I gotta take issue with the premise here, that means-testing the poor is a necessary evil. In fact it's an inevitable dumpster fire in at least half a dozen different ways, _by design._

An AI "solution" whose inner workings are incomprehensible to even the most tech-savvy of humans is not a well-recommended fix. (Think: Air France flight 447: https://www.vanityfair.com/news/business/2014/10/air-france-flight-447-crash)

It's far more efficient and effective to simply deliver universal benefits and tax them back very progessively from top income recipients. Because _we already means-test the rich._ It all gets sorted out on your tax return every year.

The straightforward approach of Social Security, ACA health-insurance premiums, etc. are a direct result of that simple universality. And most recently the Child Tax Credit. (Infuriating that the Dems didn't pick up/adopt Romney's version which was hands-down the best one out there...)

The idea that we should fix effed-up programs by throwing cool (and insanely abstruse) tech at them, rather than...fixing the goddam programs, seems pretty starry-eyed to me.

Means-test the rich, not the poor. (Oh! We already do!)

Expand full comment

If we’re able to simplify government aid programs to the point where we don’t need some kind of assistive interface to it, then so much the better! Maybe we disagree over whether such programs really *can* be simplified (I am not an optimist on this point, people have very different values and what government in history has ever aligned on this type of thing?).

But I didn’t mean to get into an argument about whether government programs are complex! My thought was that so long as we have complexity (maybe it’s not necessary, maybe it’s just hard to get rid of), then an LLM is a great way for us as humans to interface with it.

Expand full comment
Oct 6, 2023Liked by Jason Preston

What an uplifting framework! From your public thinking to a generation of policymakers, please.

Expand full comment