#198 Sethuraman Panchanathan: How The U.S. National Science Foundation is Shaping the Future of AI

0:00This cannot and must not be just about federal investments only. It's got to be everybody coming together. And I expect as we go to full scale, that is the approach that is going to be the one that's going to scale, make it scale, is that everyone is saying, we need to scale up what we are investing in. Cloud resources, GPU resources, model resources, other kinds of algorithmic software resources, right? Other kinds of tools and other resources. Everything is going to have to scale. AI might be the most important new computer technology ever. It's storming every industry and literally billions of dollars are being invested.

0:35So buckle up. The problem is that AI needs a lot of speed and processing power. So how do you compete without costs spiraling out of control? It's time to upgrade to the next generation of the cloud, Oracle Cloud Infrastructure, or OCI. OCI is a single platform for your infrastructure, database, application, development, and AI needs. OCI has four to eight times the bandwidth of other clouds, offers one consistent price instead of variable regional pricing, and of course nobody does data better than Oracle. So now you can train your AI models at twice the speed and less than half the cost of other clouds.

1:20If you want to do more and spend less, like Uber, 8x8, and Databricks Mosaic, Take a free test drive of OCI at oracle.com slash IonAI. That's E-Y-E-O-N-A-I, all run together. Oracle.com slash IonAI. That's oracle.com slash IonAI. Okay, great. So, Dr. Pantranatham, I hope I'm pronouncing it correctly. Very close. I'm really delighted to have you here. And I wanted to talk to you specifically about AI, about how AI fits into the National Science Foundation's work, into the funding work that you do. But I thought maybe we would start by having you introduce yourself to listeners and tell us how you got to the National Science Foundation.

2:21I know your background, but listeners may not. and then we'll talk about how AI fits into the NSF's mission. Yes. So go ahead. My name is, thank you, Craig, and thank you for the opportunity to speak with you. This is an exceedingly important topic, needless to say, and that's something that in everybody's minds these days, which is a good thing. At the same time, we need to make sure that we are able to talk about this in ways that people can understand, relate, and therefore apply it in ways that is more appropriate for what it is that they are looking for. in their daily lives and their professional lives and beyond.

2:57So my name is Setu Raman Panchanathan, and I'm the director of the National Science Foundation. And, you know, maybe I can dial back and go to my PhD student days. Yeah. Right. And my own research was what it was not called as machine learning those days, but it was machine learning. I worked in vector quantization, which was essentially about how do you get image or sub-images to be clustered and then find representative images so that you might use it for compressing images through vector quantization. So these concepts of clustering, labeling, training, all of these things were part of what, you know, I was engaged in quite extensively in terms of, you know, in test set training without the test set being part of the training.

3:45All of these things that people talk about these days as part of what happens in machine learning and AI is something that I've seen firsthand. And I'm now dating myself, late 80s. And then the neural networks were starting to become more prominent. And we worked a lot with those kinds of models. Then from then on, I became a faculty member in computer engineering, computer science engineering. I founded a center for cognitive ubiquitous computing. And this was in early 2000s. in terms of how do you look at AI, machine learning, multimedia, and various media, you know, sort of assets. To be able to look at how do we ensure that people with a range of abilities can have enriched lives, can, you know, excel in their performance, things of that nature, and most importantly, make opportunities available for everyone.

4:39And through that journey, I have seen amazing things happen. I initially worked with individuals who were blind and visually impaired. And the kinds of things that we were able to make happen because of their involvement, because of their engagement and design and ideas that they brought to Frugian. And these were students who were blind and visually impaired that were engaged in mathematics, in computer science, in engineering and so on, coming and working in my research center. I saw through that amazing possibilities of what can be done with machine learning, AI and multimedia. But even more importantly, what can happen when you provide opportunities to everyone and that unleashes amazing possibilities, not only for them, but for everyone.

5:24And so that's the background I was sort of through the years and what I was engaged in. And in 2014, President Obama appointed me to the National Science Board. And before that, of course, and during that, I had the good fortune of securing a number of NSF grants of various types. You know, single PI grants to NSF I-Core grant to NSF I-GIRD grants to NSF Major Research Instrumentation grants to Central Level grants to, you know, the NSF ITR, Information Technology Research Grant and the National Research Training Grant. So you get a good sense of how NSF is. But when I came to the National Science Board, I got a view of NSF from a different perspective, which was highly enriching and rewarding for me.

