So I've heard you say many times that at Flagship Pioneering you take a future backwards approach to innovation. You envision better futures and then you work backwards from there to achieve what you want to achieve. So how is AI changing the way you think about the future? What futures are you most excited about?
In a nutshell, I'll say that over time, what Flagship does, what I've been involved with has taught me one thing, which is you can live your life thinking from the present forward, or you can live your life thinking from the future backwards. The advantage of present forward is that people think you're realistic, you're pragmatic, you're rational, you may even be an expert. There's no experts in the future. It turns out there's only experts of the present. But if you're willing to think first about the future and then do things in the present that will achieve that future, you literally flip the time arrow. And I think with science, technology, and many other areas, leadership involves being able to think future backwards and tie it to the present reality. Because often, at least we find it's daunting to think how are we going to overcome the major challenges of the world viewed from the present. But if you can just lead to a future state, imagine, get people collected around that future state, agree on what's worth building, then the road to that becomes pretty simple. You no longer have to fight over direction, you just have to fight over destination. So while I know that lots of people go around saying the destination is not important, it's really the journey. I think the destination is really important because it can actually inform, instruct, reinforce the journey. So that's what we mean. And AI, just to end, is a tool like none other to help you envision destinations, envision alternative paths to get to those destinations, and even prescribe and help you navigate the journey to those destinations. So I can't, since the 39 years since I got my PhD at MIT, I can't imagine a gift like AI for people like us, innovators, journeymen in life sciences to be able to use to go further, faster, and have more impact. So this is really like I feel 20 years younger as a result singularly of AI. All these people talk to me about longevity. As far as I'm concerned, using the tools of AI is a form of longevity, if you can do a hell of a lot more in the time you've got left. So there you go.
Wow. I have to say Noubar is on the board known as the Corporation at MIT and he is our board's optimist and so I always come away invigorated from these conversations. And one thing that you've spoken a lot about, written about, is poly intelligence, the synthesis of human, machine, and nature's intelligence really is the next frontier. So, how do you see these sort of poly intelligence systems shifting tools, mindsets in how science is going to be done over the next decade?
Very much so. So, Sally already revealed me as being an optimist. I go around saying that I'm a paranoid optimist. I'd invite you to think about yourself as a paranoid optimist. You can be as optimistic as you want, provided you're equally paranoid. I just came from another session where people were talking about skepticism about AI and I also said to them we should invite ourselves to be skeptical evangelists. As long as you can be an evangelist at the end of the day, be as skeptical as you want, but if you're just skeptical, especially about AI, and we'll come back to that, then you're like a film critic. It's really easy to criticize a film. Everybody can do it. It's really hard to make a film or make it better. So I would invite us to find a balance. Now, poly intelligence in a nutshell, those of you who are interested, is a term we coined a year ago to try to reveal that while humans are obsessively now engaging with artificial intelligence, something that we created, we've now made room in the language for another form of intelligence than our own. I think you'd agree with me that humans having controlled language also have controlled what gets to be intelligent. And for the longest time until recently, we believed humans are the only intelligent beings on earth. In fact, we made sure animals were not intelligent. I want to say parenthetically, those of you who are not following the field, that about three months ago, a pretty definitive paper came out demonstrating Bayesian reasoning in chimps, adjusting conditional action based on adjusted priors. I don't know if that's not a basis for intelligence, but I'm going to leave that aside. I think there is even neuronal intelligence in other animals. But in any case, we said there's human intelligence and now there's machine intelligence, AI. And what we pointed out from the life science world is that now that we've made room for one form of intelligence, how about we think of cells and everything in nature as forms of intelligence. Now, you might say, how could that be intelligence? It doesn't think like us. And of course, that would be human conceit because why does it have to think like us to be intelligent? If you think about intelligence as largely a form of reinforcement learning, whether it's done on GPUs or whether it's done in our neuronal circuits in our brains or reinforcement learning in the form of DNA. What's the reinforcement learning in DNA? Darwinian evolution. It's all it is. It's coded in a substrate of DNA. It makes its own processor called a cell. And it embodies all of history's learnings up until that point and acts and adapts and continues to learn. If that's not intelligence, if a virus that can almost take down all of our immune systems is not intelligence, if a T-cell that could find two, three cancer cells and kill them is not intelligence, I don't know what is. But what I just point out is what by poly intelligence we mean humans, machines, and nature coexisting as three forms of intelligence. And the punchline is if we could use machine intelligence to emulate nature's intelligence in ways that we've never done before and in ways that our brains cannot properly capture and process, I would argue we may well have a chance in our lifetimes very soon to begin to talk about nature, the complexity of disease and preconditions and treatments in ways we have not had available for the entirety of the biotech pharmaceutical industry. So imagine a mechanism that involves 5,000 proteins interacting in hundreds of ways. Which paper are you going to describe that in? That's something that machine models can do just fine. And it may not be accurate in the beginning, but then give it time. So I would say at least a prediction for the future, and I'm hoping MIT leads the way in this regard, is humans deploying machine intelligence to reveal nature's intelligence for our benefit. The only thing is what has to change is our belief that we're the supreme being and our belief that we need to understand everything in order to act on it. None of which is true. I don't understand how my iPhone does what it does, nor do I understand how the airplane does what it does. But boy, do I use it. I think you're going to see AI reveal all sorts of systems level information that we're going to act on. So with that said, let me flip over to asking Sally a question. MIT, Sally, does not have a medical school, but by any means with AI, it could certainly make massive medical inventions and advances. So how does AI change the way you're thinking about life science, biomedical science and the practice of medicine altogether?
Yeah, you know it's interesting. I think that as you point out MIT doesn't have a medical school yet we are surrounded by hospitals and I think in the past we did a lot of work where we were pioneering in the fundamentals of AI or the fundamentals of quantum or the fundamentals of all sorts of areas of basic science but the direct application to health was more difficult for us. And what we did in the past year with the help of generous donors, Analog Devices and others, is we've set up pilot programs now, for instance with Mass General Brigham, where we have asked clinicians to tell us, help us define what problems they are finding thorny where the MIT AI expertise could actually be applied to solving real world problems. And you know, we put out a call and I think we got like 117 applications. I mean this was a sort of latent need out there. I mean there's sort of low hanging fruit which as you know, all of the imaging modalities are being profoundly affected by AI. You know, we have folks like Regina Barzilay who can now detect lung cancer, you know, six years earlier using AI analysis of CT scans that were not evident to investigators. But there's a lot more that we're starting to think about. So for example, Jim Collins and others thinking about one of the reasons that our lifespans are so long, as you know, from something like, like now, is we can deal with infectious disease now in a way that was difficult but we're starting to see a rise in antibiotic resistance. We're seeing fungi now with global warming becoming a pathogen that we have to worry about. And so using AI to think about how we improve on things like antibiotics, thinking about how we design new therapeutics and also just thinking about how patients' experience is different. How do we pioneer with hospitals to connect people to the right care? How do we mine the electronic health record to look for the harbingers of disease? Pancreatic cancer is a good example. So basically we are continuing to innovate in AI but rather than just pull the problems out of thin air on the ground we are making partnerships with the clinicians to really figure out where AI can help them and where innovation in AI can make a difference in their world.
