S5E22: Lifeboats: Kim Polese

This is Larry Lessig. I met my guest Today first in 1998 when she spoke at the Berkman center for Internet and Society. Kim Palazzi had been an early employee at Sun Microsystems, where she was the founding product manager for Java. She then left Sun Microsystems and co founded Marimba, where she served as CEO until 2000 and chairman of the board until the company was sold to BMC Software. She then served as CEO of Spike Source, a company that sold software to automate open source application management. But the startup that we're talking about today, or maybe better celebrating today, is AI driven technology for understanding and and surfacing common understanding within a community. That idea me celebrating AI might strike you as odd, at least recognizing what I've said earlier in this series because so much of this season has been doomsplaining AI. But the key to recognize is that AI is good at what it's told to do. And if what it's told to do is good or valuable or edifying for our society, it can be very good at doing useful or valuable or edifying things for society. And the edifying thing that the technology behind Kim's current company is doing is finding a way to build common understanding or recognize common understanding within a diverse public. Or put better differently, it's a way to understand what we believe. And that's better than the standard way we understand what we believe today, which is polling. In this sense, this episode is a perfect complement to the last. In the last episode we saw how the idea of representative democracy is not exhausted by elections. There are other ways to craft representative democracy that don't depend on elections. In this episode we'll see how the idea of understanding what election a people think is not exhausted by the invention of George Gallup Public Opinion Poll. It is available through other technologies too, technologies that turn out to be better and fairer in understanding at least complicated ideas than a simple poll. This is a hopeful story today about technology in a space filled with doom. And again, I know this because my brand has been doom and I like this technology. Stay tuned. Kim, thanks so much for talking with us. You're building and have deployed an extraordinarily interesting application of AI technologies. Why don't you introduce us to what you've built built and where you think it can go?

Absolutely. Happy to, and thank you. Thanks Larry, for having me on the podcast. Maybe it might help to take a step back and just give you a little background on my own involvement in AI. In fact, my first job was at the first AI Company to go public, first job in Silicon Valley. And this was back in the era of symbolic computing, symbolic AI and an era called expert systems where we were actually encapsulating human knowledge into these AI systems. And they were powerful but on a very small scale and ultimately could not scale due to bandwidth, compute power and the fact that they relied on a small group of experts. But the kinds of problems we were solving were interesting. For example, reinventing the way factory floors worked, helping I was deployed out to NASA to help them implement these systems for navigation, a whole range of different use cases. But that that first era of AI was interesting because we were actually relying on human beings to understand how to make things work better at that company. Actually the CTO, Chief Technology Officer and ultimately the CEO Tom Taylor is now my co founder at CrowdSmart. So let me fast forward to CrowdSmart. The AI winter descended after the expert system era. I went off and did a variety of different things and then we came back together and co founded CrowdSmart a few years ago. CrowdSmart is actually using expert human expert knowledge, but in a way that instead of relying on a few experts and actually integrates an infinite number of people together, reasoning, deliberating, and the AI acts like a facility and ultimately helps those humans solve problems, make decisions, make predictions more accurately with less bias. And it does this by actually optimizing for diversity, cognitive diversity, the diversity of ultimately, very importantly, the group of people that are participating and the diversity of ideas. And so the system is actually presenting different people's ideas, asking open ended questions about whatever the decision they're trying to make or prediction or problem they're trying to solve, and then letting people respond with open ended answers, sentence or two. And the system very quickly is identifying different ideas, churning, using large language models actually to turn those responses into meaning for diversity of meaning. So as you're using the system, you'll see once you submit your ideas or your answers, you'll see seven other ideas. And if you can rank, upvote, rank those ideas, the ones you agree with, if you don't agree with any, you click skip, get another seven, you get another seven. And very quickly what the system is doing is learning where the points of alignment are. It's finding points of alignment, but it's constantly optimizing for diversity of ideas. So it's also seeking outlier ideas, it's making sure that every response gets an equal quote, unquote at bat, that everyone's idea has an equal chance to be presented and ranked. And as it's doing this Very rapidly. It's learning really what the best collective insights of that group is for whatever again, the decision they're trying to make or problem they're trying to solve. By the way, this is, identities are masked. So again, you're optimizing for cognitive diversity and reducing ingrained bias, which is one of the greatest challenges of course, to accurate decision making. We all have, we all carry ingrained bias with us is the facilitator.

