How to Improve Democracy? Audrey Tang and Justin Rosenstein on #InnoMinds S2EP4 (Part 2)

If you explained to an alien how the economic incentives of planet Earth work, they would be able to predict for you all of the different catastrophes that we're experiencing on planet Earth. That has not changed. Because the social media companies are still subject to that exact system, because they're still working within the profit maximization context, because they're still ultimately beholden to a board of directors that's beholden to shareholders and indeed have a legally binding fiduciary duty to maximize profit. And all these problems that, again, naturally arise from the fact that the incentive is to keep people staring at their phones. At the smallest level, although this is itself a massive change, is bringing a whole new design paradigm to the way that we're designing technology, in which the focus is not on these very simple metrics that are not aligned with what people actually care about, but to really, it's almost, it starts with a philosophical project to get really clear about what is meaningful, what is valuable, what is the actual service that we're trying to provide, what is the way in which we want people's lives to get better individually and collectively. Silicon Valley historically has very little in the way of tools to understand qualitative experience and the subtleties and nuance of phenomenology of our subjectivity.

And this idea of 'communing,' of the people who are using and being affected by some system being the ones who govern it and are in control of it, is such a critical concept. It's so pivotal to us moving as a society in a direction away from the incredibly inequitable racist, sexist problematic structures that we have today, in which you have this elite capitalist class that is in control of so much of the power and wealth in the world, to a world in which power and resources are equitably distributed, and communing, there was essentially a smear campaign in the 20th century against commoning. A famous essay about the tragedy of the commons that had the idea of, 'Oh, if you don't have the state or if you don't have market forces and private ownership, you're inevitably going to get the destruction of these resources.' And that's just not true. There was an intentional smear campaign that was perpetrated by the people who benefited from it. And Elinor Ostrom is an economist who won the Nobel Prize in economics who really looked into the actual history of commoning, which is that for basically the entirety of human history, it has been the most common form of the way that we manage resources and power. And it is highly, highly effective because you only get a tragedy of the commons in the cases where you have a bunch of people who aren't coordinating with each other. When you have coordination, when you have people creating rules norms and self-governing, then you're able to -- whether it's a forest or a fishery or the air -- you're able to effectively have democratic, participatory control of these common resources. And historically that's been done at small scale because that kind of coordination was only possible at small scale. But as we live in a more globalized society where we have more common resources like the atmosphere as a whole or internet commons that are not just at local scale, but at global scale, we fortunately now have the digital tools to also be able to coordinate and govern at global scale.

I actually now work full-time on a social venture, a nonprofit called One Project. I'm still involved with Asana as a board member and advisor, but primarily work on One Project, whose purpose is precisely to be working with communities in order to develop new models of economics and democracy that are more participatory, more effective, more fair and equitable, more in harmony with ecological regeneration. And it's for precisely this purpose that the dominant economic and political models, we believe, are just fundamentally broken and are essentially the operating system deep down there as a root cause, from which you can then understand all of the social ills that we're experiencing, as arising from this basic set of economic and political rules and incentives. The first step is to start by taking the existing experiments that are happening all over the world, sometimes experiments in new novel forms, in some cases just the continuation of pre-capitalist forms of much more participatory, fair systems of economics and governance, and bringing more power to them, helping them to become more resilient against the dominant system, and to become more scalable and use, in some cases, social power and people power, but in other cases, use digital technology to be able to take these more effective, equitable ecological systems and enable them to be more and more of a meaningful competitor to the dominant, hegemonic, capitalist market state system.