6:15And so I did that for six years as part of my appointment. And then I got appointed by the president to become the director of NSF. And through this journey, I've also participated in national policy kind of frameworks like the National Advisory Council on Innovation and Entrepreneurship, the Secretary of Commerce, I was part of that. I chaired the Association of Public Land-Grant Universities Council on Research, which essentially was all the vice presidents of research and innovation. And so you were able to see various things through my own experiences at Arizona State University, where we did a lot of transdisciplinary institutes and schools and so on, and through my own research.

6:56and then through these engagements nationally, I was able to get a good idea of what is possible. And coming from a small state like Arizona, I got to see firsthand that when you make possibilities for talent from a range of backgrounds across the socioeconomic demographic, across the rich diversity of our nation, across the geographic diversity of our nation, then amazing things can be unleashed in terms of ideas and innovations. At the same time, I also saw firsthand that innovation is possible anywhere in context. Every region of our country has the possibility of finding that thing that makes it unique for that region and that you're able to now unleash that innovation that then makes possible prosperity, economic security, national security, prosperity for everyone everywhere.

7:51So this is something that I found firsthand and I cherish that a lot. So when I came to NSF, that was one of the things that I wanted to make sure that NSF, while it has been doing amazing things, can that become also a focal point for unleashing opportunities for everyone everywhere, unleashing innovation everywhere for everyone. So that has been something that I've been working on quite a bit. And AI is a great thread. Yeah, well, that's what I wanted to ask. So NSF is a funding organization. And I think your budget at this point is$10 billion or something? $9 billion this year, approaching$10 billion, yes.

8:33And the funding for AI from NSF is about a tenth of that, is that right? Correct. $828 million just in FY2023, yeah. That's right. So how does the board decide how to allocate that funding? How do you decide how much to give to AI, how much to give to biotechnology, how much to give to whatever the other buckets are? And is the AI portion growing? Excellent question. So, you know, as you know, NSF is a very unique agency. It is the agency responsible for unleashing fundamental basic research as well as applied research and translational activities. Right. And the core mission is fundamental and basic research, right?

9:22Vibrant, fundamental, and basic research. So when you look at all the inputs that you receive, you know, from scientific community, we get decadal surveys, for example, that the astronomy community released recently, right? You get that kind of input that comes from the scientific community. You get, you know, the advisory committees of the various directorates. You know, I meet with them regularly, right? And then they tell us, what are the kinds of things the scientific community is engaged in? What are the kinds of things that they're especially looking forward to doing? And what does it mean in terms of needs out there?

9:56And then you look at the amazing amount of ideas that are constantly being sent to NSF. We get about several tens of thousands of proposals every year. Wow. Every year, right? I think we got some 45 ,000 proposals last year, right? So these are proposals that come from individual PIs, groups of PIs, small companies. All of them send the proposals to NSF with fantastic ideas. So you have that to understand how is the community looking at and advancing ideas, fundamental ideas. Then you have the Office of Science Technology Policy and the administration releases strategic plans. And the key investment areas that they are thinking about.

10:39We get guidance from the Office of Management and Budget on the administration priorities. Then we get congressional input and interest in terms of specific things that we need to advance. Then we look at the international landscape. All our international partners, like-minded partners, they are also evolving scientific plans and priorities. So we get a host of this. It's just a few of them I'm mentioning to you. We get a host of inputs that helps us synthesize our own thinking about the strategic plan that we put out. strategic thinking that we put out. It is not only on an annual basis, but what we do with our National Science Board in terms of looking at five, ten year kind of horizons.

11:18What are those kinds of things that we need to look at? And then we look at all of these things, you then get a distilled understanding of the priorities. And I've got amazing leadership team, the individual assistant directors of the various directorates and then their leadership teams. they all bubble up these amazing inputs also that comes up. And we sit together, we talk about it, we have retreats, and then we decide how we are going to allocate these resources, right, in a strategic way. Some of them are focused on those disciplinary kind of investments. Some of them increasingly are focused on transdisciplinary kind of investments.

