You’re a genuine hero, in the fact that your work is not only an actual plan and possibility space to upgrade democracies, but is also factoring in the race for AI itself, and what it will take to correct that. So my hope is that people will turn around this ...

Your work as an embodiment of foregrounding and making primary the vulnerabilities, the fragility of society so that we can care for that, and instead focus the incentives on the bridging, the health, the transparency, that the strengthening aspects of society.

So, I think part of what I want to instill, hopefully, at least my sort of take in this conversation is, and listeners, is that we need to create a broader sense of prudence and caution, and correct that 30-to-1 balance, from 30 times more people racing to increase the ...

The difficulty, I think, is the ambiguity of where those lines are, and the many different horizons of harm, and the different kinds of risks the range because some could argue that without the demand that Audrey Tang upgrade plan to democracies, the existing AI that we have proliferated, is ...

And in this, I think, you know, we deeply agree that we are currently racing towards a very dangerous, uncontrollable, dark outcome. And we agree that there needs to be some form of international cooperation, international agreements, and, you know, agreements between really any of the top labs that are ...

And even more so, given the Taiwanese government supports TSMC, more than, say the US government would support any US companies. So, do you see that what’s what’s the tension there, and I guess it’s just, I mean what Tristan’s is getting at, I think, is it’s a particularly unique ...

I think what we’re getting at with this question, is that Taiwan is at the fulcrum of the sort of geopolitical rivalry, in terms of the situation with China, and also the fact that it manufactures 90% of the world’s chips.

So, I’m actually not sure where I’m going with this question, because we just think it’s a fascinating and super important irony, that I’m not teeing this up effectively…

Let’s see I’m trying to find here. When Japan occupied French Indochina 1941, the US actually retaliated by freezing all Japanese assets in the States, preventing Japan from purchasing oil. And it was actually this that many historians think had caused Japan to decide to attack Pearl Harbor, hoping that ...

And, you know, this is interesting sort of moment where US-China tensions and the CHIPS Act and the US trying to prevent China from getting access to these chips. You can see from basically selling these most advanced chips to to China. And this is interesting parallel in history, which ...

Well, AI is that for cognitive labor, sort of paying humans to think to draw, to do the market research, to trade on the stock market, to illustrate something, to make a video. I can just take an AI chip, and it’ll do that instead.

You know, it has this metaphor that. Yeah, well, is is wonderful metaphor that you know what oil was like a barrel of oil to automating 25,000 human labor hours of physical labor, so you can take 25,000 humans and get them to do something physical labor for an hour, ...

What is the sort of upgrade plan what’s the takeaway that I can have about how I get this long suite of, you know, Audrey Tang upgrades, so that I have an AI resilient defense dominant digital democracy?

If I’m a policymaker from any of the democracies that are having elections this year, and I’m realizing that this may be the year of the last human elections, because of the impact of AI.

…but it’s, I mean, it is both ironic but, you know, to Audrey’s point, it’s using the technology in a defensive way, rather than the technical offensive way; bring in the 21st century technology to make sure that everyone sees the same thing at the same time. It creates a ...

There’s a grand irony in saying here’s the 21st century upgrade plan for 18th century democracies…

Can I get one of you guys to ask Audrey, a question please, about what are the other learnings for this for other democracies, like if you were going to go, and I know other leaders speak to you about what you’re doing in Taiwan.

Okay, I think this is great. I think it’s an excellent level of accessibility for people. I’m super excited that this is yeah, this way that brought it down.

So that watermarking won’t be a way to deal with it at the content level; we have to move to the actor or behavior level. That’s what you meant by shift left.

So I just want to make sure to clarify for listeners that when you said “shift left,” you don’t mean shift left politically, you’re talking about the ABC model, the actor/behavior/content, and where we used to say how do we defend on the content level — “this is a deep ...

So think about it we left the cats out of the bag, those are the small open source models, but if I like reverse engineer the cat it tells me how to unlock the lion from its cage. And that’s really dangerous. So, how do we tease apart, how we ...

We want to open source things. What’s confusing about AI is that it’s dual use it’s both offensive and defensive. And one of the things is that when you open source AI, it actually teaches you things about the really advanced closed source models I believe there’s papers on, you ...

How do we know who should have access to that? But we don’t just open source it. And actually, I just want to really highlight this for listeners, because as they hear other parts of our conversation, they might hear us disagreeing on open source, actually I don’t think that ...