I think policy plays a critical role, and in a way, a very necessary role in order to contain the collective action problems. So, collective action problems, sometimes called multipolar traps, are situations in which, if you don't have coordination between actors, and each actor acts in its own self-interest, you get these effects that are bad for everyone. So, nuclear proliferation is an example. I think every country would prefer that there were not any nuclear weapons, but because there's no airtight way to ensure that no country has nuclear weapons, it's in each country's, not just interests, but they're almost required for national security to build that up. Or at least in America, there's so much sugar in food, and I think that's the result of any food company that doesn't invest in that, their food is not competitive in the market. And I think you see the same thing in technology, and I see this excuse over and over again, or this reason, where you'll have leaders at social media companies or AI companies that say, 'Yeah, we wish that we could make this less addictive.' 'We agree it would be safer if we were deploying AI at a slower rate.' 'But if we don't do it, someone else will.' And so in a lot of cases, those companies, people behind the scenes actually are asking for regulation because that will enable them to do the right thing without worrying that an unscrupulous competitor won't then just take advantage of them being an ethical actor. So, regulation is the tool that we so far today have, to be able to bind those kinds of traps. In terms of what that should look like, I think, one, we should be using the existing regulatory model, as broken as it is. Within AI, for example, you're seeing the EU AI Act, or for applying to both AI and social media, there's in America the Algorithmic Accountability Act. I would love to see these pieces of legislation actually move to be passed and be able to help bind these problems. At the same time, there are things that are fundamentally broken about the existing governance models, that they are quasi-democratic. You have, in America, this two-party system. It's very hard to get things done. There's data that shows that what the United States government is likely to do is highly correlated with what the wealthiest people in America want, and has extremely little correlation with what the average person in America wants. So ideally, what we would do is use the existing state regulatory powers to create laws that enshrine a new, more participatory system into law, where the people are more directly involved in defining the rules of these systems. Which is to say, like many of the experiments that Audrey has done, or there's various forms of new systems or very old systems of democracy that are more effective, more collaborative, engage more people's voices than the kind of 51% vote system that we've come to think of as democracy. If we could develop those participatory governance systems and essentially have companies report not to a board of directors, but report to the people, report to these democratic participatory governance processes, and then have regulation require companies to, at a minimum, take into account the voice of those participatory democratic processes. That would be a step in the direction of us moving away from everything about profit maximization and toward the people being in control of these systems that control their lives.

Audrey and I are both advisors to the Collective Intelligence Project that is using, at first, Polis, but is looking at using these kinds participatory democratic tools in order to get input from the people on what should be the direction of artificial intelligence. And OpenAI itself recently started an experiment on democratic inputs to AI that I advised on and I think it is a good start, although there's a lot further for them to go. They're starting with having people participate in deciding basic things about the political orientation or willingness of ChatGPT to say certain things or not say certain things. But ultimately, I hope that is a stepping stone toward a much more radical democratic experiment of allowing people to decide deployment schedules and when harms need to be considered and evaluated and who should be the beneficiaries, how should there be compensation to people who provide input. It's a whole host of deep fundamental questions about what is AI for, where should these incredible supercomputers and superintelligence be applied, that if they remain in the hands of the free-for-all, free market capitalist system, the very incentives that we already have will continue. And to the extent that we already see that companies use the intelligence and resources they have at their disposal to continue to hurt the environment or to strip mine our consciousness in order to stare at ads, artificial intelligence makes that incredibly accelerated. But if we can make sure that democracy is at the heart of deciding and participation and wisdom, a wise democratic process is at the heart of deciding what AI is used for, then AI could be this incredible benefit to humanity.

Yeah, there's a general problem. And I think in many ways, democracy over time has gotten a bad rep. Because when you think of the ways we do democracy today, generally, as 51% of a popular vote of a citizenry that has not had time to think carefully about the issues, or responding to charismatic candidates, then it feels like a populism that leads to very poor decisions. In order for democracy to be not just legitimate, but actually wise, you need systems that empower the participants to have the information that they need to have the real … and the time to think things through deeply. So, citizen councils are a great example of a democratic system where you take a representative sample of the population, and pay them, and give them the time and space to come and meet together in person. For potentially months at a time, they learn from the experts, learn deeply about the systems that they're going to be governing, deliberate, talk things through, and then come to decisions that are much wiser -- more thoughtful, more mindful -- than if they were just voting when they didn't understand. Now you're pointing out that in some cases even the experts don't understand, but we can understand at least, so people making those decisions can at least understand to the extent of the experts understand. Within AI, I think one of the most exciting things I've seen is Anthropic's model, Anthropic being one of the main competitors to OpenAI, of constitutional AI, where you essentially provide a document to the LLM, to the AI system that explains the values, that explains: 'You should do this and not this.' 'You should optimize for this.' 'You should always be good in these ways and avoid being evil in these ways.' And while it's not perfect, what I've seen in the research so far is that it's quite effective in shaping the behaviors and apparent attitudes of the systems. So who should be empowered to write that constitution? Ultimately, the people. Ultimately, the people who are going to be affected by these AI systems should be democratically defining the constitutions that then define the behavior of the AI systems.