11:57Now AI, let me remind you, that AI of today has been made possible by sustained investments by NSF for over five to six decades. And I constantly remind people that even during AI winters, when people walked away, not sure where this is going, right? Should we invest or not? NSF continued to invest because it believed that this has the promise and the potential and the ideas out there. That's what NSF is really good at. Sustaining these amazing ideas that then mature into what we see as of AI today. And now a lot more investors are coming in and so on and so forth. So that's one of the unique facets of NSF.

12:40So when you talk about where we are with AI today, in the fiscal year 2023, we had about$828 million of investment in AI. And that's obviously going to grow with time. And we're also looking to the amazing interest, the bipartisan support in Congress now. You look at the Senate AI caucus, you look at the House AI caucus, You look at the Biden-Harris administration and President Biden's interest in ensuring that we are in the vanguard of AI progress. You know, he put together this executive order last October, which is very comprehensive, very visionary. And so you look at all facets of this. You see the interest in industry wanting to partner.

13:21So I'm looking towards the future of these investments increasing because there is interest in wanting to do that. And I can give you many examples of that. And so based on what your questions are going to be, or I can talk about them. I just don't want to make this long so that you are not able to interrupt me and ask what you need to ask. Clearly, we invest heavily on core fundamental research, single PI proposals that come to NSF. Great ideas, machine learning, AI, AI applications. And now these days, you get people in geosciences, biosciences, mathematical, physical sciences, social behavioral economic sciences education directorate and every one of them every directorate of NSF folks are saying how do I advance my disciplinary science by bringing the AI-ness in my discipline and it does two things right Craig it is not only that the disciplinary aspirations are advanced in a rapid manner but I constantly remind people it also brings the new problems that AI needs to solve, the AI community needs to solve, or new problems that machine learning and AI community need to solve so that there can be more vibrant progress, not only because of what a geodirectory problem might have broadened, but that solution is going to be applicable to all of the directorates or all of the disciplines that are presented with the directorates.

14:51So that's the symbiotic nature of how we make progress. Okay. So I talked about the single PI kind of grants. The next conceptualization is what we call center level grants or institute level conceptualizations. What is it about? That's about bringing researchers from across disciplines and across institutions, across sectors, working together around applications, around basic research themes in AI. So we launched this concept of AI institutes three years ago, in 2020, 2021. Now, three rounds of AI institutes. On the AI institutes, we are having three rounds of institutes launched, 25 of them.

15:41They span the entire nation. 46 states are involved. But if you look at our AI investments in total from NSF, all 50 states are involved. I'm very proud of that because the talent ideas are democratized and they're everywhere and it's our responsibility to invest and inspire and make possible that talent those talented ideas so these 25 AI institutes NSF invests 300 million dollars out of the 500 million dollars in these 25 institutes each institute is 20 million dollars scale so 25 institutes half a billion dollars scale The other$200 million is from our partners. You'll be interested to know that five of the AI institutes in agriculture are invested in by USDA's NIFA.

16:32And we're very thrilled by that. Department of Transportation, Department of Homeland Security, NIH, DOE, NASA. Everybody is a partner. NIST. We have one AI institute, NIST, which is focused on law, ethics and society. So we're very excited by the fact that these partnerships from interagency and one of the AI Institute scale of investment was the consortium of industry, Amazon, Microsoft, Google, Accenture. Yeah. Actually, I had a question about the institutes. Are those existing institutes that are receiving funding or are you standing up institutes? And are the institutes physical locations or are they networks of researchers across institutions?

17:24They have a physical footprint in the main institution that coordinates it. And then the partners do have smaller footprints. So it has got a physicality, but also a virtualness in terms of a network of people, a network of institutions and entities working together. Yeah. And the funding goes to supporting staff personnel primarily, or are you funding compute resources or data collection activities? I mean, where does the funding go in the institutes? To all of the above. It depends on the particular project and what their needs are. I'll give you an example of a project. In this cycle, I was with Senator Schumer launching this project in University of Buffalo.

