Building an instrument other people get to use

Some of the most useful research infrastructure isn’t built by the people who eventually use it. Telescopes, particle accelerators, light sources, supercomputers, sensor networks: each one is a piece of capability that took years to design and stand up, and the science it enables comes from researchers who arrive after the work is done. The instrument is what the previous generation hands the next.

That framing is what I keep returning to when I look at this new NSF testbed grant. The award funds the design and construction of A National-Scale Testbed Supporting Artificial Intelligence Research Spanning the Computing Continuum. Pete Beckman is principal investigator, leading from Northwestern, and I’m a co-PI alongside Ilkay Altintas, Eugene Kelly, and Nicola Ferrier. The $3.2M is the line in the press release. The thing that matters is what the line buys.

What the line buys is a continuation. The testbed builds on years of work with the Array of Things, and the Sage cyberinfrastructure that grew out of it, into something larger and open to a much wider set of researchers. The point of opening it is straightforward. The kind of distributed AI research the award envisions, where edge sensors, regional infrastructure, and centralized supercomputers are treated as one continuum, is something you cannot do without scale. Most groups don’t have that scale. A shared testbed gives it to them.

Democratizing access is the part I keep coming back to. There are AI questions worth asking that come from researchers who don’t sit next door to a leadership-class facility, and they shouldn’t have to relocate to ask them. A national testbed is a way of saying: the instrument exists, you have a key, go.

Award details are at the NSF award page.