0:00

Leading the Shift—a podcast from Microsoft Azure Episode 6: Ade Famoti | Microsoft Research [MUSIC] Susan Etlinger (00:05): Welcome to Leading the Shift, a podcast from Microsoft Azure, where we hear from the people at the forefront of data, AI, and cloud technologies. I'm your host, Susan Etlinger. In our first few episodes, we spoke with leaders about how they're building and what they're learning with data, generative AI, and cloud technologies, but today we're going to switch it up a bit. We're going to talk about the future, and specifically we're going to share some of the research Microsoft is doing to advance science and technology for the benefit of humanity. A few months ago, I had the opportunity to interview Ade Famoti at a tech conference. It was such a mind-blowing conversation that I knew we had to share it with you. Ade is a senior leader in the Microsoft Research Accelerator division. He focuses on exploring the intersections of science, technology, business, and society on a global scale. And one of the things I most appreciate about Ade is that he bridges two distinct, but really interrelated areas of Microsoft, our research lab and our commercial business. He and his team are bringing the next wave of AI and advanced computing innovations into Microsoft's ecosystem. And every time I speak with Ade, I walk away with a longer and longer reading list, so brace yourself, we're going to go deep. Ade, welcome to Leading the Shift. Ade Famoti (01:24): Thank you, Susan. I was really excited to be here. It's such an important moment to be discussing AI, and I'm really looking forward to this conversation. Susan Etlinger (01:32): So, let's start with you. Let's talk a little bit about you. Tell us a little bit about yourself, how you got into this field, what you do, all of that? Ade Famoti (01:41): Yeah, sure. I'm a computer scientist, Susan. So, my background is in AI and robotics. So, I kind of started early building x86 Windows PCs on my own programming in GW-BASIC right out of high school. I joined Microsoft subsequently 24 years ago, and I worked as a messaging and email transport expert. So, I kind of dabbled in a lot of things, including Unix, Novell, IBM, Lotus Notes, and the even did some cryptography, but my AI interest really began when I was an Exchange Ranger architect. I was really fascinated by Naive Bayes spam filtering in the early 2000s. But I think over the years, I've also held several leadership roles across Microsoft, from sales and strategy, helping enterprise customers with their digital transformation journeys, to product, and now recently research. So during the pandemic, I was invited to lead outbound product for robotics and AI simulation startup. This is where I really got a ringside view of building an AI product. And I think more recently I was invited to build and lead Microsoft Research Accelerator's AI focus and incubations team. So this is working with a team, a phenomenal team, focused on turning research into impact across real world innovations, across robotics, material science, drug discovery, and a spectrum of research frontiers. Susan Etlinger (03:05): That's amazing. Okay, so let's get into it. I've been thinking about and talking about this idea of a platform shift with so many different people and the conversations that you and I have had been just, I think, so inspiring for me. And it strikes me that in some ways, the time that we're in right now mirrors that sort of 50-year period between 1870 and 1920 when so many different things happened, but the world could not have imagined what the implications of those things would be. So for example, X-rays, the discovery of viruses, the theory of relativity, the combustion engine, the first airplane flight, the commercialization or the mass production of steel, all these different things happened that created these sort of secondary effects that really changed the way that we live in the world, and you've said that AI is now having its physics moment. So, I'm curious to hear your perspective on that. What does that mean to you? Ade Famoti (04:02): Yeah, so look, every major platform shift, so this could be PC or the internet or mobile cloud, there's been social media and AI in its numerous winters, every single one of those shifts has brought a ways of innovation and disruption, but the GenAI moment feels fundamentally different. It's almost like a seismic paradigm shift, and it's got broad implications across domains. So, when I say AI is having its own physics moment, reflecting back at the golden age of physics in the late 1800s and early 1900s where we saw Albert Einstein's special theory of relativity just totally append a new tone in thinking. Quantum mechanics totally changed how we understand how the world works and really expanded our worldview around time and space. I think similar