18:19And this is called the AI Institute for Exceptional Education. Now this institute is focused on working with speech pathologists to be able to provide them the expertise that is needed for screening kids who need speech pathology services. Also providing augmentation to speech pathologists. Why is this exciting? It turns out there are 3 million kids who need speech pathology services. We only have 62 ,000 speech pathologists. And so how do you bring AI to be able to now screen the kids that need help to understand what kind of help they need and being able to provide that help along with speech pathologists so that we might have unbelievable outcomes of these children participating in a full force in terms of the talent that they can express for their own lives but for the betterment of the nation.

19:22Truly exciting and so this is a consortium of universities and other entities working together. So if you look at each AI Institute. I can go through every one of them. It is just truly exciting to see how people have come together and really want to advance the next wave of what the AI innovations have to be, but also looking at solving problems for application domains that need AI infusion to be able to strengthen their approaches at speed and at scale. So I'm very excited by that. So that's a conceptualization, AI institutes. The other one that I will talk about is the National AI Research Resource called NAR.

20:04Right, that was going to be my next question. Yes, happy to talk about that. So if you look at this, as you see AI, you see the amount of compute resources, model resources, AI tools, software, right? And you look at data. You look at all kinds of needs that are there for people to be able to connect, express their ideas, configure their ideas, and find real solutions. And through that, training the talent for the future. Okay. Now, if you look at the, just take compute resources, just one example of that. It turns out that most of the people that are able to do work are people who have access to large computational resources.

20:52your chat GPT-4, GPT-5, and so on, you can clearly see that the compute resources availability is an important requirement for people to be able to be engaged. And then it essentially presents a situation where there are haves and have-nots. And as I said earlier, to other general forms of talent and ideas, AI, talent and ideas are also democratized. They're everywhere. They are institutions, community colleges, minority serving institutions, research two institutions and some research one institutions who don't have the resource capacity. So if we do not provide these resources, we would miss out on a huge pool of ideas and talent.

21:46And that's not a good idea, clearly. So we configured through a task force that was put together by the National Security Council and the Office of Science Technology Policy work together. And NSF and other agencies were part of that. And out of this Nair report came out one of the recommendations is the launch of a Nair pilot and a full scale Nair. And I was so glad and grateful to the executive order of President Biden. where he clearly outlined a number of things that we need to do in AI. And he tasked the agencies, different agencies to do different things. For NSF, one of the things that we were tasked was in 90 days, we had to build this NairPilot in partnership with other federal agencies and other entities.

22:36And within 90 days, I'm very proud to say that we have 11 federal agencies, Department of Energy, NASA, NIST, US Department of Agriculture, you name it, right? 11 federal agencies. And then 25 non-governmental partners. The who's who in AI. NVIDIA came. We put in$30 million for the Zener pilot from NSF. NVIDIA came in with$30 million of resources. Microsoft came in with$20 million. Amazon came with$10 million of resources. And the who's who in AI, every one of them brought in their own capacities to contribute. Right. And so with that, we just launched 35 projects. We launched this in the White House Office of Science Technology Policy event that happened in the White House on Monday.

23:28And some of them presented what they are doing. It's really transformative work that people are doing. And the access to these resources, they were saying, is going to help them to move at speed and scale. That's the first thing. We've got 50 more such projects who have been reviewed positively, waiting for resources. We already now released the next wave of requests for people who may have good projects that they want to put into place. But I'm very happy to say that more partners are joining the fray now and wanting to work with us. Because this is about partnerships. This is about the ecosystem working together.

24:06And as I said earlier, with the congressional focus and the administration's focus and interest in wanting to invest in the full-scale Nair, I am very bullish and positive that this is going to make possible real building of infrastructure and resources that's going to help keep us in the vanguard of innovation as a nation. Yeah. Well, I wanted to jump in and give a shout out to our sponsor, Vanta. Whether you're starting or scaling your company's security program, demonstrating top-notch security practices and establishing trust is more important than ever. Vanta automates compliance for SOC 2, ISO 27001, NIST, AIRMF, and more, saving you time and money while helping you build customer trust.

24:55Plus, you can streamline security reviews by automating questionnaires and demonstrating your security posture with a customer-facing trust center, all powered by Vanta AI. Over 7 ,000 global companies like Altasian, Flow Health, and Quora use Vanta to manage risk and prove security in real time. Get$1 ,000 off Vanta when you go to vanta.com slash IonAI. That's vanta.com slash IonAI. IonAI all run together, E-Y-E-O-N-A-I. So that's Vanta.com slash IonAI for$1 ,000 off. I wanted to ask about Nair. One thing, I did some reading before the pilot. There were a lot of ideas. But is it a pool of money to grant credits, like cloud credits, to research projects?