So to be clear, because I think it's really important to emphasize this point, the system itself doesn't necessarily know, doesn't know anything about the thing it's going to be facilitating. It's just facilitating the facilitation of whatever the problem is you're trying to solve. So we could sit down and try to figure out what's the best way to deliver clean water in Rwanda, or we would thinking about, you know, what's the best way to balance the budget for the federal government. If we had a sufficiently diverse group of people, this size system would help surface the places where we have some kind of common ground. Is that a fair way to characterize it?

That's right. That's right. And you are optimizing for diversity of experience and knowledge. But at the same time, the people who are participating in this will have relevant insight and could be from any angle. Let me give you an example. We actually started the company crowdsmart with a problem we wanted to solve, which is if you're an entrepreneur and you're raising money, you need to know someone in the venture capital industry to raise your sort of founding capital. Otherwise you're out of luck. Which is kind of crazy when you think about all the talented entrepreneurs. All the problems need to be solved, all the capital, but it flows through this tiny relationship straw. So we thought, could we create a collective intelligence AI powered model of actually doing diligence so we could assemble groups of people who have expertise, relevant domain expertise for whatever the given company was, and then use the AI as a facilitator to identify the best, highest insights of those people, that group of people for each company. So we actually built the core technology, built a community of 2,500 people, raised a small fund because we knew we needed to be able to deploy funds to test the predictive accuracy and then track whether that combination of human and machine intelligence was better than typical VC returns. And we deployed capital to about 30 companies over a period of about 18 months. Then we watched what happened. And again, the way this worked is the group of people, usually 35 to 45 people would be presented with a particular startup company. They would get access to all the typical diligence materials. They could have live Q&As with the founders, all the stuff that usually goes on when a venture capital firm is evaluating a company, but in this case it was 35 to 45 at a time, it was all virtual, it was all over zoom, usually two to three weeks. So this is asynchronous as well. And they would participate in this open ended discussion about whether this company, what the odds were of a profitable return for this company. And it was drilling down on all the stuff that matters like team traction, product market fit and so forth. It turns out this approach was over 80% accurate at predicting which companies would go from seed stage funding to the Series A for a significant increase in valuation. And Interestingly also over 40% of the high scoring companies ended up being underrepresented founders leading these companies, which was a side effect of reducing ingrained bias. It's not that the teams doing the evaluations couldn't see these were underrepresented founders. It's that instead of the questions about where'd you go to school, who do you know, why are you here? Instead of the relationship being the focus, it was tell me about team, product, market, fit, traction, so forth. So based on the strength of that result, we realized there's something here that's, there's a very strong signal, there's something that's really working here. Typical venture returns are more like 20 to 30% versus over 80% accurate. That's an example of instrumenting a decision making process, optimizing diversity. For example, one of the companies company out of UC Berkeley called Cocoon Cam, using AI vision technology to detect sudden infant death syndrome, infant heart rate and breathing. We had a vision experts, we had professional investors in the medical device space, we had people who understand how to take a product like that into Walmart and big retailers. We had NICU nurses, parents. So each person had equally relevant but diverse perspective and knowledge about an insight about whether this product might work in the market and whether the company might actually succeed. So that's an example of instrumenting a reasoning process, a collective reasoning process, which is essentially what the AI was doing. Very rapidly it's doing sampling, that's what presenting seven reasons or seven ideas at a time is doing. Very rapidly it can identify what are the key points of alignment, ensure that the quiet voices are heard, the loud ones don't dominate, and identify the best, highest knowledge of that group about the given outcome. That they're seeking. Yeah. So the success of that, we decided to productize the core technology and make it possible to use for anything, not just startup invest.