to physics, I think AI is kind of navigating the same trajectory. So, we're seeing breakthrough after AI breakthrough, we're seeing lots of debates, we're seeing lots of skepticism, and I think ultimately AI will navigate those same hurdles that physics navigated, and totally, completely change our understanding of how the world works. But at the core of this is generative AI, I would say at the forefront of the shift. So now we have these machines they have, and these algorithms they have remarkable ability to generate things like text, and images, video, code, and now we're even seeing scientific hypothesis by these models using inductive biases or perhaps what I call speaking the language of nature. And now we're seeing AI taking more complex reasoning tasks. So, we're seeing these reasoning models, and this totally just expands what we think is possible. Susan Etlinger (05:42): Wow, okay. There's a lot here. Scientific discovery, AI understanding the language of nature. Can you elaborate on what you mean by that? And let's talk about that in terms of materials, discovery, and drug discovery and all that kind of stuff. Ade Famoti (05:58): So, AI is accelerating scientific discovery, I would say, in ways we've perhaps never seen before. So, think about accelerating the material discovery process. So, materials and discovery materials, you could argue it's been a cornerstone of all of civilization, from steel to iron, to copper, to perhaps even sand or in silicon. But one area that we're working on is using artificial intelligence to discover, identify these new materials. And I'll tell you that the material space and the chemical space for these materials, it's pretty vast. Some folks will argue it's 10 to the 180. Susan Etlinger (06:41): So wait, 10 to the 180 what, opportunities or what is the number? Ade Famoti (06:46): Novel materials that are yet to be discovered. Susan Etlinger (06:47): Oh, wow. Okay. Ade Famoti (06:50): Yeah, absolutely. So at Microsoft Research, we have two deep learning models I'd like to talk about perhaps for the listeners. One is MatterGen. So, MatterGen is a deep learning model that simply proposes new stable inorganic materials or crystals, and Madison, just going by its name, is a deep learning model that simply simulates the properties of these new materials. So both of these models have huge implications for novel materials, for semiconductors, superconducting magnets, sustainable energy. So, now they can help us imagine new lithium ion chemistries to help us usher in this new world of electric vehicles, and there's even great opportunity for space exploration also. So, this is traditionally where traditional material science has really not really lived up to par because it's relied a lot on trial and error, but now with artificial intelligence, the possibilities are just immense. Susan Etlinger (07:55): So, break that down a little bit though, what's different about the way that we can discover materials using generative AI? Ade Famoti (08:02): So, now what we have is the deep learning models and the artificial intelligence toolkit that we have offers the opportunity to search this large space across the periodic table. So, there's 118 elements in the periodic table, so now we could start to imagine at the atomic level, new substrates, new compounds, new alloys, just new materials that we've never even seen before that could... To think about the energy transition materials that could help us with decarbonization, materials that are more carbon-free, materials that are more sustainable. There's a wealth of possibilities out there with AI. Susan Etlinger (08:45): I got it, that's amazing. So, any examples you want to share in terms of just going deeper into some of the things that you've seen? Ade Famoti (08:54): So with materials, if you look back at traditional materials, the discovery of graphene, was it thin? This was a big deal in the materials world, but now with MatterGen, there's possibilities, like I said, using AI to offset... They set up early trial and error discovery. One of those discoveries was a new material we just synthesized using MatterGen called tantalum chromium Oxide. So, that is a material that has never existed before. It's discovered in MatterGen, synthesized in the wet lab with a partner, and now we have a discovery. Susan Etlinger (09:39): So, it strikes me that as you're talking about energy efficiency, we can also talk about the impact of novel materials on the earth and optimize for the things that we want to see, right? Ade Famoti (09:50): Absolutely, there's a wealth of opportunity in this space. So, now with artificial intelligence and these deep learning models, we could start to imagine new energy chemistries, so think lithium ion chemistries. Lithium ion's been around for seven decades. Now with AI, we can imagine new