Moral Graphs: Interview with OpenAI Grant Winners! Meaning Alignment Institute: Aligning Humanity! | Artificial Intelligence Masterclass

Um, thank you for having us, David.

Um, our organization comes from almost a decade of research from one of our co-founders, Joe, um, who was early on in this humane technology way, realizing that we have all of these social systems optimizing for engagement, and it's leading to all kinds of bad societal outcomes.

So he went on a research question trying to understand what should we optimize for instead, um, and that's a very difficult question, um, that has some answers in philosophy and some answers in economics.

So he did all of this research, uh, originally intended for recommender systems, but we're now applying it to LLMs because it's a sort of continuous problem.

Um, this is also part of a larger, I would say, uh, social vision that Ellie's been working a lot on, um, which involves how you can think of, um, a, a post-AI world, um, what would give us meaning in such a world.

And, uh, a lot of our work is around values and meaning and understanding how you can interview people to find out what their values are and how you can build AI systems that are optimized for, uh, meaningful lives.

Wonderful.

Um, Ellie?

Yeah, great intro.

And yeah, thanks, Dave, also from my side.

Um, yeah, I guess I'll just add that we're, we're a bit of a, um, weird organization, uh, for our work in AI, 'cause actually we are, like, philosophers, and our quest is really, like, understanding what the good life is and what makes life, uh, beautiful and meaningful.

And, um, I guess our specialty is, like, we get, like, super nerdy on that, and that's where our, like, work on, like, values articulation comes from.

Um, and then it happens to be the case that this nerdiness around the good life is actually super useful, both in, like, how to design, um, AI systems that respect values and respect the good life, but not just for that.

It's also kind of like it helps envision a new, uh, way for arranging economic systems, like what would be an economy that was, like, really centered around meaning and values? And not just that, also, like, communities, but also kind of, like, more broadly, like, a new way of, like, understanding society and, like, how do we fit in it in a way that's kind of, like, really centered around meaning.

Excellent.

No, I, I love that, the, the intersection of economics and philosophy, 'cause that's kind of where I've, I've found [laughs] myself.

[laughs] Um-

Hmm

... just like when you, when you, when you confront the possibility or maybe even inevitability that AI is gonna change stuff, right? It really forces you to question-

Mm-hmm

... what do you care about? If it's gonna reshape stuff and we have some, some authority, some agency over the trajectory, where do we wanna go? If you can go anywhere, where do you go? That was actually the very first question that I asked myself when I was using GPT-2.

Hmm.

I said, "We have something that can think about anything," uh, obviously in its limited capacity, "so what do you want it to think about?" [laughs] If-

That's good

...

Um, thank you for having us, David.

Um, our organization comes from almost a decade of research from one of our co-founders, Joe, um, who was early on in this humane technology way, realizing that we have all of these social systems optimizing for engagement, and it's leading to all kinds of bad societal outcomes.

So he went on a research question trying to understand what should we optimize for instead, um, and that's a very difficult question, um, that has some answers in philosophy and some answers in economics.

So he did all of this research, uh, originally intended for recommender systems, but we're now applying it to LLMs because it's a sort of continuous problem.

Um, this is also part of a larger, I would say, uh, social vision that Ellie's been working a lot on, um, which involves how you can think of, um, a, a post-AI world, um, what would give us meaning in such a world.

And, uh, a lot of our work is around values and meaning and understanding how you can interview people to find out what their values are and how you can build AI systems that are optimized for, uh, meaningful lives.

Wonderful.

Um, Ellie?

Yeah, great intro.

And yeah, thanks, Dave, also from my side.

Um, yeah, I guess I'll just add that we're, we're a bit of a, um, weird organization, uh, for our work in AI, 'cause actually we are, like, philosophers, and our quest is really, like, understanding what the good life is and what makes life, uh, beautiful and meaningful.

And, um, I guess our specialty is, like, we get, like, super nerdy on that, and that's where our, like, work on, like, values articulation comes from.

Um, and then it happens to be the case that this nerdiness around the good life is actually super useful, both in, like, how to design, um, AI systems that respect values and respect the good life, but not just for that.

It's also kind of like it helps envision a new, uh, way for arranging economic systems, like what would be an economy that was, like, really centered around meaning and values? And not just that, also, like, communities, but also kind of, like, more broadly, like, a new way of, like, understanding society and, like, how do we fit in it in a way that's kind of, like, really centered around meaning.

Excellent.

No, I, I love that, the, the intersection of economics and philosophy, 'cause that's kind of where I've, I've found [laughs] myself.

[laughs] Um-

Hmm

... just like when you, when you, when you confront the possibility or maybe even inevitability that AI is gonna change stuff, right? It really forces you to question-

Mm-hmm

... what do you care about? If it's gonna reshape stuff and we have some, some authority, some agency over the trajectory, where do we wanna go? If you can go anywhere, where do you go? That was actually the very first question that I asked myself when I was using GPT-2.

Hmm.

I said, "We have something that can think about anything," uh, obviously in its limited capacity, "so what do you want it to think about?" [laughs] If-

That's good

...