Okay, so. But in the startup investment context, so you have a particular company time, from exposing your rankers or deliberators to the point at which you've come to a decision about it,

you can decide, do you want to do it in real time, do you want to do it over a few hours, days, weeks, months? It's asynchronous. It can be synchronous, but it works best when it's asynchronous, because one of the power, one of the powers of the benefits of this approach is people can think, respond, then see other people's ideas go away, think, change their mind, come back, disagree, and actually engage in a collective deliberative process around whatever the topic is. It can be as fast or slow as you want.

In the 80%, the greater than 80% return case, what was the amount of time that they were deliberating on those subjects?

Two to three weeks. Two to three weeks, yeah. Yeah.

And so it's asynchronous in the sense that you're just getting like a text message on your phone that sort of says, here's where we are, or what was the technology? Precisely.

It's an application. It's very intuitive and easy to interact with. It's basically, you're presented with an opportunity to engage. You click, you engage, and you're immediately provided with a question. Do you think, in this case, do you think this is a compelling investment opportunity? This is after you've seen the. The pitch from the founders. Do you see? Do you think this is a compelling investment opportunity? If so, yes. If yes, why? If no, why not? And that's, that's the initial prompt. You provide your input, again, a sentence or two, not a lot of, you know, effort required. Then once you do that, you're presented with other people's ideas seven at a time. And this is using a sampling algorithm. And those seven ideas or responses from others are optimized for diversity of thought. And you can rank those ideas. If you agree with any, you upvote, you rank them higher. If not, skip, get another seven. Again, that process repeats and very rapidly using this something called Bayesian learning. The system is understanding what the points of alignment are, but it's continually optimizing for diversity of thought. So it's doing these two things at once. The best human facilitators do, which best human facilitators are always Seeking points of alignment. Where are we agreeing? But what do we miss? What about the crazy ideas? What about the ideas that aren't yet popular? Make sure those get a chance to be heard. And ultimately after the system does this a few cycles, it's able to identify the, essentially the collective mind, a map of the collective mind of the group about that issue.

Okay, so in some sense it's perfectly aligned with what we imagine the best democratic systems are trying to accomplish. In that you've got a wide diversity of people who are participating. But the system is supposed to be identifying the places where we are aligned, but shaking them up a bit by making sure that we're considering the widest range of input in coming to that decision. So in a democracy we might have an initial view and then somebody comes along and says what about the worker? And they're oh yes, what about the worker? Indeed. And so then you think again about it like in, in the face of that. But after this process of deliberation, we should be able to know what we in fact have all coalesced around. And that's what this system is building.

That's exactly right. The reason why we have found this difficult to do, I mean, sort of broadly speaking, we haven't been able to do this without before large language models is it's really difficult to manage multiple lines of communication. And Brooks Law says that when you have about 100 people deliberating, that's 50,000 different lines of communication. That is almost impossible to make sense of unless you have large language models that can actually make meaning out of human language. And so what we've done, this is Tom Kaler's work, he's our chief scientist, co founder and this is his life's work, is bring together large language models and collective intelligence and use instead of using large language models to ingest existing text, historical existing text, which is what ChatGPT and these tools do, very powerful, but often can be inaccurate. Instead of using it on historical text, what we're doing is using large language models on conversation, human conversation, generating human conversation through open ended questions, letting people in fact discuss collectively, deliberate. And the large language models are turning that conversation into meaning and again optimizing for diversity of thought, which is how you get to greater accuracy, you reduce bias and you get to more accurate results.

