At the National Science Foundation (NSF), artificial intelligence has become a high priority agenda item. Recently, it appointed a new special assistant to the director to focus on AI.  The Federal Drive with Tom Temin  spoke with the new NSF AI Director, Tess deBlanc-Knowles.

Interview Transcript: 

Tom Temin All right, so tell us what this job is all about. You have another job already at NSF, and we’ll get into that. But now, your special assistant to the director for artificial intelligence. That sounds like AI matters a lot around there.

Tess deBlanc-Knowles It does. Yeah. And I think it’s important to start out with the point that NSF has long played a really significant role in sustaining AI research in the United States. NSF investments were instrumental in the development of machine learning, reinforcement learning approaches that are really at the core of today’s large language models. And today, NSF support for the field totals more than $800 million annually. And this funds innovation activities across 50 states, D.C., Puerto Rico. There’s support for things like research projects, institutes, fellowships and traineeship opportunities, education and research infrastructure. And we support the whole pipeline of innovation. So from foundational discoveries to the applied work of applying AI to other fields like agriculture, education, climate issues. And we also support the really critical work to advance trustworthy AI. And that it’s really important as increasingly, AI products are coming to market, there’s still a lot of work that needs to be done in the research space to advance the verifiability and reliability of AI systems. And then we also support researchers as they move their ideas to market, kind of along that full pipeline of translation, in terms of developing the idea, the business plan and then supporting startups. And then our programs also support some of these foundational elements that are critical for AI innovation ecosystem. So we play a pivotal role in supporting education and training, from k-through-12 education to making experiential learning opportunities available to advance education. And then we facilitate access to the research infrastructure. And really do this all at a really national scale, making sure that those opportunities are available for all Americans. So my role in this is focused on facilitating coordination across these many AI initiatives that we have underway, and then also ensuring timely implementation of NSF task in the recent AI executive order released by the administration. And it used to be the case that our AI funding was almost all focused through our Computer Science and Engineering directorate. But now you see that AI applications initiatives are being funded across our eight directorates. And I think this really reflects the importance of AI across areas of science and engineering. And that’s something that we want to continue to double down on.

        Learn how federal agencies are preparing to help agencies gear up for AI in our latest Executive Briefing, sponsored by ThunderCat Technology.

Tom Temin So your job then is to what, make sure that all of the eight directorates that have a piece of AI are operating in some sort of consonance with one another.

Tess deBlanc-Knowles That’s right. So we can take a foundation approach to all of these issues, because it’s really hard to kind of just separate one portion of AI when it really touches so many areas of our computer science or social behavioral engineering director or our engineering director, and kind of each of these areas of research have roles to play not only in the future of AI technology itself, but in the application of AI to really drive scientific discovery and innovation.

Tom Temin And there is so much talk about AI. And every day you hear about another deployment, either in industry or even in government of AI algorithms to do something or other. And so you might get the impression that it’s all done. It’s just a matter of deployment now and making sure you do it right. I guess maybe that’s not really the case. What are some of the research areas in particular, what’s yet to be developed and explored in AI?

Tess deBlanc-Knowles Yeah, and I think that’s an important point because, I think that we’re all so impressed by the capabilities of today’s generative AI systems. It’s supremely impressive on how far they’ve come and what they can do, but we still have a lot of work to do as a research community to advance some of these foundational capabilities of AI. So like perception, representation, learning and importantly, reasoning. And these are all the capabilities that are going to really expand our ability to harness AI for some of these kind of big potential application areas and uses that we see. And particularly, in how we can use AI systems kind of in tandem in collaboration with humans in the task that we want to take on. And I mentioned it before, but there’s also really critical work needed to make AI more reliable and transparent and really to provide assurances related to the performance of the model as well as its safety. And this is increasingly important as AI is being deployed to market, and we really need to advance our ability to assure model behavior, the line that behavior with human values and safety guardrails. And then we need reliable methods to continue to test and continue that assurance as a model learns as it develops and its deployed, and then importantly, integrated into AI systems of systems. As this becomes more complicated, this question of assuring behavior and trusting the behavior of the AI system becomes even more complicated in critical. And then just one last piece I’ll mention is really how critical I think it is to advanced the application of AI to other fields of science and engineering, and really to try to apply AI some big challenge problems like environmental sustainability, agriculture production, health care. There’s really a lot of untapped potential here that our research community can help move forward on.

