Lord Speaker, my lords, honourable members, distinguished guests, it's a huge honour to address you in the Palace of Westminster, a place steeped in history. Today I want to speak about a future that is not on a distant horizon, but is taking shape right now within the universities and research institutions across the UK and all around the world. A future where artificial intelligence is accelerating science and the discovery of new knowledge. We have become accustomed to think of AI as a chatbot. However, the case I want to make here today is that this view misses a bigger picture.
For centuries, the pursuit of knowledge was a painstaking endeavour, a testament to human curiosity and perseverance. From Newton to Crick, and from quantum physics to the meticulous work that unraveled the structure of the human genome, progress has been driven by human effort and ingenuity. Indeed, it was scientific greats like Darwin and Turing that motivated me to study computer science. In a 1960s paper, the Nobel Prize-winning physicist Eugene Wigner marvelled at the unreasonable effectiveness of mathematical equations for modelling important natural phenomena such as planetary motion. However, over the past half century, models that rely on sets of equations or other deterministic assumptions have struggled to capture the full complexity of systems in biology, economics, weather, and elsewhere. Why is this? This is because of the sheer number of interacting parts that make up these systems, as well as the potential for emergent behaviour. The challenges in modelling these systems impedes scientists' ability to predict or control how they will behave.
The brilliant minds of today must also contend with the burden of past knowledge, an ever-expanding library of past discoveries that they must master before they can push the frontier forward. To stand on the shoulders of giants, you must first climb a ladder, and a ladder which is getting longer by the day. The systems we seek to understand, from the intricacies of a single cell to the dynamics of planetary climate, are almost infinitely complex. It's like asking an explorer to navigate the entire world with one hand-drawn map and a compass. In short, we are running against the limits of what the human mind can view, analyse, and understand.
This is the transformative role AI is beginning to play in scientific discovery, and one which in time we may come to see not only as an optional extra but as a universal scientific instrument comparable to the microscope. Not as a replacement for human intellect, but as an extraordinarily powerful assistant, an intelligent tool that can navigate vast oceans of information, identify subtle patterns invisible to the human eye, and accelerate the journey to discovery. And this is the work I'm lucky enough to lead at Google DeepMind.
DeepMind's story is a British one. We began in London, founded by Sir Demis Hassabis with a belief that solving the riddle of intelligence and applying it to science was the most worthwhile thing anyone can work on. Our journey started by teaching AI to master games. It was back in 2016, during the historic match against the world champion Lee Sedol, watched by over 200 million people worldwide, that our system AlphaGo made the now famous move 37. A move so surprising that human commentators at the game at first thought it was a mistake, but which ultimately revealed that AI could not just imitate human intuition but play an ancient game of Go with creativity. From those early games we have grown, but our roots have remained firmly here in London. We now apply the same core AI technology to some of the greatest scientific challenges of our time, and we are proud to contribute to the UK's world-leading innovation ecosystem.
Let me begin with a story that encapsulates this revolution. A story that began with one of the greatest challenges in biology: the protein folding problem. Proteins are the building blocks of life, crucial for its functioning and also a cause of disease. I'm sure you will all remember the coronavirus spike protein during the pandemic. Proteins are formed of a sequence of characters that represent the different amino acids. Their function is dictated by their intricate 3D structure, and understanding the shape of even a single protein is something which has historically taken a full PhD's worth of work for nearly 50 years. Protein folding, finding the 3D structure of a protein from an amino acid sequence, remained one of biology's great scientific challenges.
Since it's the 3D structure that actually carries the function, it was like we could read the sequence, we could read the letters, but not understand the word. So to accelerate the search for a solution, the scientific community established a critical assessment for structure prediction, in short CASP, a biannual competition like an Olympics. But this was not just an academic exercise. It was a global head-to-head challenge to see if anyone could crack the protein code. The very design of this challenge, its biannual rhythm, its blind assessment, its singular focus on a grand challenge, provided the perfect environment for AI research.
In just one year, our team at DeepMind developed a system called AlphaFold. By training the model on structures from the Protein Data Bank, we built a model which was able to successfully predict the shape of over 200 million proteins, virtually every one known to man, and solved a 50-year grand challenge. This was not just a win for our scientists. It was a landmark moment for science, recognised last year with the Nobel Prize in Chemistry for our founder Demis Hassabis and one of my teammates, John Jumper. I often joke that I must be the only person in the world who has a boss and a member of my own team who are both Nobel laureates.
The true impact, however, lies not in the accolades but in the empowerment of scientists worldwide. I often say that protein folding is a root node problem on the tree of knowledge, problems which, if solved, unlock new branches and avenues of research, and this is precisely what we have seen. AlphaFold made over 200 million protein structure predictions freely available to the research community. Over 2.5 million researchers in over 190 countries, including more than 170,000 based here in the UK, have accessed AlphaFold predictions. If this is not a testament to the progress that science and humanity as a civilisation has made, I don't know what is. That 170,000 people in the UK are studying proteins, the building blocks of life, things which I could have counted the people who were studying that on my fingertips a few decades ago.
This is enabling research that could have big impacts, such as new vaccines for diseases like malaria, enzymes that can break down plastic pollution, and fundamental insights into conditions like cardiovascular diseases that affect millions. And this scientific renaissance doesn't stop with proteins. The ability to apply the power of machine learning to data-rich, complex problems holds huge promise for some of the biggest challenges that governments and scientists across the world are seeking to tackle. In genomics, AI models like AlphaMissense are helping us to understand the impact of tiny variations in our DNA, the misspellings in our genetic code, classifying which of the millions of possible human genetic variants are likely to be harmful. This is crucial for diagnosing and one day potentially treating genetic diseases and common conditions like cancer.