Okay, so you did it originally in the context of this really dramatic example of venture capital investing and you showed that it was better than, than the standard efforts to do the same thing. Many people at that point would have said, wow, we've built a printing press that's printing money. Let's just grow the fund and just do this for the rest of time and we'll become the richest people in the world. You decided not to do that, which I know you, that's not surprising to me, but it might be surprising to others. But I don't want to understand your psychology. I want to understand what you think you've facilitated by opening it up. Like, where else now is this being used? Or where do you want it to be used and what are your dreams? Like, how do you think this changes the world?

So we realized this is a powerful technology and we could in fact become venture capitalists, make the technology proprietary, raise a big fund. But that was not interesting because the need for people to collectively problem solve is greater than ever. Not to state the obvious, we have complex great challenges, broadly speaking, that require human beings to deliberate effectively. And whether it's climate change or democracy or impact investing or, you know, scientific discovery, you know, the list goes on and on. The most important element in achieving breakthroughs is our ability to collectively reason to together to be able to identify our best, highest knowledge. And so this is a system that does that. We, we are particularly interested in using this. There are two sort of avenues. One is commercial. The other is, I call it social impact. The commercial applications are again really almost endless. Companies working together with their customers to innovate products, investors making investment decisions with less bias and greater accuracy, a very wide range of use cases. And in fact, CrowdSmart is engaging with a wide range of those kinds of use cases. At the same time, when we think about democracy and participatory democracy and community engagement, this is an incredibly powerful tool that really needs to be available at the local level, at ultimately any level, to help people problem solve in their own communities. We actually recently founded a nonprofit, 501C3 information called Common Good AI. And we're collaborating with a variety of different organizations, other companies, researchers, to engage in these initiatives and also to ensure that there is a deep reservoir of research that is developing around what really is third generation AI. And this is as defined by darpa, but it's human empowered AI. It's a combination of human machine intelligence and deliberative democracy, to me is one of the most important areas we need to focus third generation AI.

So, so you would, for example, imagine you had a community. I live in Brookline. We have Town Meeting in Brookline. That's 300 people who are elected to be members of Town Meeting. And it discusses problems which everybody here in Brookline is aware of. We can imagine a parallel process that used common good AI to address the same issue. Right? So let's say they have a question about what's the right way to what's the right school building project we should adopt? And you could imagine this then engaging with. There's no selection up front, right, about the representativeness of who's participating. It's a diversity of participants. And so how many would have to be participating? What's the way you would structure that decision? How do you imagine it would flow?

So you want a minimum of around a dozen, 15, 20, then the system starts to be predictably accurate. You want to have at least that many people because you want to have that cognitive diversity. You do want to optimize for diversity of thought. You can invite as many people as you want. So the system is scalable, infinitely scalable, essentially. It doesn't matter how many people. The more people you have, the more accurate it is. Ultimately a couple dozen, three, four dozen people. You've got enough of a sample size that it is consistently predictably accurate. The way you instrument the process is decide on the question. It's sort of the open ended question you're trying to answer. So if it's, you know, a new housing project, should we build whatever the project is in this area, what do you think? Yes or no, what's the why? It's always about the why behind what you think. Then again, first question you answer without seeing other people's input, but as soon as you do answer, then you start to see other people's input. Then you can start the ranking process, then you can drill down on specific questions, sort of sub questions. So should it be in a different neighborhood, should it be a different size, you know, all, and if so, you know, if yes, why? If not, if no, why not? So that process is very open ended, simple, intuitive process, and it's optimizing for diversity of thought. So you're constantly presented with ideas that may not agree with yours. The best outcomes are where people might have very divergent points of view coming in and they might not really fully agree on the perfect outcome, but they're able to align around an enlightened consensus, which to me is one of the most important elements of, well, problem solving and participatory democracy. And a functioning, healthy democracy is finding a solution that might not be everything precisely that you think should happen, but it is a consensus that is something you can live with and that in fact integrates new ideas. And that's when I Say enlightened consensus. It's a process that brings you to a new outcome that you might not have, you wouldn't have been able to come to without engaging with others and learning what they know and understanding why they think what they think. So that process of ultimately finding this enlightened consensus is what the system is doing. And that is to me, a core element of healthy democracy.

