We are working on more seamless and intuitive ways to collaborate with artificial intelligence

The mouse pointer is a constant companion on computer screens, across websites, documents and workflows. Despite how technology has changed, the indicator has barely evolved in over half a century.

We’ve been exploring novel AI-powered capabilities to lend a hand the indicator not only understand what it’s pointing to, but also why it’s crucial to the user.

Our goal is to solve a common frustration: because a typical AI tool lives in its own window, users must drag their world into it. We want the opposite: intuitive AI that meets users in all the tools they operate, without disrupting their work. For example, imagine pointing to a picture of a building and asking for “Show me the way.” Nothing else is needed once the AI ​​system already understands the context.

Today, we present the fundamental principles that guide our thinking about future user interfaces and share experimental demonstrations of an AI-enabled indicator powered by Gemini. For example, you can visit Google AI Studio to edit image Or find places on the mapjust pointing and talking.

Thanks

The research project was led by Juraj Gottweis and Vivek Natarajan, as well as Alan Karthikesalingam, Annalisa Pawlosky and Yunhan Gokturk, Byron Lee, Dan Popovici, Dina Zverinski, Eeshit Dhaval Vaishnav, Elahe Vedadi, Fan Zhang, Felix Weissenberger, Florian Hasler, Frankie Garcia, Gary Peltz, Grzegorz Glowaty, Ivor Rendulic, Ivan Budiselic, Jacob Janson, Jacob Janberg, Jan. Jeremy Ratcliff, Joel Fenster, José R Penadés, Katherine Chou, Kavita Kulkarni, Keran Rong, Khaled Saab, Luka Rimanic, Marina Boia, Mathias Voges, Matthias Bellaiche, Nenad Tomašev, Ottavia Bertolli, Paige Kunkle, Petar Sirkovic, R Surocker, Tino Tuzy, Tao Tuzy RD Costa, Tom Sheffer, Victoria Langston, Vikram Dhillon, Yuan Guan, Ziyue Wang, Amin Vahdat, James Manyika, Demis Hassabis, Yossi Matias and Pushmeet Kohli.

Thank you to our teammates: Ali-Cowen Rivers, Anna Trostanetski, Barnaby James, Bill Byrne, Boon Panichprecha, Charlie Taylor, Diego Ballesteros, Hussein Hassan Harrirou, Ieva Grublyte, Ivan Lee, Jakob Oesignhaus, James Walker, Jorge Barrios, Laurynas Tamulevičius, Luka Važić, Meet Shah, Mihai Ciorobea, Natasha Latysheva, Nicolas Stroppa, Nir Kerem, Saz Basu, Sebastian Nowozin, Taylor Applebaum, Team Rakket, and Thomas Wagner and Yaniv Carmel for technical support.

We would also like to thank Carmela Sidrauski, Clare Bryant, Filippo Menolascina, Jonathan Gootenberg, Katherine Labbé, Matthew Onsum, Omar Abudayyeh, Ritu Raman, Ryan Flynn, Velia Siciliano for their cooperation.

Finally, we would like to thank Ali Eslami, Andy Berndt, Ankur Jain, Anna Koivuniemi, Clemens Mayer, Dale Webster, Greg Corrado, Jason Freidenfelds, Jeff Dean, Joelle Barral, John Jumper, John Platt, Josh Woodward, Karen DeSalvo, Koray Kavukcuoglu, Michael Brenner, Michael Horiol, Oriol, Noriol, Vizels, Vizels Parthasarathy Ranganathan, Ronit Levavi Morad, Royal Hansen, Scott Huffman, Srini Narayanan, Susan Thomas, Thomas Kurian, Zoubin Ghahramani, and Sundar Pichai for supporting this work.

3.5 Flash: Gigantic-Scale Agent Tasks

This balance of speed and performance makes 3.5 Flash ideal for solving long-term agent tasks. What once took developers days and auditors weeks, Flash 3.5 can now support you complete a project in a fraction of the time, often at less than half the cost of other pioneering models. Plans, builds, and iterates quickly to solve real-world problems, whether it’s building fresh applications, maintaining code bases, or helping prepare financial documents.