Now let's think about materials. Materials have in some sense defined human civilisation. We describe the different eras of human civilisation from the Stone Age to the Iron Age to the Bronze Age, and depending on who you ask, maybe we are living through the Silicon Age or the Plastic Age. They are so important. Discovering new materials with specific properties has traditionally been a slow and painstaking process of trial and error. Our AI tool GNoME, which stands for a Graph Networks for Materials Exploration, acts like an AI prospector sifting through a universe of possibilities. What it has done is expanded the set of known inorganic materials, which was roughly 30,000, to 2.2 million potential possibilities. And these 2.2 million new crystal structures include hundreds of thousands of stable materials that could form the basis of new-generation batteries, more efficient solar panels, or even new catalysts to clean our environment.
The economic and societal potential for the UK here is significant, and in the critical quest for clean, limitless energy, AI is playing a vital role. Nuclear fusion, the process that powers the sun, holds incredible promise. But containing plasma hotter than the sun is an immense engineering challenge. In collaboration with EPFL in Switzerland, our AI model has learned to sculpt and confine these ultra-hot plasmas within fusion reactors, a significant step towards making fusion a reality, an area where the UK already possesses world-leading expertise.
We are also creating all-purpose AI tools to help scientists everywhere. A great example is Co-Scientist. You can think of it as an AI research assistant that can draw upon vast scientific knowledge to propose new ideas. The clever part is that it uses a team of internal AIs that debate these ideas with each other to find the most promising ones, which we hope will help human scientists to find breakthroughs much faster than was possible before. Even the abstract worlds of mathematics and computer science are being transformed. Our new AI model, AlphaEvolve, is an AI-powered coding agent that rapidly discovers complex algorithms. Using evolutionary programming, it generates, tests, and refines thousands of solutions to improve existing algorithms and discover new ones. AlphaEvolve has enhanced data centre efficiency, aided computer chip design, and accelerated AI model training.
These specific breakthroughs, remarkable as they are, point to a more profound transformation. AI is becoming a universal instrument for science, augmenting human capabilities at every stage of the discovery process. It can help us manage knowledge. For instance, by using systems like Gemini to read, synthesize, and curate information from hundreds of thousands of scientific papers in a day, a task that would take human teams years. It can help us generate and understand scientific data on an unprecedented scale, creating vast, high-quality data sets that become the bedrock of new discoveries. It can turbocharge experiments, allowing us to simulate complex phenomena, guide the design of physical experiments to be more targeted and efficient, and even automate aspects of our lab work.
And it can help us find solutions in vast search spaces, whether it's designing drug molecules from a near-infinite number of possibilities or identifying solutions to matrix multiplication problems, a fundamental problem in computer science and mathematics.
What does this revolution mean for the United Kingdom? It means opportunity. An opportunity to drive economic growth, create high-value jobs, and tackle some of our most pressing national challenges. From improving healthcare outcomes in our NHS to achieving net-zero ambitions, the future scientific and economic success of this nation will increasingly depend on our ability to fully harness AI in the pursuit of discovery. Building on the enormous strengths the UK already has today, we have the talent. We have the heritage. Now we must double down on the potential of this technology.
To fully capitalise on this moment and to usher in what we believe can be the new golden age of discovery requires a concerted national effort. A partnership between government, world-class academic institutions, and industry. First, we must sharpen our focus by identifying the grand scientific challenges where AI can make the most profound difference for the UK and for humanity. Think of these as rallying points for talent and investment. Perhaps even new global competitions to spur innovation. The protein folding competition which I mentioned earlier was so successful in catalysing AlphaFold because it brought together the ingredients necessary to make it work. It was a worthwhile challenge with a high-quality data set, with a massive search space in which to experiment, and a clear, measurable function to optimise for. When we think about designing our own grand challenges for AI, we should look to replicate what the protein folding competition was able to achieve.
Second, we must ensure our scientists have the essential fuel for these powerful AI engines: large, high-quality, and accessible scientific data sets. This means strategic investments in curating existing data sets and creating new ones, making the scientific world truly readable by our AI systems and researchers. Third, we must equip our brilliant researchers with AI tools and teach AI as the next universal scientific instrument. AI should become as fundamental to a scientist's toolkit as statistics or microscopy. This means embedding AI training into science education at all levels and supporting upskilling for current researchers.
And finally, we must be bold and experiment with how we organise and fund science. We need to foster environments where interdisciplinary collaboration thrives, where high-risk, high-reward research is encouraged, and where breakthroughs can be rapidly translated into real-world impact. This involves learning from and experimenting with new models of research institutions and funding.
Lord Speaker, my lords, honourable members, the United Kingdom has a legacy of discovery that is second to none. We are a country that has consistently pushed the boundaries of human knowledge. With AI, we have the opportunity not just to continue that legacy, but to amplify it beyond what we might have imagined a few years ago. The path to this new age of discovery requires vision, investment, and collaboration. It requires us to embrace the transformative potential of AI with ambition and with responsibility. Let us work together to ensure that this future, a future of accelerated scientific progress and enhanced human well-being, is a future forged here in the United Kingdom. Thank you.