Okay, so when we say that, we get to an answer, what are the ways that we are confident it's the right answer? So, for example, with the housing project, we could imagine taking a vote at the beginning, should we build the housing project or not? And that vote could be 60% no and 40% yes. But at the end of this deliberation, you might have a completely different result. So what's the way in which as a decision maker, you say, actually the deliberation answer is the one I should embrace and the voting answer is the one I should reject? Because they obviously won't even most of the time be the same thing.

So. And one thing I did not mention, but it's important to say is this is both quantitative and also qualitative response. So you can score what you know you're. Do you agree with this? If so, on a scale of 1 to 100, what do you think? The system is also scoring responses quantitatively. And it's. And then you're providing the qualitative reasons why you think you think. And it's, the system is turning that those qualitative reasons into themes or points of alignment. So to answer your question, you could take a vote at the beginning, then people engage in this deliberative process. You take a vote at the end and you are generally going to be in a different place because there's been an exchange of ideas. The system has identified what people are actually lining up around. It's again ensuring the outlier ideas have been heard. And so you can at the end ultimately take another vote and see where people are. But the system is doing that in an integrated way throughout. So it is turning that conversation, that qualitative conversation, into a quantitative score, into ultimately a collective vote of the group, a map of the collective mind. So it's inherently doing what you set out to do at the beginning, which is take a vote.

So we've done in the context of deliberative and infrastructure that's facilitating face to face but virtual deliberation. We've done experiments around different topics. And for example, we've done a lot of deliberation around the question of the electoral college. And the deliberation around the Electoral College is very distinctive because we've got three solutions, but let's just put two on the table right here. One solution, which is basically national popular vote, and the second solution is fractional proportional vote. So you basically say if you get 43% of the vote in California, you get exactly 43% of the electoral vote in California. The first solution, national popular vote, is something that everybody's talked about, or at least if anybody knows what the Electoral college is, they've talked about. The idea of replacing it was just a national popular vote. And that solution is very partisan. Republicans don't like it and Democrats do like it. But the fractional proportional vote is a solution most people have not even thought about. And because they've not even thought about it, there's not really a strong partisan alignment. And in fact, by the end of the conversation, we find overwhelmingly everybody likes the idea, regardless of, of their partisan base. So deliberation doesn't really matter much in the first case because people are kind of committed to their partisan view, and at the end they're still committed. But in the second case, deliberation does a lot. So if we applied the same technology, we applied your technology to the same questions, is there any reason to believe you'd get over the partisan commitment, or would you be vulnerable to the same kind of part partisan bias that we saw in the context of our deliberation?

So this is an important element of the system, which is you as a participant have the opportunity to learn about whatever the background is on the topic. So in the case of startup investing, for example, all the participants got a chance to meet the founders, see the diligence materials, see the business plan and the pitch deck and the background profiles of the founders and so forth. That's important context for them to be able to then engage in that deliberative process. In this case, they would have the opportunity to learn about a different form of the Electoral College or a different application of the Electoral College. And upon learning about that, now they have a new insight and are able to come to the collective deliberation, the collective reasoning process, with that understanding and new knowledge. So that's an important part of this whole system is presenting people with important information and background and insights that might enable them to come into this collective reasoning process with a new perspective or a different perspective or deeper understanding.

Yeah, but I guess it's an empirical question whether in fact fact or it's an empirical psychological question whether in fact they hear the other side or have they even internalized. You know, one of the things psychology teaches us is especially smart people are really good at working around contrary evidence. Like, so, you know, some people, like, look at climate change and they think, oh, it's the ignorant people who don't believe in climate change. No, that's not true. It's in fact smart people, people who are really good at, like, hearing the data, hearing the evidence, and then saying, actually, this is probably because X, Y and Z. So I'm not going to believe the evidence even still. So I imagine that, you know, once you've done 100,000 of these in many different contexts, we'll have a better sense of its ability to move people beyond their partisan priors. But there's no reason, I guess, to believe that we have certain confidence that it would move them beyond their partisan bias. Or do you have that confidence?