26:05Is it actual clusters of GPUs that researchers can access that are operated by the NAR? Or is it some combination of that? So, for example, when NVIDIA contributes$30 million, what is that in terms of GPUs? Is that in terms of GPUs? In terms of resources, compute resources. Is that time on their GPUs? Correct. Or are there physical GPUs that are being put together in a data center that are devoted to Nair? These are GPUs. They're given GPU time that is given clearly for us to be able to allocate to projects. So to answer your question, the first part of your question, it's all of the above. You know, when the Nair was put together, we had a couple of things, focus areas, Nair open, which means there are resources that are available for anyone.

27:04You know, this could be, you know, open source model resources, open source tools and things of that nature that people can dial in and get access to. And then there's Nair Secure, which, you know, for example, Department of Energy and NIH partnering on the secure side of things. And then Nair Software, the software resources that are being made available and so on. And then there is the near education piece, right? So we are able to provide resources for educators to be able to, near classroom resources, all facets. So what you're looking for here, if there are financial resources, compute resources, obviously.

27:46Financial resources is targeted at people who are going to be using these resources to advance ideas. There are some of the partners who are providing model resources, right? or algorithmic resources. So it's all of the above. And so depending on the project specification, we are able to then parse out these and assign it to people so they can make progress in their individual projects. I see. So let me ask just, if I'm a researcher, because I hear this all the time from researchers who are not affiliated with the big AI giants, that they can work on toy problems. It's frustrating. They can't work on the problems they would like to because they don't have the compute resources.

28:31So somebody like that applies for compute resources at NAR. What are they given? Are they given, you know, a cluster in the cloud that they can use? Are they given time at a data center, a physical data center that they plug into? I mean, how does that work? So when I talked about the first 35 projects, 27 of those projects were funded through NSF's resources and partner resources. Eight of them was funded through DOE, for example, and partner resources. So DOE gave access to their exascale computing systems, right? And so they're high-performance computing systems rather. And so depending on the project and the needs, you figure it out.

29:25You say this project needs high-performance computing resources or specifically GPU resources or specifically modern resources or a combination of them. So depending on that, we then pick up those projects. And so this committee of people who are looking at this are looking at what are the characteristics of the requests and how does that match up with what each one of us is able to do? That's how we are looking at this right now. We know what the needs out there are. We know what we have. It's a mapping process. But we also use this. We also use this to say, we know what the needs are. We know what we don't have yet or what we don't have yet at scale.

30:06Let's go to the appropriate partners and say, can you pledge these resources? And I have to say that I'm very impressed with the community. small startup companies to mid-sized companies to large-sized companies to agencies the sort of the open way in which they want to come and be part of this really is very energizing and is exciting okay and this portends a great future for us sure and clearly those who are nowhere near what the need out there is as you rightly point out you know there are people who are sort of tinkering with stuff and they feel that they are not able to do enough. Right. Right.

30:48And so we want to make sure they have enough of that. Yeah. And the pilot is, how large is the pilot? How many millions of dollars? So right now, you know, because of the fact that we put in$30 million and then there is this partner resources, I mean, I would safely say it's probably three times the hour investment, right? So the overall size of it. So that's where we are right now. But the way we are growing quickly, and it's going to be multiple more times of that. But as we get near resources allocated through new investments, the congressional investments appropriations are made, then this is bound to scale very rapidly.

31:32Right. And is there a goal? Is there a target that by 2030, you'll have a billion dollars worth of compute resources in there or 500 million or some, is there some number out there? Yeah. You know, Craig, this field is going screwing so fast that whatever number we put there, the needs will outpace. Yeah. So what we need to focus on is to say, are we configuring ourselves? You talked about a billion dollar number. Yes, it's got to be a billion dollar scale. That's not a question at all. And how much more is going to depend upon not only what we get from the federal government as investments, clearly, but that's why I'm taking this approach and we are taking this approach.