Effective Altruism and A.I.

Okay, so, um, recently [sighs] there's been, uh, warnings about AI, uh, potentially, uh, wiping out humanity by the end of the decade.

Um, and, uh, there's, there was a call to, to set some sort of regulations in place to, uh, uh, get, you know, to prevent this from happening.

And at a White House meeting a few weeks ago, they, uh, decided on a voluntary, uh, four-part system that is supposed to, uh, prevent AI from getting out of control.

Okay.

Now, a lot of this really revolves around ethics.

They talk about alignment.

Um, and in the AI world, what alignment means is making sure that the AI is aligned with the ethical standards of the company.

Okay, so that, that's the immediate situation.

Now, what we need to look at is what exactly ethics means in this context.

Um, I did a breakpoint, uh, actually the week that we're, um, recording this, I, I had a breakpoint out, uh, commentary in which, uh, I pointed out that no matter how tight your ethical standards are, they don't work.

Um, Enron, for example, passed every compliance thing in the book.

They had Lehman Brothers, one of the, the, the Big Five, uh, accounting firms saying that everything was fine, and they collapsed, um, because they found loopholes and found ways around the problems.

Okay, so the question becomes, what is the ethical framework that they're using to try to align the AI with? And that is a thing called effective altruism.

This is really the big ethical theory that motivates a lot of your tech people and most of your AI companies.

And along with that is also th- there's a supposedly independent board, although it's got a lot of overlap in personnel, uh, with Anthropic, um, th- that is also basically an effective altruism, uh, board.

So what is effective altruism? Effective altruism basically says that we should view our, um, our ethical activities in terms of, uh, return on investment.

That is to say, for every dollar that you spend, you wanna get the biggest bang for your buck, okay? So it is constantly looking for ways to help the world, uh, efficiently, and efficiency is sort of the, the key goal here.

Um, now all of that sounds really good, but the question is, how do you measure things like that? There are several questions.

Let's start with how you measure them.

The goal is to quantify everything.

So you wanna take into account all of the various, uh, uh, benefits, uh, that might accrue, uh, quantify them, uh, and then multiply that by the probability of success, and then that gives you a fixed number, and the higher the number, the better, uh, you are, uh, at pursuing that particular, uh, project.

So that's, that's the general theory.

Now, it seems to me that we've got a number of kind of practical issues to start off with here.

Um, the first of them is, well, let's just start with how do you quantify things? Okay.

And in the effective altruism world, uh, what they are looking at-- I mean, they, they use things like, um, uh, how much is it gonna prolong life? You know, things like that, numbers along those lines.