Combined with the updated anti-gravity beam, 3.5 Flash becomes a powerful engine for deploying collaborative subagents to solve large-scale problems in the most demanding employ cases. Under supervision, it can reliably complete multi-step workflows and coding tasks while maintaining pioneering efficiency.

Thanks

Thank you: Ferran Alet, Tom Andersson, Ilan Price, Stratis Markou, Andrew El-Kadi, Dominic Masters, Amy Li, Samier Merchant, Natalie Williams, Gregory Thornton, Ken MacKay, Olivia Graham, Ben Gaiarin, Elinor Kruse, Akib Uddin, Juanita Bawagan, Armin Senoner, Devaja Shah, Jacklynn Stott, Remi Lam, Aaron Bell, Paul Komarek, Matthew Willson, Alvaro Sanchez-Gonzalez and Peter Battaglia.

We also thank the teams from the National Hurricane Center, the British Meteorological Office and the Co-operative Institute for Atmospheric Research.

Finally, we thank Raia Hadsell, Zoubin Ghahramani, Yossi Matias, and Demis Hassabis for supporting this work.

Liver fibrosis is a scarring process that can lead to cirrhosis, which causes more than 1.4 million deaths per year. Geneticist Gary Peltz of Stanford University School of Medicine is using Co-Scientist to accelerate the search for drugs that can tardy, stop or reverse the process.

IN research published in Peltz’s team explored whether a co-scientist could support efforts to identify drugs from the extensive literature on existing drugs that could be repurposed to treat fibrosis. Peltz asked his co-scientist to propose three candidates and explain their reasoning. Sam also identified two drug candidates based on their notable presence in the liver fibrosis literature.

Peltz then subjected all five drugs to a fibrosis test in his lab, which consists of living human liver cells. His two drugs showed no benefit against fibrosis. In turn, of the three drug candidates selected by the co-scientist, two blocked fibrosis and promoted liver cell regeneration. Only a few articles linked one of these drugs to liver fibrosis, which represented a needle in a haystack of scientific literature.

The co-scientist’s outstanding choice was vorinostat, an anticancer drug. In Peltz’s experiments, it blocked 91% of the damage responses that can cause liver scarring. The co-scientist’s suggestions pointed to drugs that change the shape of gene activity, rather than a single fibrosis pathway. Peltz argues that such drugs deserve solemn consideration as treatments for liver fibrosis and could ultimately support bring a recent generation of antifibrotic drugs to market.

Ritu Raman of MIT and Ryan Flynn of Boston Children’s Hospital approach human biology with very different sets of tools, but a co-scientist brings their labs together. Raman, a mechanical engineer, builds living nerve and muscle tissue to model diseases that affect voluntary movement. Her husband Flynn, a chemical biologist, maps RNA on the surface of cells to see how it affects cell communication and how pathogens invade.

When Raman decided to research ALS, which was outside her usual domain, she was exposed to a extensive, conflicting literature that usually took months to understand. The co-scientist compressed this work, quickly helping Raman interrogate the evidence against her tissue model, transform ideas into testable hypotheses, and rank potential directions according to the trade-offs the labs face, such as feasibility and potential risk-reward.

But the co-scientist’s best leads had a catch: they concerned what was happening on the surface of cells, where much of their communication was mediated. Raman could manipulate tissues and measure results, but decoding the molecular interactions responsible for these signals was beyond her area of ​​expertise.

This gap became a catalyst for collaboration. Raman introduced Flynn to modern research directions, and the pair used the Co-Scientist program repeatedly, combining his best ideas into imaginative research paths that connected their different toolkits. To develop modern therapies, the search is currently underway for modern RNA-based mechanisms – and potentially RNA-based drugs – that could be used to combat ALS.

Biomedical research generates a flood of information that no scientist can realistically absorb. At the University of Edinburgh, bioengineer Filippo Menolascina uses Co-Scientist to search the literature for missed connections and generate novel hypotheses.