An important element of this is the question why? So why do you think this? And in this process of collective reasoning, people express their. That background around their reasoning behind their belief that effort can actually move the needle. Because I actually, as a participant, happen to have insight on that. I thought, in fact, like you, I thought xyz, but because of my experience doing abc, I, in fact learned that's not true. And here's why. So that's what people do in real life, and that's how a conversation can lead to an outcome that's a productive one. So that's what the system is constantly probing for, is why? Why do you think that? And when you're presented with the reasoning behind someone's idea, so they're not just. It's not about your identity, it's not about, you know, clinging to some political party, but instead it is your experience and what you've learned out there in the real world about this particular topic and how you were moved from one point of view to another point of view, when you share that, that's when minds can change, when people feel it's safe to disagree and say, look, I don't agree with you, and here's why. Suddenly, breakthroughs can happen. So it's not just mudslinging, but it's actually understanding why somebody thinks something. That's what the system is constantly doing and probing for.

But technically, they're free to express that view because they're not in any direct way identified. Right? So if we imagine the most polarizing, partisan type of debates that exist in American politics today, like, is Donald Trump guilty of some crime? You could express your views and not fear the other side punishing you. For those views. But I wonder whether you've run experiments where it's questions like that that are being deliberated on. And have we seen movement or people kind of just digging in their heels because they have lots of ways of rationalizing the counter evidence.

So people do dig in their heels. But this is, what's important, is that the people who don't just dig in their heels, but actually open their minds and say, well, you know, I thought that. And there might be a different way of looking at that point. Here's another angle. That perspective, that person's perspective might attract a lot of upvotes or a higher rank because others participating in this collective reasoning process see that someone's not just digging their heels in, they're actually thinking this through, or they've had an experience or an insight that changed the them and they're sharing what changed their mind about this. That idea then might start to get momentum and that rises in rank in the system in a way that someone digging their heels in and very clearly digging their heels in, their ideas won't rise. And so the system, the AI facilitator, is on the back end, creating a whole map of who's most influential and as ranked by other people's votes. And that system will not keep teeing up someone who's just digging their heels in because they're, you know, they're sort of stuck on an idea. It will, it will rank the ideas that are voted up by everyone else. And those ideas are more likely to be someone who's not digging their heels in, but who's actually thinking in a way that. Or expressing, addressing an idea that is richer, deeper, provides context and again, the why behind why you think something.

So give us a sense of how many of these deliberations are happening and what's the range of them. And again, I'm not interested in the commercial context, which is obviously itself very interesting, but in the more public facing context that you're committed and open in. I'm sorry, a common good.

Common good AI.

Yeah.

So we're just launching several projects right now in the process of getting them off the ground. One is in the city of Cincinnati and an MD there at Children's Hospital. Grew up in the neighborhoods in, you know, was in the life himself on the streets dealing with gun violence. Went off and got an md, came back, became an ER doctor at Children's Hospital and was finding himself sewing these kids up at night. Then they go off into the war zone the next day in the streets. And he realized this is a system level problem. We need to get everybody involved in this who has a perspective on the problem, what are the root causes of gun violence in Cincinnati? What are some of the corrective actions we could take? So this is just, again, we're raising money right now to get this off the ground. But it is a combination of community leaders, kids who are actually in the life, members of gangs, parole officers, social workers, doctors, a whole range of people, each of whom has an equally relevant but important perspective. So that's one initiative. The second one is a small northeastern town, not too far from where you are, a logging town historically, and there's deep red and deep blue. And they now are dealing with a situation where some very wealthy investors have come in and bought the land around this town. And they all want to preserve their forest, but for different reasons. If you're, you know, a hunter, you want to make sure that forest is available. If you're an environment environmentalists, you care about carbon capture. Whatever your reasons are, however, you share a common goal. So that's an initiative that is also in the process of getting off the ground. There's a third initiative right now at Children's Hospital, which is focused on health equity and getting the entire employee base of Children's Hospital in Cincinnati to participate in a collective reasoning session around how we increase services available to everyone in the community, how we provide a more equitable health service offering to the entire city. So these are multiple different projects. There are others that are in the process of getting funded. So we're right at the beginning of launching these initiatives. But there's tremendous interest in engaging people in expressing their understanding, knowledge, experience about whatever the topic is they're trying to solve or address or resolve. All of these initiatives have a common theme, which is everyone's coming at the problem from a different perspective, but each person's voice is equally important and everybody's voice will be heard. So we're very excited about the potential to apply this to community engagement, participatory democracy at both the local level and beyond.

