Behind The Filter Bubble

Everyone's got a hard question about AI, your job, your kids, where it's going. Anthropic was built to surface those questions and share what it finds along the way because there's hope in hard questions. Ask yours at clod.ai/spotify and keep thinking. That's clod.ai/spotify. Find rentals here for Mint Mobile, the message for everyone paying big wireless way too much. Please for the love of everything goodness world, stop. With Mint you can get premium wireless for just $15 a month. Of course if you enjoy overpaying no judgements but that's weird. Okay, one judgement. Anyway, give it a try at MintMobile.com/switch. Up front payment of $45 for three month plan, equivalent to $15 per month required. Intro rate for three months only. Then full price plan options available. Taxes and fees extra. Simple turns at MintMobile.com. Afternoon glare totally ruined my watch party for the big game. That's when my buddy recommended Blinds.com. With custom window treatments made to fit your style, space, and budget, customers who shop at Blinds.com experienced overwhelming satisfaction. Side effects may include a fuller, deeper night sleep and a sudden urge to host the watch party again. Blinds.com isn't a medication, but it can improve how you feel about your home. Visit Blinds.com to order your free samples today. When you spend enough time around great companies, you notice that the best ones treat trust as a growth strategy. Not a box to check. Vanta helps companies build and prove trust from day one. Its agentic trust platform connects compliance, risk, and trust at enterprise scale. Vanta delivers a 526% return on investment over three years and pays for itself in just three months. We learn why over 16,000 of the fastest growing companies use Vanta, Advanta.com. By the time I hit my 50s, I'd learned a few things, like how family is precious. Work can always wait, and 99% of people over 50 already have the virus that causes shingles. Not everyone at risk will develop it, but I did. The painful blistering rash disrupted my life for weeks. Don't learn about your shingles risk the hard way. Talk to your doctor or pharmacist today, sponsored by GSK. Open source code of any kind is an invitation to adversaries to understand how to gain the algorithm for their own advantages. Having transparency in the algorithm, the algorithm is only part of the story. Part of the story is also what are the content moderation guidelines that the platform follows, and that can open itself up to all kinds of issues around interpretation and where the platform wants to be, and you have to understand that also has second order effects around who actually stays on the platform. Welcome to Politicology, I'm Ron Steslow. Shortly after Elon Musk announced his hostile takeover bid for Twitter, he said one of his goals in the takeover would be to make Twitter's algorithm that recommends whether a tweet gets promoted or demoted, open source. Musk said the move would reduce the risk of "behind the scenes manipulation" in the form of shadow bands that Republicans have said is a result of Twitter's political bias. And I really wanted to dive into what algorithmic transparency is and what it could mean for Twitter users and for social media platforms more broadly. So today I am pleased to be joined by two of the authors of System Error, where Big Tech went wrong in how we can reboot. Returning to the show, Miran Sahami, Miran is the James and Eleanor Chisbro professor in the School of Engineering and Professor and Associate Chair for Education in the Computer Science Department at Stanford University. Prior to joining the Stanford faculty, he was a senior research scientist at Google. He was appointed by California Governor Jerry Brown to the state's Computer Science Strategic Implementation Plan Advisory Panel, and he is one of the inventors of email spam filtering technology. Miran, thanks for making the time again. Thanks for having me, Miran, good to be here. And making his Policology debut, Rob Reesh. Rob is a philosopher and a professor of political science and, by courtesy, professor of philosophy at the Graduate School of Education at Stanford University. He's a co-director of Stanford Center for Ethics in Society and the Associate Director of the Institute for Human-Centered Artificial Intelligence. Rob, it's great to meet you and welcome to Policology. I'm really glad to be here, thanks for having me. Support for Policology comes from Quince, which has quickly become my go-to place for the fashion staples I always find myself reaching for first. Quince focuses on high-quality wardrobe staples made with premium materials, like 100% Mongolian cashmere, organic cotton, and marino wool, all in timeless, versatile styles designed to last. Their Mongolian cashmere sweaters start at just $60, and they are all the softness and quality you'd expect from a luxury brand, but without the luxury price tag. My current favorite is the Mongolian cashmere Johnny Collar Polo Sweater in Navy, which has been perfect for this in-between weather, and everything at Quince is priced 50-80% less than similar brands. They work directly with ethical factories and cut out the middlemen. So you're paying for high quality, not the brand market. Even my highly-choosey husband has become a fan, because the quality of their products is more like what he's used to finding in European boutiques. I recently ordered him one of their excursion Italian leather sling bags as a gift, only to find out he had actually already ordered one, so now we match. Find your next fall favorites at Quince. Download the Quince app for app exclusive offers, or go to Quince.com. Get free shipping on your order and 365 day returns. Now available in Canada and the UK, too, and when Quince asks where you heard about him, let them know it was politicology. It's the best way to support the show. Port for politicology comes from AG1. October has a way of sneaking up on us. Shorter days, darker mornings, and a routine that suddenly feels off. You're now up before the sun with a full day ahead, and that's when AG1 are in this place. Your body's already working hard to adjust to the season. The last thing it needs is a nutrient gap, making it harder. I'm headed out on a long road trip this month, which means my diet is going to be, shall we say, less than optimal. But AG1 next gen has 75-plus ingredients, vitamins, minerals, pre- and probiotics, superfoods, backed by four separate gold standard clinical trials. So wherever we wake up, I know that with one scoop, cold water, every morning, before anything else, my bases are covered. Go to drinkag1.com/ron, and for a limited time, get a free immune support bundle, including a bottle of omega-3 fish oil, and a bottle of