Anybody attempting to emulate DARPA needs to be willing to sit with a lot of discomfort for a long time

Heuretics enabled by DARPA take a long time to get out into the world and the attribution is usually mixed because of DARPA’s role as a creator of fields and not as a commercializer of technology.

While there were intermediate products, the Mother of All Demos happened in 1968 - six years after J.C.R. Licklider joined DARPA and it took another five years after that to make its way into a commercial product in the Xerox Alto and arguably another ten years after that to actually have a massive impact in the Macintosh.

The DARPA grand challenge that arguably kicked off autonomous driving was in 2006 and of this writing in 2020, autonomous vehicles are still not mainstream.

Adding to the uncertainty is the fact that a big part of DARPA’s role is to get the ball rolling, not get it to the goal. This role is important but muddles attribution. You see this historically in how it seeded some of the best CS departments in the country and how often its money is used to derisk research enough that the NSF is willing to fund it. (DARPA provides a derisking role for people in other organizations.) Almost by design there will be arguments about who deserves credit for things DARPA was involved in. Any organization following that path will suffer the same problem.

On top of uncertainty, there is going to be discomfort because trying to emulate ARPA will entail working on weird shit. DARPA funds wacky things that go nowhere. Over time, DARPA directors have needed to fight to keep DARPA from being toned down and turned into a normal R&D org.

It is very easy to say “we accept failure” in theory but in practice it’s really uncomfortable.


Web URL for this note

Comment on this note