Imagine you were imagine town meeting of Brookline said, okay, we want to try this for a question about where to place a school or whether to change the form of government to become a mayor, system, whatever. And they wanted to come to you and engage you to do it. What would they need to bring to you? I mean, obviously they need to raise resources to make it possible, but what kind of resources and what would you give them to make this engagement possible?

So they would need to identify who's participating in this process could be the entire community. Again, no limit on the number of people who can participate. They would need to decide on what the question is they're trying to answer or the decision they're trying to make. So that's pretty simple. There would need to be some set of background materials or, you know, information that would be relevant for everyone to have access to, to get a little informed about the issue, the topic, and that's it. I mean, people question background material and that's it. And the system is, once it starts, is acting like the best human facilitator and enabling people to engage in that deliberative, collective reasoning process. It's pretty simple.

So the materials could be like a YouTube video, like that sort of lays the issue out.

That's right.

So they have the YouTube video. They have a list of 500 people from the community who say they want to participate in this. Could actually be town meeting itself. Town meeting could say it wants to do its deliberation through this. Is there a website right now where you. Or is there an app right now that they would come to you and say, we want to do this, and you say, okay, we'll set it up and you'll do it? Or is it self generated or how exactly technically would it work?

Yeah, it's seamless, essentially. The technology enables you to click a button and be instantly in the app. So the product exists and is available for them in this case to use. They would integrate it into whatever website or application they already have, or just have it used as a standalone application. So it can be both embedded and also exist as a standalone app, whatever form they preferred. But that's essentially, that's it. There's no elaborate setup, there's no kind of set of hurdles they have to get through to use the technology.

So is there a place right now people can go and see examples or are you still standing up the public facing part of this?

Yeah, we're in the process of getting the public facing part of common good AI set up on CrowdSmart. You can learn about the technology and that's again set commercial applications. But the combination of crowdsmart and Common Good AI will give you a background on what this is, how it works and how to get involved.

And so this podcast, I'm not sure exactly when this episode comes out, but it's the. This year, do you imagine, like in 2024, we're going to have a bunch of different communities struggling or working through local issues on Common Good AI?

I'm very hopeful that that will be the case, and that's certainly our intent and what's happening is that we're being approached by a number of different organizations, communities that are learning about this, hearing about it and want to use it. So right now we want make sure we've got the structure, the resources, the foundation to be able to deliver on doing that, making it available. And that is exactly why we founded Common Good AI. We realized this is more than one company. The potential is immense, the need is huge. The urgency is there because we need better tools. Social media media has failed us in many ways and in many ways been destructive to a healthy functioning democracy. We need new tools. And so the answer to the question is absolutely. That is really what's fueling us is the belief that the world needs this, we need to get it out, we need to make it available to communities. Healthy democracies need new tools. And we actually now can use large language models, AI in completely new ways. It was not possible before. And so we're 100% committed to making that happen.