His team focused on a common liver disease called metabolic dysfunction-associated steatohepatitis (MASH). Therapeutic development is challenging because MASH involves interrelated biological processes, including liver inflammation and metabolism, which means that single-target drugs are not sufficient. This is pushing researchers toward combination therapies, but the number of potential drug pairs is overwhelming.

Faced with this combinatorial explosion, Menolascina used the Co-Scientist option to narrow the search area. In his hands, the co-scientist synthesized evidence from liver biology and pharmacology, highlighted mechanisms worth focusing on, and flagged potential combination therapies his team could test.

In one iconic case, a co-scientist tackled a timely, practical question: Why does the drug resmetirome, a recently approved drug prescribed for a specific stage of MASH, only aid a petite group of eligible patients? The system generated a hypothesis pointing to the NLRP3 inflammasome as a specific molecular bridge linking inflammation and metabolism in disease – a link that had never before been pieced together into a single actionable explanation. The hypothesis, later verified experimentally, could pave the way for targeted dual therapies.

Few problems in medicine are more intricate and more consequential than aging. Overlapping biological processes often determine how long people can live in good health. At Calico Life Sciences, head of AI/ML Matt Onsum and chief scientist Katherine Labbé apply Co-Scientist to connect scattered findings about the biology of aging and turn them into hypotheses worth testing.

This is not an basic task because the biological literature is full of discoveries of varying quality, dead ends, and results that cannot be replicated. Onsum says the co-scientist impressed Calico experts with his scientific insight, helping cut through the noise and helping them identify recent ideas that are truly worth exploring.

One example comes from Calico’s work on the integrated stress response (ISR), a protective cellular mechanism that, if constantly turned on, can also contribute to disease. Calico’s team used Co-Scientist to formulate a recent but plausible hypothesis regarding the regulation of ISR by metabolism, which is known to change with age and various diseases. The researchers also engaged with a co-scientist to refine the experimental design, test the hypothesis, and provide recent information as results became available. The experiments led to recent discoveries that have crucial implications for the role of ISR in health and disease, and the team plans to publish the results.

Most emerging infectious diseases are caused by pathogens that jump from animals to humans, such as Ebola, HIV, influenza and Covid-19. Professor Clare Bryant from the University of Cambridge is using her co-scientist position to look for the molecular switches that cause severe disease in humans, such as sepsis, when pathogens jump between species, and to find novel ways to prevent such phenomena.

Bryant, a testing associate, provided a summary of one of her grant applications to study influenza in birds and humans, outlining her lab’s research questions. The tool generated and classified a set of promising hypotheses—some she had already considered, others she had not. The unknown ones were the most thought-provoking.

When the grant was awarded, Bryant presented a full and detailed proposal. Later, while reading the text on the train to Brussels, she thought “aha!” moment: The co-scientist prioritized a protein that wasn’t on her radar, connected to several signaling pathways she was already interested in. She spent the rest of the week wanting to give him more data.

After returning to the lab, she added unpublished material, kept secret by Co-Scientist. With each iteration, the hypotheses sharpened, moving from candidate proteins to specific amino acids on which her lab could focus its experiments.

Bryant’s team is now building cell lines containing amino acid mutations to test their refined hypotheses. It would typically take two to three years of experimental work to get to the point of identifying the exact amino acids. However, her lab is now on track to complete the project within six months if working with a co-scientist leads them to the right goals.

Collaborating to apply pioneering AI to advance health and life sciences, reimagine classrooms and support the workforce of tomorrow.

At Google DeepMind, we believe that pioneering artificial intelligence can be a powerful force for good. Last year, we opened a fresh research lab in Singapore to expand the impact of our work across the Asia-Pacific region. Today we are taking a gigantic step forward. As part of our National Artificial Intelligence Partnerships initiative, we are launching fresh programs in the country while supporting a key pillar of Google’s fresh domestic AI partnership with the Singapore government.

