We’ve been here before.

With the recent issues around the Navier-Stokes problem, it has never been clearer to me that we need regulation around the development and use of superscale AI, and we need it yesterday.

My background is physics. My professors at the university I attended were largely of an older-school, experimental physicist lean. Many of them were half-traumatized by their advisors, and many of their professors were fully-traumatized by WWII. Science was quite literally weaponized, and many of the contributors faced life-long struggles with the ethics and outcomes of their choices and works.

My undergrad training was filled with hardcore math, some entry experimentalism (see: undergrad), and a lot — like very much a lot — of ethics in science. There was a palpable ethos in our courses, lectures, seminars, and even in the offhand comments of our professors: “make sure you can live with the down-stream results of anything you choose to work on”.

I’ve always taken it seriously, so much so that I’ve taken it for granted. I thought all branches of science were as aware of the need to responsibly consider the outcomes of their research.

I was, in two words, unbelievably wrong. Or maybe naive. Probably both.

My excellent professors at one tiny, excellent, private university did not represent a real slice of scientific ethics, and especially not cross-field. And then I went and crossed fields, and I was simply not prepared for how different a world it is out here.

Computer science as a machine has not really slowed, at any point, long enough to consider the ethics of their products. Our products. Sometimes, there are mutterings about ethics, or murmurations about policy, but so often, the need for responsibility and care is shooed and scorned under the ever-mounting pressure of harder, better, faster, more, more, more.

Here’s an example.

How often in CS or tech are we hearing:
“if we don’t do it first, someone else will, so it might as well be us”?

Now where have we heard that before?

I’m not arguing for cessation. If AI is here, then it is here, and the new arms race is underway. But like the last go-around, this field needs safeguards, regulation, care, consideration, and above-all a purpose, and one that is outside profit margins. I’m certainly not saying the government can be expected to always make the best, most ethical choices with a touchy branch of research. But I am saying “move fast and break things” should emphatically not apply to superscale AI.

The idea of letting the characterized amoral tech-bro sell nuclear bombs to whomever can afford it is the literal premise of many a kitsch novel and movie. Look at the messes with social media — do we want those same morals in control of this generation’s atomic weapon? No? Because that is what we are doing. We are simply playing with fire and hoping the house doesn’t burn down.

We’ve been here before. We arguably got lucky that the world is in one piece after the last time around, and we were being orders of magnitude more cautious back then.

We can and must do this better.

Photograph (of J. Robert Oppenheimer) taken from the very excellent 1965 NBC News televised discussion of the trinity test and Manhattan project: The decision to drop the bomb. NBC-TV, executive producer : Irving Gitlin, 1965.