Tom Temin We are speaking with Tess deBlanc-Knowles. She is newly appointed as special assistant to the director for artificial intelligence at the National Science Foundation. And you mentioned reasoning a moment ago as an area of research. And if you look at the large language models, really how they work, it’s basically highfalutin pattern recognition and reproduction. So they’re really more mimicry than reason. I mean, if you just take away all the fancy verbiage, what does reason and is that something possible in the field of AI? The same way traveling at the speed of light is for rocket ships?

Tess deBlanc-Knowles Yeah. And I think it’s still to be determined if we can develop an artificial intelligence system that can reason in the way that humans reason. But that’s one of the areas where it’s really critical for NSF investments. There’s no market incentive for private sector company to invest in this big question of, can we get these systems to reason? But that’s where NSF is investing across these kind of foundational areas of AI, kind of pushing the frontiers around. If we could get systems for reason, how would we go about doing it? And we’re kind of making those bets across different areas that we think that’s gonna lay the foundation for the next breakthrough that might not come from these big machine learning based large models.

Tom Temin And for those receiving such grants, I mean, I would think there need to be a lot of oversight to make sure that they’re not doing the artificial intelligence equivalent of Theranos, which never did really have the technology it said it would. People went to jail and they mimicked it and sort of made it look as if something was going on in this box when it really wasn’t. I would think that’s a big danger for the grant makers.

Tess deBlanc-Knowles Yeah. And NSF has a really well oiled mare review process. So even from the very beginning of the grant process, when researchers are putting their proposals to NSF, it goes through a really rigorous process in which each proposal is reviewed by a panel of experts from the community that kind of kicks the tires on do we even think that this is feasible? Is this a worthwhile technical path to go down? And then each of the program directors oversees the program if it is chosen to be awarded. We do a lot of work with the AI community around thinking through early in the project, some of the ethical implications and building that into some of the reporting processes as well.

        Read more: Artificial Intelligence

Tom Temin And what do you bring to this? You just came out of the white House, where you were a couple of years at Office of Science and Technology Policy. There seems to be a theme here in what you’ve done in the government.

Tess deBlanc-Knowles That’s right. Yeah. So I actually I started my career in national security, but moved into the AI space around five years ago through a position at the National Security Commission on AI. And so what I bring, particularly to the work that NSF does, is this kind of national picture of our policy environment, kind of some of the big levers that we’re moving towards as a country. I’m trying to integrate our efforts so that we are aligning with these big pushes in terms of what we need to do to make sure that we have this really strong national AI innovation environment, which is really the foundation for economic growth, for continuing to push the technology forward for national security capability. So I strongly believe that this is kind of the critical ingredients for our AI future. So I’m bringing that perspective, knowledge of the policy environment and helping integrate our activities. And so they can plug into a lot of these national efforts.

Tom Temin Yeah. This is really one of those fields that is not just commercial or just military. It’s not like developing a tank which has no commercial application. We hope we’d all like to drive one through a parking lot sometimes. But AI is really a blanket almost for every domain. Fair to say?

Tess deBlanc-Knowles Yes, I think that’s very fair to say. I think it’s one of these quintessential dual purpose technologies that the very essence of the technology can be used in so many different application environments, which I think is why it’s so exciting. That’s also why there are so many risks and considerations around AI, and how we safely integrate it into all of these different domains and applications. And I think that’s why folks are so excited about it. They’re also worried about it because it’s going to have this impact across scientific domains, application areas, all parts of life really.

Tom Temin And when you do a Google search and you get their newly generated AI large language model answer at the top of the screen, do you believe it?

Tess deBlanc-Knowles Sometimes, but I like to double check. I think, as with all of these large language models, it’s great to get inspiration from them, but always prudent to check answers.

Copyright © 2024 Federal News Network. All rights reserved. This website is not intended for users located within the European Economic Area.