cathode chemistries other than nickel, cobalt, and aluminum, or nickel, cobalt, and manganese. So, this brings tremendous opportunity where we can now have high energy densities in the batteries for electric vehicles and great storage. We can have cars that charge a lot faster. Now, we could really now usher in this world of electric vehicles. Susan Etlinger (10:30): Wow, so let's shift gears again to drug discovery, because I just find this fascinating. I mean, I think a lot of everyday people think about generative AI in the context of video and audio and text and all these things, but you're thinking about generative AI in the context of CT scans and X-rays and MRIs and blood work and all these different sort of pieces that can potentially come together to help us better understand and treat disease and discover novel drugs. So, could you talk a little bit about that research? Ade Famoti (11:04): Yeah, so one of the most promising applications of AI, and this one really excites me, one of the most promising is just identifying new drug candidates. There's a lot of diseases out there. The conventional drug process takes a long time. We've got 7,000 genetic diseases out there we need to find cures for. So, what AI is doing is helping us develop new drug candidates simply. And what is a drug? A drug is typically a small molecule that binds to a biological target in the human body, and that's typically a protein. So, it alters the target's function and produces a desired therapeutic effect. So with that definition, a drug could be anything from caffeine to nicotine, but the drug discovery process is about finding the right molecules and getting the desired effects, as I said. So, there's a lot in there, of course, in between what AI can do, we still have to test for it. Again, is the drug still absorbable in the human body? Is it distributed in the right way? Can it be metabolized in the right way? Can it be excreted and does it have the right toxicity or not? Or can it mitigate for toxicity in the human body? So ,what we have at Microsoft Research, just accelerating discoveries in this space, is a deep learning model called TamGen, so, TamGen stands for target-aware molecule generation, and this model helps accelerate small molecule discovery. It works by helping predict how different compounds intersect the molecular level, it works to inhibit and bind to the ClpP tuberculosis protease. So in essence, it's really just helping... It's fighting tuberculosis. And if you think about the impact of TB, as they call it, this is a disease that has tragically killed millions globally. So, there's tremendous promise here with TamGen and other discoveries in this space. But the traditional discovery process, and I think here's the meta point, it's an incredibly slow and expensive process, and AI now could really drastically reduce and shorten the cycle by identifying viable and de novo drug candidates much faster. So yeah, this space is really exciting, and I think the impact on healthcare, health outcomes, and medicine could really be transformative. Susan Etlinger (13:26): So, it's the discovery part of it, and that part takes obviously a tremendous amount of time, and then of course, the rest of the process as well, but you're focused really on the discovery piece. Ade Famoti (13:39): Absolutely, there's still a lot in the direct design process. There's identifying the target as one using AI or other conventional techniques. There's preclinical, there's clinical, there's phase I, II, III, and there's FDA, so there's a lot in there. AI identifying the drug does not mean it's ready for human consumption. There's still a lot in there, but now we can really accelerate and catalyze this whole process. Susan Etlinger (14:02): And thinking about it, I think one of the things we talked about earlier was sort of what machines are good at versus what humans are good at. And the paradox there, is it Moravec's paradox? Ade Famoti (14:17): Yeah. Susan Etlinger (14:17): Yeah, and so the idea that a human researcher, a human doctor cannot possibly ingest all of these data points, no brain, I don't think, but any brain is capable of doing that and then generating the result. Obviously, and that's what we call, I guess, human trial and error, but the scale upon which we can actually act seems to be just expanding exponentially, which is really, really interesting and just exciting, I think. Ade Famoti (14:51): Absolutely. No, it's a great time to be alive, Susan. It's just the wealth of possibilities. The moment we're in, it's just absolutely just mind-blowing. Susan Etlinger (15:01): Yeah, so we've also talked about embodied AI, and I'm just fascinated about this too, because I think for a lot of people, of course, a lot of people sort of who are not physicists, who are not scientists, or computer scientists, their