The majority result would be that it would have an offensive, harmful effect on society, and the principle I’m hearing you speak to, Audrey, is just when we’re releasing something that is offensive, we should be in the conversation about preparing for the risk regulation.

But if you open source, let’s say gun designs, or open source bags of anthrax in every Walmart so anybody could buy a little bag of anthrax. They could do their own biology research, but that would be open sourcing in a way, or democratizing something that is offensive to ...

If you open source a mask, more people will have access to a defensive technology, so society gets more resilient, more robust, more protected.

So I actually just want to pause and make sure that listeners really get this, because it is such an important concept, is the technology that we’re open sourcing defensive or offensive?

I’d like you to do the best shot.

I don’t know if you agree with that distinction but I think I’m trying to figure out a way. I think that honestly if I’m really honest, I think listeners can’t follow this, I think this is like so over the tops of most people’s heads, because it’s unfortunately just ...

And then if we open source offense dominant things were open sourcing things that have more ability to take down and erode and destroy, then we have, we’re open sourcing things that have the ability to strengthen continue to maintain to grow to, etc.

And maybe, as you answer this Audrey, if you can introduce the idea of an AI, releasing AI that is offense dominant, and its effects versus defense dominant, because I think if people have that tool like what is an example of an offense dominant AI. And what makes, what ...

So I’m wondering like a simpler way to talk about what we think the philosophy of what we’re talking about is the from an abstract level like just trying to think of the way we want to do this, because basically what your claim is… Go ahead.

So I’m just self aware of how much jargon and technical detail people need to know, to understand everything that we’re talking about: Mixture of expert, “Mixtral”, 7 billion, 13 billion, it’s like I’m just aware that like most people aren’t going to get it.

Yeah, we can also rerecord that version of that question in a separate session. But, yeah, do you want to just take a stab and then we can… Audrey can you show our way through this?

Could you give your view about what open source, your views about the risks of open source and the benefits, and where you think that both aligns with our views and maybe disagrees, and how that relates to the broader race for artificial general intelligence?

And these frontier models from GPT-4 to GPT-5 and from Claude-2 to Claude-3, all the way to super intelligence that we have some shared concerns there.

In the previous episode Aza and I did a big episode on the dangers of open source AI and open source AI in your work has been key to actually strengthening democratic societies. But there’s also a difference between open source, AI development, and where you, I think, and I, ...

I think we should go to a place where it may look from the outside that we have a disagreement or divergent views which is around the question of open source AI.

And I think part of this episode is meant to point at this long list of solutions, these long list of projects that if you imagine every democracy in the world were to implement this whole suite, you get your kind of big upgrade from your vulnerable Windows 95 operating ...

And so there’s a whole sort of suite of tools that are open source tools that are part of this democracy upgrade plan that include, you know, the petition system, the deliberative polling, Polis, no reply buttons.

Yeah. And for those who aren’t tracking the community notes feature which is live on Twitter that Elon is using, it was actually built before Elon got there, but it was by another collaborator of Audrey’s, Colin, who helped build Polis.

What I love about Audrey’s work is she’s about changing the incentives, so that we get to the different outcome. And for those who aren’t tracking the community notes feature on X, or Twitter, was actually created by a parallel sort of project called Polis.

And per our frequent referencing of Charlie Munger’s quote, “if you show me the incentives, I’ll show you the outcome.”

And just to link this for listeners who know our work on social media, this is instead of rewarding “division entrepreneurs” who are identifying new creative ways to sow division and inflammation of cultural fault lines, this is rewarding the bridging and synthesis between these entrepreneurs.

Did I describe that correctly?

And then it’s no longer just citizens working to come up with policy, then it’s the stakeholders and the government and the experts working but beholden to the agenda set by people in that way.

And so it is that set of statements that bridge in a non-voting sense, just based on people’s values, that set the agenda that say this is what we, a set of the population, believe that becomes what legislators have to and stakeholders have to respond to.

And then the statements that are bridging are the ones not that are voted for, but are the ones that sit across many different tribes and those tribes are segmented based on what people’s actual statements are about what they believe.

And then instead of people voting, which has all sorts of ways that can be attacked, you describe a system where every positive statement of value that you make sort of tells the system more and more about who you are and which set of tribes you belong to.

For instance, there is no reply button in in your systems. And you’re like, OK, how do you have a conversation without a reply button? And instead it’s not that people are replying, it’s that when you see something you disagree with instead of getting into a flame war by ...