proposals using natural language.

We recently introduced a multimodal version of AMIE, a research AI agent for medical diagnostic conversations published in Nature. It can intelligently interpret and reason about visual medical information, working towards more accurate diagnosis. AMIE builds upon our earlier work on MedPaLM and subsequent language models fine-tuned for the medical domain. We continue to add assets to Health AI Developer Foundations to help developers build medical AI applications and we released TxGemma, a collection of open models designed to improve the efficiency of therapeutic development, building on our embeddings models for chest x-rays, digital pathology and dermatology.

We keep advancing research into genomics, too, to better understand individuals' genetic predispositions to illnesses and to diagnose rare diseases. REGLE is an unsupervised deep learning model that helps researchers discover associations with genetic variants. And we open sourced new DeepVariant models as part of a collaboration on Personalized Pangenome References, which can reduce errors by 30% when analyzing genomes of diverse ancestries.

Advancing neuroscience research to demystify the brain

Over the past decade, we’ve made fundamental contributions to the field of connectomics and furthered scientific understanding of the workings of the brain. Just yesterday in Nature, a team from Google Research and ISTA published on LICONN, the first-ever method for using commonly available light microscopes to comprehensively map neurons and their connections in brain tissue. LICONN will enable more labs around the world to pursue connectomics studies.

Building beyond neural connections, in collaboration with HHMI Janelia and Harvard, we introduced the Zebrafish Activity Prediction Benchmark (ZAPBench) with recordings of more than 70,000 neurons from the entire brain of the larval zebrafish. This allows researchers to investigate the relationship between the structural wiring and dynamic neural activity across an entire vertebrate brain for the first time. We’ve open-sourced the dataset and benchmark to help neuroscientists develop more accurate brain activity models.

We’ve also explored the similarities and differences in how the human brain and deep language models process natural language through a series of studies in collaboration with Princeton University, NYU and HUJI. Our findings indicate that deep learning models could offer a new computational framework for understanding the brain's neural code.

Advancing geospatial science to tackle planet-scale challenges

Google Research is also accelerating geospatial problem solving by making critical information more accessible. We recently launched the first FireSat satellite to help address wildfires. As we expand the constellation to more than 50 satellites, the high-resolution imagery, updated globally every 20 minutes, will help emergency responders detect wildfires at an earlier stage and help scientists understand how wildfires spread. We’ve also driven climate resilience and crisis response with our advanced AI models for Flood Forecasting and WeatherNext models.

Geospatial Reasoning is a new research effort that seeks to combine the power of our geospatial foundation models with generative AI to uncover powerful and actionable information — all through a simple conversational interface. It builds on prior models including those for floods, wildfires and weather, as well as Open Buildings and SKAI models and expands our previous Population Dynamics and trajectory-based mobility foundational models. Geospatial Reasoning can be a critical tool for advancing public health, urban planning, integrated business planning, climate science and more.

Advancing quantum computing towards real-world applications

For more than a decade, we’ve been making progress toward building large-scale quantum computers that can solve otherwise impossible problems. Our new Willow chip is a major milestone, demonstrating error correction and state-of-the-art performance. On World Quantum Day, we highlighted how it’s bringing us closer to real-world applications. For example, in collaboration with Sandia National Laboratories, we demonstrated that a quantum algorithm could more efficiently simulate the mechanisms needed for sustained fusion reactions. This could help make fusion energy a reality, with its potential for clean energy at scale. We continue to advance our quantum research, and recently shared a novel hybrid approach to quantum simulation that paves the way for further scientific discoveries.

The promise of AI is now becoming a reality across scientific disciplines. We’ll continue to ask the biggest questions and address challenges that were previously unsolvable in pursuit of scientific breakthroughs that can benefit billions.