Through our work with a wide range of organizations, we work together to support public sector transformation, support business growth and enable employees to thrive in an AI-powered world. The partnership is focused on tackling real-world challenges – including improving healthcare, accelerating scientific discovery and reimagining education – to ensure that these technological breakthroughs translate into meaningful progress in Singapore’s diverse communities. Together, we are strengthening Singapore’s national AI strategy to responsibly deploy AI at scale for economic growth and public benefit. Ultimately, by accelerating learning and innovation, AI can create an additional A$3.3 billion ($2.5 billion) in economic value through faster research and development by 2040.

Dealing with complicated social challenges

At the heart of this partnership is a shared commitment to accelerating local research and development that transforms the pioneering potential of AI into real, inclusive progress. Our team in Singapore works with local experts to responsibly apply pioneering AI solutions in areas where it can have the most significant impact on and with the community, starting with healthcare and life sciences.

• Extending your care with an AI co-clinic: We are exploring the possibility of partnering with public health clusters as part of our global co-clinician AI research initiative. This work examines how artificial intelligence can enhance and support physician knowledge to provide higher quality care. It also examines the evolution of healthcare towards “triadic care,” in which artificial intelligence agents support patients through all stages of care under the clinical supervision of a physician, using systems that provide information from clinical guidelines and scientific literature.
• Increasing pandemic preparedness: WITH Google.orgWe will support initiatives that advance infectious disease research and enhance pandemic preparedness. Using pioneering artificial intelligence like AlphaFold i Google Earth and other cutting-edge AI tools for science, we are working to better understand the epidemic in Southeast Asia. This is part Google.orga $7 million financial contribution to the Philanthropy Asia Alliance’s Health for Human Potential coalition.
• Innovations for inclusion: We are working on a Gemma-powered running assistant designed for blind and visually impaired athletes. Using spatial reasoning that provides real-time understanding of the environment, this tool helps athletes run independently, without physical lines or human guides. We are working with SG Enable – Singapore’s central disability and inclusion agency – to test and refine the product to meet the real needs of visually impaired runners.
• Accelerating scientific discovery: We are working with the National Research Foundation to train local researchers in agentic AI for scientific tools such as co-scientist hypothesis generation, which are already showing promise in a range of biomedical applications. We will also host workshops to lend a hand Singapore’s local science community leverage these pioneering tools to unlock fresh breakthroughs.

Improving education and building a future-ready workforce

As we transition to an AI-driven economy, our goal is to empower communities across Singapore to thrive. We work with the ecosystem to co-design programs that enable teachers and students to responsibly navigate fresh technologies.

We delivered Gemini for Education to all teachers, from primary to middle school, and will provide training in these pedagogically grounded AI tools. This will lend a hand teachers plan lessons and adapt training material, freeing up valuable time to focus on what’s vital: teaching. We will continue to work with the Ministry of Education to strengthen its capabilities in artificial intelligence in teaching and learning, including teacher training and upskilling programs.

Implementing innovations for sustainable development

Building a truly prosperous future requires a responsible and collaborative approach to global sustainable development. We are launching the Google DeepMind Accelerator: AI for the Planet program in the Asia-Pacific region to engage with the next generation of climate visionaries. This program aims to support startups, research teams and nonprofit organizations using pioneering artificial intelligence to unlock fresh approaches to complicated environmental challenges in areas such as energy, water, agriculture and more. Selected organizations will receive expert mentorship, tailored support and lend a hand integrate pioneering AI into their work, helping turn visionary ideas into significant progress.

Together we are pioneers of responsible artificial intelligence

We believe that for innovation to serve as a true force for good, it must be built on a shared foundation of responsibility. Google DeepMind is collaborating with the Infocomm Media Development Authority (IMDA) and MLCommons to study multimodal and multilingual security metrics. This collaboration focuses on supporting the safe and sound and responsible deployment of AI while respecting the nuances of local languages ​​and cultures, helping to ensure that the digital future we build is designed for everyone and with everyone.

Shaping a prosperous future for all

This partnership is a significant milestone in our journey with Singapore. By combining frontier research with local expertise, we can collectively address complicated societal challenges and create fresh opportunities for growth.

We look forward to working with researchers, doctors, educators and students to ensure that AI serves the diverse needs of communities across Singapore.