Okay, so let's imagine, let's stand up four models for deliberating about an issue and again, keeping at the town meeting level. So we start with the one that exists right now, which is a meeting where people show up and people stand up and they make their arguments. That's, that's one thing. A second would be what Jim Fishkin describes, which is to say, let's take a random representative selection of the public that's bring them together into a space, that's give them the information they need to deliberate and then they deliberate. A third would be the version of Jim Fishkin's work that we've been developing, which is to allow virtual deliberation. It shares with Fishkin the idea that people are going to be deliberating face to face. But it suffers the same weakness from your perspective the fishkins would present, which is that it's like, you know, all the internal biases that we bring to deliberations would be present even if you're doing it virtually versus in real space. But there's some advantages. I think one of the advantages we hadn't thought about, but became really obvious was the safety sense. Like, there's a lot of people who would not be eager to like show up on a Sunday night at a basement of some church to deliberate about some issue because they don't know who's there and they don't themselves feel confident or safe to go to that place. But if all you had to say is like just, you know, log on it'll be like a zoom platform and if you don't like it, you can turn it off. But it's in your own house and nobody's going to stalk you or create any risk for you if you do that, that's probably an advantage. Then the fourth example is yours. You have the same materials, but everybody is working from an app and it's asynchronous and you have time to reflect and think about it. I take it you would say the advantage of the app is that you're neutralizing all these implicit biases that we know people bring to deliberation about that. I imagine it's also much cheaper in the sense of people's time. I mean, synchronous time is extremely costly. It's really hard to bring the right kind of people together at any particular time. But if you could do it over the course of three weeks, we all can find the moments when we need to to sort of log in and do it. So it would lower the cost of time. And I imagine it also just lowers is the actual cost, I mean the cost of running the deliberation. I mean, I imagine you're using a lot of electricity to make the AI's work, but it's actually objectively a cheaper process. Is that a fair summary of that?

That is fair. And there's one other important point, which is that you can't promote your idea, your own idea. The system doesn't allow you to do that. Other people can promote your idea, but you can't promote your idea. So that means that you're eliminating one of the biggest challenges to in person deliberation. And, you know, sort of the traditional social media approaches too. The, the algorithm is deliberately developed so that you cannot put your idea out there again and again and again. And the system again is enabling others to do that, but not you. So that's an important point that I think is really critical to both the accuracy of the system and also the reduction of kind of the bias that comes from people who are very effective at promoting their own ideas. One other thing to mention, and that is that these systems don't take a ton of energy because what you don't have to do is train it on, train the system on massive amounts of text. Instead, the system is working adaptively, learning from evidence and doing sampling. And that process does not take tons of co compute power. But that's just a side point.

So, Kim, it's extremely exciting in some way. As I mentioned in the introduction, you've had this incredible career of always kind of showing up at the most exciting inflection points of arc of technical development. And here you are again. And you know, as someone who's deeply committed to figuring out what's the best way to revive a concept of democracy that we all could be excited about, this is extremely exciting project to begin to play with and experiment with. And so I'm going to be an adopter, early adopter, frequent player to try to push it into different contexts and let's see where it goes. But it's an incredibly hopeful example of AI Very different from the dark stories that we've been talking about so far.

Yep, indeed. And thank you, Larry. It's really honored to be here and to be able to share with you and others what we're working on. And we'd love to collaborate with any and all who want to be part of this.

So people who want to follow you, is there a website yet or a

way that they can common goodai.org okay, great. Common goodai.org and then the company is Crowdsmart AI.

Yep.

Yep. Fantastic. Kim, thank you so much for your time.

Thank you, Larry.

This has been the 22nd episode of the Season 5 of the podcast Another Way. These podcasts are produced by equal citizens in an abstract sense. They are produced literally by Josh Elstrow of Elstro Productions. You can find out more about Equal Citizens at equalcitizens. Us. Give us your thoughts, your feedback. Share this podcast there as broadly as you can and if you can, donate to help us keep this work going. What I do, I do pro bono, which is fancy lawyer talk for without pay. But we have a team that needs pay to keep them going, to keep equal citizens alive. So check the donation button at equalcitizens.us. thanks again and stay tuned. Sa.