Which is exciting, I'll say, and I'm curious what you think about that, it also is very concerning to me, the democratization of LLMs, and the fact that -- I knew that GPT-3.5 level was possible to run locally, this is news that you can even be close to GPT-4 level. Does it concern you that even if the main labs are doing their best to create good constitutions and have ethical models, that you could have unscrupulous actors all over the world, just taking those models and re-tuning them for more nefarious purposes?

The fact that we now live in a world in which we have these devices that are connected, where every single one of us can transmit information back and forth is incredible. And so, if you imagine that we were developing systems on top of that basic internet infrastructure, that were refined to be in service of -- not maximizing us clicking, not maximizing us staring -- but maximizing for our values. We could take each one of those values and consider, what is the potential? So a value, like 'connection' or 'community,' while social media gives you this fake version of that, where you quasi-feel like you're hanging out with your friends because you're seeing them doing fun things, real community comes from being together in person, from a depth of connection, and there's ways in which social media could be really helping people to come together in person. To the extent that we value knowledge and learning, social media, rather than amplifying the most radical, outrageous, click-baity voices, could be helping us to make sense together and learn together and amplify the things that are most important for us to learn. To the extent that we value a fair and wise society, social media could be helping us develop empathy for each other's perspectives. There's all these ways in which we could be really introspecting and doing the philosophical exercise and the democratic participatory exercise of what are the things we care about, and then that becomes the design exercise of how do we make these systems ever more continuously improved, to refine them to be in service of our deepest values. Ultimately, of course, that is a different incentive than what the companies currently have, which is to the things that will cause people to be more engaged are the things that maximize profit. And so, I think I'll just end by saying, because that profit maximization incentive is so the water we swim in, we are the fish in the water that is capitalism, it's hard to even imagine what a different world could look like. And yet, I love this quote from Ursula Le Guin: 'We live in capitalism. Its power seems inescapable, but so did the divine right of kings.' And that's not going to change overnight, but to the extent that we can collectively keep in our imaginations, and keep as our goal the dismantling of these existing power structures that serve a small number of people at the expense of the many, we are able to move more and more in a direction where if we can change the incentive structure, that will lead more naturally to people developing these technologies in alignment with the social good. But even prior to that, we can, as individual designers and creators, start to be more mindful about what are the decisions we're making, and be willing to make less money, in exchange for aligning with the things we actually care about in our hearts, and that we want for our society and for our children.

Audrey, you've had so much success bringing new forms of democratic participatory tools into Taiwan. I think there's people like myself all over the world who are so inspired by your work. Do you have a vision for how you'd like to see that work become more popular and expanded? Are you taking any steps in that direction, or are you focused on just make Taiwan systems as good as they can be, and allow other people to be inspired by that, and adopt what they see fit?

Yeah, I mentioned the citizen council model or citizen assembly, sortition as it's technically called, which you get a representative sample of the population together to deliberate. I think deliberation is such a critical -- I'm inspired by a story I've heard about the constitution editing process in Ireland, in which there's controversy about gay marriage. And if you had taken a popular vote of people in Ireland, the conclusion would have been that gay marriage would be illegal and unconstitutional. But there was a process of sortition of a group of people in Ireland coming together and discussing, and some of the people in that group were themselves gay. And in the course of deliberating, and in the course of not just intellectually deliberating, but getting to know each other as people, there was a sharing. In the course of getting to know each other as people, people's minds changed in that group. And by the end of the process, they decided that gay marriage would become legal in Ireland. That's an incredibly inspiring example of how people, not just voting and sort of yelling at each other on social media, but actually getting together in person, hearing each other's stories, hearing each other's perspectives can create incredible outcomes. Similarly, one of the projects that One Project has sponsored is called OF/BY/FOR ALL, where they did an experiment of bringing together a sample of people to discuss COVID. And it was an incredibly controversial topic. They did this right in the middle of the pandemic. And when they started, people were wildly divergent in their perspectives, but by the time they had gone through this multi-month facilitated process, there was an incredible amount of convergence and understanding. I think that this is a process that can be at the heart of new democratic models, that then can be assisted by much more modern technology, like artificial intelligence. AI presents this opportunity, where you could build a system that literally could talk to every single citizen in a country, or every single person in the world, and allow them in their native language in the way that they think to express their desires, their values, their needs, their wishes for the world, and to integrate all of that into a plan that accommodated as many people's desires as possible. The potential there for using AI to do that kind of participatory governance, even economic planning and resource distribution, is incredibly exciting.

Thank you, Audrey. My pleasure. I'm Justin Rosenstein. See you on TaiwanPlus.