vision of what AI is, a lot of it was really influenced by science fiction and kind of that 70-year-old history and this notion of embodied AI, we see a lot of it sort of confused sometimes or conflated with robotics. And it was really helpful for me when you explained the difference, and I think our listeners would appreciate that, too. Ade Famoti (15:40): Yeah, that's a great question. So, I kind get this a lot. So, robotics and embodied AI, somewhere i overlapped, but absolutely not the same thing. So, robotics is more about physical machines, and you could consider embodied AI as entails given AI, artificial intelligence, a physical presence. So, whether that's physical presence in the physical world or in the virtual world, so physical world, we're talking autonomous robots, self-driving cars, we're talking robo-taxis. And in the digital world, that could be embodied AI in virtual assistants and AI-powered avatars. Yeah, so that's embodied AI. But to give your sense how we got here, so Butler Lampson is a Turing Award winner for computing and a former Microsoft technical fellow, and it's fascinating how Butler Lampson describes computing. He describes it in three epics. So in the first epic, he speaks about machines that were built. We built machines that compute faster than humans. And in the second epic, we connected those machines to share information globally, so think about the birth of email and story technologies. And now in the third epic, it's all about interaction, machines that understand, perceive, respond, and act within the physical world. So, that really just encapsulates embodied AI. Susan Etlinger (17:10): And so let's talk a little bit about how that must be very culturally dependent, because what people would want in one region of the world might be very different from what somebody might be needing or wanting in another place in the world, either because of geographic concerns or cultural concerns or other types of things. So, can you give some examples of additional places that's coming to life? Ade Famoti (17:36): So, culture plays a big role. I was recently in Japan, spent some time in Kyoto, in Tokyo, and it's fascinating seeing those humanoid robots. Humanoid robots, they kind of look like humans. They have arms and legs like humans, but obviously not human, and they have almost human-like dexterity. And so, it was fascinating to see these robots accepted and integrated into social setting in Japan. But if you think about the Western context, robotics and automation or embodied AI tends to be more function-driven, like industrial robotics. So, it's interesting because these cultural nuances really shape how AI is developed or adopted within those cultural regions. Susan Etlinger (18:22): And it must be economic as well, obviously. Ade Famoti (18:25): Absolutely. I mean, in the Japanese context, they have a near existential crisis where they have an aging population and a declining birth rate, so that also is a major cultural influence as well. Susan Etlinger (18:37): Yeah, I mean, I think this is a really fascinating point, which is that the way that embodied AI, robotics, even drug and materials discovery, all of these things, many of them have probably global implications and global impacts, and some of them are going to be really regional, really specific to culture. And I feel like there's space for all of that, and that one of the things that we need to do as everyday people, and then of course all of you need to do as scientists, is to keep those ideas in your head simultaneously, which is a little bit challenging, but I think also really to the good, because when we look back at the things that have happened in the past, one of the things we can learn is, where did we make decisions that ended up being beneficial in the long term? And where do we make decisions that maybe seemed beneficial in the short term, but in the long term we learned something else and then we needed to make course corrections, and it just feels like a lot of the thinking that you're doing in your team is doing is around trying to keep both those short and longer term considerations in mind. Ade Famoti (19:45): Absolutely, I think you're absolutely spot on. I mean, there's been a lot of advances in this space, but there's also tremendous opportunity. So take Moravec's paradox, so this is a paradox why Hans Moravec, so a computer scientists, and Hans Moravec describes a tasks that are easier for humans, like walking or recognizing emotions, these are incredibly difficult tasks for AI and robots. If you think a complex task like chess, those are relatively easy for AI. So, this also has its own cultural philosophies that it influences also. Susan Etlinger (20:22): Yeah, so as you think about the future, and of course the future can be six months, it can be six years, and you think about