Birju: Let me invite in the broader scope here, because my understanding is that there seems to be a few different views held by those close to AI. And some voices, such as what I'm hearing here, talk about—I don't know actually what you're saying, but I've heard this view, I think espoused by the effective altruism movement, to slow down. And then there's others that talk about speeding up. And others that talk about the importance of not having just a few players in the game, and being able to expand who those players are.

And others talk about not doing that, and making sure that China doesn't win. I've heard people like the former CEO of Google talk about climate change being so late in the game that AI is our Hail Mary that could make a difference in a way that humans can't. So, how do you try to frame all of these views so that they don't obfuscate from how we as a society move forward on this?

Randima: Yeah, it's okay, there's a lot of pieces to this. I'll try to one-line, just give a few sentences on each one. The overall principle is matching power with responsibility. There's a principle there, right?

And that covers both the concentration side and the diffusion side. So concentration, what I mean is, for those listening, do you just want OpenAI and Anthropic and Google, a small number of companies to control everything, right? And NVIDIA. The other side is, let's make this technology available to everyone, right?

And so that's democratize everything. And of course, you have these obvious trade-offs with both, right? One of them, everything's kind of closed, you don't really know what's going on, there's not transparency. When they make mistakes, it's hard to be accountable, and there's regulatory capture.

That's one side. The side where everyone has everything, the problem is malicious actors get a huge advantage. And there's an inherent mismatch between offense and defense, where if you want to break into something, it's a lot easier than defending. Because when you attack, it doesn't take as much discipline.

You're like, okay, cool. And then, if the defense side finds a bug, it still takes weeks or months to fix the bug. So even if you have software that can find the bug on both sides, the offense just needs to use the bug to get in, right, to break into the system. The defense has to identify the bug and fix the bug.

And that is much, much harder. So there's always that asymmetry, right? That's just going to be that. And so people misrepresent that because they don't tell that story accurately about the need for implementing the bug fixes.

They always kind of forget that. Yeah, then there's a lot of complicated things, like U.S. versus China, right? The thing to remember here is the race, if you're saying there's a race, the race is to deploy the new technology.

In a way that's supportive and that does the best for your society, for your nation, and the world, right? So it's not how fast can you deploy it? It's how well can you deploy it? Just doing it really fast can actually have very harmful impact.

And we actually see this recognition a lot of times in China, where they have much more control. One of the things they just did recently was banning AI chatbots for relationships under 18, right? And that makes so much sense. Because there's so much harm to happen.

There's study after study that are showing this, and lots of real-world stories. There is a real loneliness. There's reasons why those things are helpful, also. I get that.

Overall, there's some real challenges, but they are very quick to say, hey, that's dumb. We're just gonna stop that, and then we're gonna deal with how to work around it. But the answer is not to just do everything as fast as you can. So that's another principle, right?

It just doesn't make sense. And then, on the speeding up, right? So the people who are really promoting acceleration. Their logic is something like, these systems are so complex, even when we try to do it right, we often get it wrong.

Just run it, and see what happens, and the things that break, we'll just fix it. We'll just outlaw the things that break, right? We'll just say, don't do this. The problem, again, is this kind of offense-defense asymmetry, where you end up creating problems that are very hard to fix after the fact.

And the second thing is that point that I mentioned about protecting systems we depend on. This is another thing they forget all the time, which is, if you break epistemology, meaning how we know it's true or not true, how we make sense of the world. If you break that, and then say, later we'll fix things. You need epistemology to fix things.

So if you've broken it, it's really hard to repair, right? And so, for example, you and I, if we're going to discuss a topic. We need to have some shared foundation of being able to agree on some facts, and then you say, well, actually, I see it this way, here's why, and then we trade some views, and hopefully we come to an understanding. If we can't agree on any of that basis.

We can't really discuss. So there's that. If people hate each other so much. That they can't even sit down and have a conversation.

The whole democracy thing, right, doesn't really work out.