AI Will Accelerate Human Innovation, and Our Goal Is to Cure All Diseases
Prof. Yossi Matias (68) Residence: Mountain View Marital status: Married, 3 children Education: Bachelor’s degree in computer science from Tel Aviv University, master’s degree in computer science from the Weizmann Institute of Science, and a doctorate from Tel Aviv University Position: Vice President at Google and head of Google Research; faculty member in computer science at Tel Aviv University (on leave) Something else: Was a pilot in the Air Force
Prof. Yossi Matias, vice president at Google and head of Google Research, we begin our conversation as you are on your way to work at Waymo. Out of curiosity, if you take the wheel, can you make it deviate from its route?
“No, there are a lot of safeguards, of course. Look, even as I’m speaking, a few Waymos are passing by me here. This went from a state where people waited a long time for it to happen, and now that it works, it’s just robotaxis and the statistics speak for themselves. It perfectly fits the well-known saying that things take longer than you think in the short term, but arrive sooner than you think in the long term.”
Let’s move to our subject, AI and science. When ChatGPT and Gemini arrived, people said this was the technology that would cure cancer and bring scientific breakthroughs. So far, the only thing it has cured is developers of access to pastries in the kitchenette, in favor of the unemployment office. When will that promise come true?
“Let’s start by saying this is nothing new. This year marks a decade since we published a paper in JAMA, one of the leading journals in health, showing that we could use machine learning to detect an eye disease called diabetic macular edema from a retinal image. This is a disease that, if left untreated, can lead to blindness. We worked with partners in Thailand and India and brought it into the clinic.
“About two and a half years ago I visited Bangkok and saw patients sitting in front of a machine, and within two minutes receiving a diagnosis that saved them from blindness. More recently we also published a research paper with the NHS in England that uses AI as a second reader for breast cancer mammography. The paper showed that AI helps identify 25% of the misses made by experts.
“Now the world of generative AI is opening up an entirely new field of opportunities. The most exciting chapter today in AI, science and medicine is how to use AI to accelerate research itself. The road is still long before we can solve all diseases, but we can see the path in which AI can accelerate scientific research toward solving the major problems.”
At Google’s developer conference in May you launched three tools meant to advance this. How will Literature Insight save the hours academics spend on research literature?
“In the area of literature there are two major problems that consume much of research students’ time. One is figuring out what literature is relevant to what you want to solve. In many fields there is an explosion and a huge amount of scientific publications. The second issue is that there are many scientific publications that could be relevant, but you do not know that because they are in slightly different fields. Even after you already have the papers, reading them and finding what is relevant is not a simple task.
“With the insights, we can take a paper or any written text, summarize it, show insights about it, create an infographic, turn it into a presentation or transform it into a podcast. One of the powerful things we examined in AI Co-Scientist is how we can help with the basic steps of literature review. Once we ask a scientific question, we need to do the whole literature review, but then the real work begins, how do I connect the dots, build a hypothesis based on everything I see in the literature, generate many hypotheses, filter them, rank them and present them back to the researcher.”
“The researcher essentially becomes the supervisor.”
Now we are talking about hypothesis generation. I thought this was a field of human creativity, but you are bringing in a tool that incorporates AI.
“Absolutely. If we go back to the traditional research path, we need to understand each paper on its own and read it, and that is certainly the right thing to do for the papers that are most relevant to what you are doing. A researcher must read and understand the paper well on their own. But think about problems where what you want is to look at a literature review, when the information you will use to generate the hypothesis rests on hundreds of publications, or 100,000 papers. It is beyond human ability to read all those papers and understand them. This is where we would like to see how we empower the researcher with AI, which not only helps read a paper but does the entire literature review and starts answering the research question.
“The researcher becomes, in a certain sense, like the supervisor. If we look ahead, I see a world in which every researcher and every research student has a virtual laboratory that does the work they do today. This allows anyone to take on a role that once required a very senior researcher: asking questions, looking at hypotheses, doing the iterations and asking the next question.”
The third tool is computational discovery, and it already touches on the research itself.
“True. Computational discovery is based on a development called the Empirical Research Assistant. When you think about research work, once there is a scientific hypothesis that you want to test, for example in medicine, you want to build models of an epidemic outbreak like COVID, find correlations and so on, one of the most time-consuming and skill-intensive tasks is building models. It can take days, weeks or months of trying different models, tuning parameters and finding the right model for the problem with the different combinations.
“This development is meant to solve that problem. It says that if I have a problem that is scorable, and I have its input, the system uses generative AI to search for the best model in the world among thousands of models, and does the same work that we would otherwise need to devote months and the right skills to. The system lets you pose a research question of, ‘I want to calculate a model for this problem,’ and get a solution that can be tested and adopted.
“One of the things limiting scientific research today is that we have focused researchers on specific fields so they can go very deep. But often the more significant things happen when you connect fields together, when you draw on insights from chemistry, physics and mathematics. One of the most important things in research is people who are able to understand more than one field and connect them, we call them polymaths. You can think of the AI Co-Scientist system as a kind of polymath in your pocket. Each of us can have a partner capable of looking across all fields and making the connections.”
“There must not be cracks in the scientific approach.”
Does a researcher still need to be someone who has done a doctorate, postdoctorate and specialization, or in the new world can a researcher be someone with only a master’s degree or even a bachelor’s degree?
“This opens up an opportunity for more people to participate in the most advanced research. Education is still extremely important, we need to accelerate education and the ability to ask research questions. All these tools are subject to the scientific method. We need very serious discipline when we use AI for scientific research, to make sure we do everything according to the scientific method, which includes an element of validation. In a world where you can generate endless hypotheses, the ability to test them is more important than ever. The ability to evaluate and give researchers tools to validate is important to ensure that we maintain reliable scientific research. The scientific approach is one of humanity’s most important achievements, it is what has allowed us to build layers of knowledge in a decentralized way so that researchers around the world can rely on them, and there must not be cracks in it.”
Still, what is the advantage of a senior researcher with decades of experience behind them and enormous theoretical and practical knowledge, compared with a researcher who completed a research master’s degree and uses all the tools you developed for research?
“The tools for accelerating scientific research are going to raise the bar for the problems we are trying to solve. Therefore, one of the most significant things for me is asking bigger questions. The fact that anyone can use the system to solve a problem means that an entire family of problems will be solved by people using these systems properly. Now we are moving to the next stage, more complex problems, ones the systems cannot solve on their own, and there you need a researcher who asks the scientific question and moves forward with it. We are not trying to solve the same questions we solved before, we are going to ask bigger questions.
“If we go back to the start of the conversation, can we cure all diseases? That is our goal. We want to understand the world, physics, the planet, the human body, cells and biological systems to the end, understand how to identify diseases at an early stage and how to invent solutions. In my view there is no limit to the scientific and medical challenges that need to be solved, and our opportunity now is to upgrade the questions. AlphaFold, which was recognized in the Nobel Prize awarded to my colleagues Demis Hassabis and John Jumper, took problems that people previously did doctorates on, solving the folding problem of a single protein, and today provides solutions for hundreds of millions of proteins.
“My hope is that AI will accelerate human innovation. That it will be something that empowers the doctor, the teacher, the researcher and the scientist. The importance of the researcher, the doctor and the teacher in this equation is greater than ever. The ability to make an impact in the realm of human creation and in places where human context is needed is critical, and the importance of the human is greatest. Many things that today seem out of reach, I believe we will be able to reach and solve, including things we all care about in health, education and our planet, and see how we empower people in every field possible.”