leaders who are making these decisions, whether they're in the private sector, public sector, academia, government, wherever they may be, you talk to a lot of different people around the world. What advice do you give them? What advice do you have for how they should be thinking about planning for understanding this shift that we're in? Ade Famoti (20:56): Okay. No, I think that's a great question. So yeah, this is not legal advice, but I'll offer some remarks. Susan Etlinger (21:03): Disclaimer. Ade Famoti (21:05): So, I think the biggest challenge for leaders today is not just keeping up with AI, it's also learning how to think differently about their industries and their domains. I'll offer three remarks for leaders. I think the first one is embrace AI as a co-pilot, so not just as an efficiency tool. Think about it's not just about automation, it's really about augmenting human capabilities. So, I'll urge leaders to look for ways AI can embrace creativity, decision-making, and innovation within their domains, within their respective industries, so that will be the first thing. The second thing is I'll urge leaders to invest in AI literacy across your organization. Everyone should be AI native and super AI literate. So, the companies and organizations that will thrive going forward, those where AI is not just in the hands of specialists, but in the hands of everyone, it's understood and used across teams, across domains, across boundaries within the enterprise. And I think third will be I will encourage leaders to experiment and adapt. So in this moment of AI, it's almost like an AI renaissance we're in, it's an era of rapid iteration. So, leaders need to foster and encourage and imbibe a culture of experimentation that could be going through small prototypes or small AI pilots or integrating AI into existing workflows, just do it, just get it done, just foster that experimentation mentality. So yeah, so again, it's not legal advice, but I urge leaders to consider those considerations. Susan Etlinger (22:54): Absolutely. No, none of this is legal advice, but to your point, a lot of the premise for this podcast is to share what we're learning here at Microsoft, share what our customers, our partners, and some other folks in our ecosystem are learning, really to just help people grasp the magnitude of the changes that we're going through, the opportunities, the threats, and to try to contextualize them within their own industry, within their own domain and discipline. So on that note, I really, really appreciate your just thinking it through with us and sharing what you're learning. Thank you so, so much, Ade. It's just been a pleasure. Ade Famoti (23:37): No. Hey, thank you, Susan, totally my pleasure. It's been great talking with you. These are exciting times, and I'm looking to seeing how AI continues to evolve and shape this world we live in. Susan Etlinger (23:48): Me too. Thank you, Ade. Wow, okay, so much to unpack here. First is we mostly think about generative AI for content and code development, but there is so much more to explore: drug discovery, new materials, and the potential of these new technologies to help us improve health outcomes, energy efficiency, and a host of other human needs. I also really appreciated Ade's perspective on how culture shapes priorities for emerging technologies, as well as how they will be adopted, and more importantly, trusted. This is becoming a through line in so many of our episodes, from Perry Hewitt on data for social impact, to Robby Ingebretsen on creating a fan remix experience for Coldplay. And finally, I love that as different as Ade's work is from some of our other guests, his advice to leaders is so consistent. Think of AI as a way to augment your strengths, invest in AI literacy, and finally, experiment, adapt, and foster a culture of experimentation. Of course, you'd expect to hear that from a scientist, but hey, we're all scientists now. We hope you've enjoyed this episode, and as I'm sure you know, new podcasts live and die on engagement. So please like, comment, share, tell your friends, and let us know what you'd like to hear about. We're listening. If you'd like to learn more about how people and organizations are innovating with Microsoft Azure, visit azure.microsoft.com. The Leading the Shift podcast is a place for experts to share their insights and opinions. As students of the future of technology, Microsoft values inputs from a diverse set of voices. That said, the opinions and findings of our guests are their own, and they may not necessarily reflect Microsoft's own research or positions. Leading the Shift is to production of Microsoft Azure. I'm your host, Susan Etlinger. Our executive producers are Hailie Meehan, Aaron Russell, Erik Williams, and me. Production support provided by the incredible team at Indigo Slate. [END]