Tuesday, December 12, 2023
Context: In my role as division director of IIS, I’m sending out a short message to the IIS mailing list on the Second Tuesday Every Month (STEM). Here’s the installment for December 2023.
Before I came to NSF, I remember reading newspaper articles that talked about a crisis that was coming: AI is moving fast and necessitating changes in policy, but there are no AI experts in the government to guide this process. That sounded right to me. I could only think of one academic computer scientist who was working in government, and he was a computer security person. As an example of this kind of messaging, Eric Schmidt said of AI on NBC’s Meet the Press, “There’s no one in the government who could get it right.” Now, that quote happens to come from May 2023, which is after I got to NSF. (Ouch.)
At NSF, my predecessor was Henry Kautz, an influential member of the AI community and former AAAI President. Now, I’d be willing to bet that Henry’s AI knowledge exceeds that of Eric Schmidt. (Actually, with Henry now serving as a Senior Advisor for Schmidt Futures, my guess is that Eric would have to agree that Henry is someone “who could get it right”!)
Anyhow, my point is that I got to NSF thinking I’d be the lone AI expert in all of Washington. And that’s not true at all. I wanted to give you a sense of what I’ve seen, since it’s so different from my expectations coming in. Let me say up front, though, that my perspective is skewed. I know the people I work with closely, and have just a vague sense of the people further away from me.
Of course, there’s a lot of AI expertise in my division. All three of our clusters, Robust Intelligence, Information Integration and Informatics, and Human-Centered Computing, are staffed with program directors (both rotators and permanent staff), and fellows who have thought long and hard about AI research in its various forms. My deputy division director, Wendy, has been basting in this topic for years and years, and is broadly knowledgeable as well as being a world expert on topics at the boundary between AI and health.
The rest of the CISE directorate includes other folks with important expertise. The Computing and Communication Foundations division funds research on the mathematics and analysis of AI algorithms. The Computer and Network Systems division funds research on AI security as well as AI education. The Office of Advanced Cyberinfrastructure funds AI computing and data infrastructure (and is responsible for standing up the National AI Research Resource pilot!). As a result, all of the CISE divisions have people working on the front lines of AI. The deputy division director, JD, had previously been in IIS. The directorate head (until this past Friday… more on that in a later message), Margaret, was amazingly articulate on AI issues despite her background being in computer architecture.
Elsewhere at NSF, there’s AI work happening in all of the directorates, including TIP (AI partnerships), EDU (AI education), MPS (AI’s use in math and physical sciences), ENG (engineering), BIO (AI for analyzing genomic data), GEO (AI for studying the atmosphere), SBE (cognitive science), etc. The NSF director, Dr. Panch, has published AI papers (including in CVPR and KDD). I work closely on NSF-wide AI issues with Tess, who has read and mastered more about AI policy today than I have in my whole life. Nearly every unit at NSF includes advisory committees of people in the field (academia, but also industry) who lend their expertise to help make sure NSF stays on top of the latest developments.
Ok, so that’s just one agency and “science” is its middle name. What about elsewhere in the government? The White House’s Office of Science and Technology Policy (also named for “science”) is juggling a lot of AI topics. Wade is there, heading up the National AI Initiative Office. He has published extensively in statistical machine translation. DARPA’s Information Innovation Office (I2O) is headed up by Kathleen. She’s not an AI person per se (programming languages), but you’d never know it when she talks about recent advances and their implications---I learn something new about machine learning every time I talk to her.
One of my roles is serving as co-chair of the AI Research & Development Interagency Working Group. Through that venue, I get to interact with experts who are grappling with AI issues directly relevant to them: People at the National Institute of Standards and Technology (evaluating AI systems), the Department of Energy (AI for science), the State Department (diplomacy around AI), Department of Transportation (safety and automation), Department of Agriculture (AI support for farming), US Patent and Trademark Office (AI to support patent examiners), the US Consumer Product Safety Commission (evaluating AI products), Department of Justice (rules around the use of AI in policing), and the 25 other agencies on the committee all have unique challenges and have risen to meet them. I’d also be remiss if I didn’t mention the AAAS Rapid Response Cohort in AI Science & Technology Policy Fellows (https://www.aaas.org/news/stpf-ai-cohort), 6 intrepid scientists who dropped everything on short notice this Fall to come to DC to help advise lawmakers in AI.
So, at the end of the day, I don’t buy the claim that there’s “no one” in government who knows enough about AI to be a responsible steward. The folks I meet with every day are smart, informed, and passionate about the topic. That said, I had a meeting last week with a public sector group who pointed out to me the many ways that agencies (IRS was one) could be using ideas from AI to make things better for themselves and all of us. They are also right. There are clearly places in government that should be on top of AI but aren’t. On deeper reflection, I’d have to say that, in the US government, the present is already here, it’s just not evenly distributed. Something worth keeping in mind.
Oh, here’s puzzle I came up with last week, in case anyone is interested. I have a piece of furniture consisting of n drawers. Each drawer holds one item of clothing. I wash all n items of clothing and then put them away in a random order. When I put away item i, which goes in drawer i, I open drawer i and close all the drawers above it (to make sure I have access to drawer i). When I’m done putting away all n items, what’s the expected number of drawers that are open?
Until next year!
-Michael
Note: The unsubscribe button is fixed!