Alex WellersteinVerified account

to be clear: I actually don't oppose funding this, but what drives me crazy is the combination of technological superiority and the utterly misguided model of how policymakers and operators could actually use this information

also this sort of thing has been around for decades and decades, and maybe the thing that's been holding us back has really been computing power, but on the other hand maybe it hasn't

I got to talk to a DARPA person a few years back; their big AI push at the time was building systems that could analyze academic publications in IR, etc., and draw inferable conclusions from them about how the world worked.

Either I or someone else asked: couldn't you just, like, fund academic work, and ask the academics what it all means? No, no, that's absurd. We need a computer program that can read the work, we can't actually just fund the work itself, or talk to people.

It's DARPA's mission to go far beyond what can be done today. Back in the day, I pitched a project to them. Sounds like they are totally taken with AI and maybe far too impressed with Silicon Valley. Anyhow, my project didn't get funded, but what they were doing made more sense.

I actually think the government should make high-risk bets. I'm not sure that the corpus of academic IR has much to say about how the world works.

The argument for a lot of the DARPA AI projects at the time was, "yes, we know that THIS problem isn't necessarily the best one to solve, but if we had an idea how to solve it, it would give us an approach that could be used for other problems." FWIW

The big AI problem (again, a couple years ago) for them was "context" — e.g., can you make an AI that can rapidly and more or less reliably construct context from scratch the way a human can. Under this heading a lot of odd stuff was being funded.