vitamin D3K2, with your first subscription order of AG1 next gen, or AG1 pro. That's drinkag1.com/ron. This podcast is supported by Anthropic, the public-benefit corporation behind Claude. Everyone's got a hard question about AI. Anthropic was built to surface those questions and share what it finds along the way, because there's hope in hard questions. Mayeron, why don't we just begin with a 30,000-foot view? How do algorithms determine what a user sees on a social media platform? What are they? And what does someone like my mom need to understand about them? Well, algorithms are basically just the code that runs underneath the hood for a social network or most other sites. And part of what it does, it needs to determine what to show you. So when you say log in to Twitter, what tweets are you going to see from the people that you're following? And so oftentimes what those algorithms do is they do things like look at who's tweets you've looked at in the past, who's tweets you may have engaged with, say to retweet or to like. And so based on that, it can build a profile of you to understand what kinds of things you like and who you tend to interact with. And so that algorithm is making decisions as to what you should see oftentimes by trying to maximize your engagement or show you things that are more likely to have you engage with them or react somehow. So an algorithm is a set of rules. Can you help us understand how they function vis-à-vis machine learning? So machine learning is a technology that basically learns from data. The idea is that the more data system has, as you're interacting with it, say, as it sees that you're clicking on various things or following particular people, it can learn from that data to understand what are the things, for example, you're most likely to click on in the future. And so when there's new piece of content to potentially show you, it can give a score to each one as to say you're likely to have engaging with it and then show you the things that you're most likely to engage with. And so really these algorithms are drawn or driven by the kinds of data that's available in the system, the sort of interactions that you have that create that data and how much sophistication they want to have in terms of the mathematics underneath underlying it to determine what those recommendations are. So what information would people be able to glean from the algorithms without having access to the underlying data Twitter uses to train those algorithms? What are the kinds of things that someone might be able to understand if they were told more information about what goes into determining the content that they see? Well, they could understand some of the factors that make a particular piece of information rank highly. So for example, how much is prior engagement weighted? So how, for example, is a particular click or a like on a tweet actually weighted relative to say. retweeting something relative to commenting on something. And so it provides a lot of information about what are the interaction modes that someone can use to try to get their information recommended or promoted by the system. - Okay, so Rob, this is where your expertise really comes in for me. When I sat down with Jeremy and Mayeron and Stanford last year, they noted that some of the biggest takeaways from system error, right? Are that Silicon Valley has asked coders to take on roles that they're not really trained to do around decisions like content moderation? We're now talking about content that users see that is determined by an algorithm, which learns by a set of rules that are written by a coder essentially. And that's been a big focus obviously on the way Elon Musk is talking about Twitter, but I want you to sort of color in the gaps here for us in terms of the ethical and moral and philosophical decisions that are being made sort of de facto by the people who are writing the rules of these algorithms. - Ron, let's begin with the idea that's now familiar, but it's an important starting point, which is that no technology is value neutral. Technologists are necessarily and inevitably putting into whatever code they're writing, a machine learning system, they're building, algorithmic curation system that exists a set of values. Sometimes these values are quite explicit. We're trying to uprank content that is more likely to be engaging to the users. Sometimes the values are less explicit and they involve things about wanting to downrank content that the company believes is either not socially valuable or politically controversial perhaps. There's a whole variety of different things that might be going on. All right, so first thing that this suggests to me is that when companies achieve tech companies achieve sufficient scale so that they have millions and possibly even billions of users, the values of the technologists who created the product become the values of the entire user base without any transparency into exactly how those values were chosen or whether or not the company sometimes even acknowledges that they're there. That's part of the controversy that Musk is addressing. Twitter's algorithmic ranking system has a set of values installed and they're not especially transparent. If you want to learn how the curation algorithm works, why does I receive my Twitter feed in the particular way it's shown to me? It's very hard to discover why that is. So I wanna add one other really important point here as well which is that the algorithm is the central feature of Twitter. While it's true, you can set Twitter in such a way as to give you a chronological feed. In other words, the tweets that show up are just in reverse order of when they've appeared on the platform. Almost no one uses Twitter in that way and uses the algorithmic curation or the algorithmic ranking content moderation to see things. And for anyone who's used any social media platforms, not just Twitter, but I'm thinking of YouTube, I'm thinking of TikTok, I'm thinking of, of course, Facebook. The algorithmic curation is the platform in many respects by which I mean the following. When Twitter says or Facebook says or YouTube says, we are just connecting users to each other. We are a system that is a neutral set of pipes. We make no judgments. So Ron, if I were to call you on the telephone and you and I talk on the telephone, we're using the telephone infrastructure and let's say it's AT&T or Bell or whatever. And you and I plot to commit a crime together. We've used the infrastructure, but no one believes that AT&T should be legally responsible for the content that traverses over the pipes. And that's a defensible point of view. Even if the content is as it were illegal or offensive or objectionable. And then contrast that with say a newspaper or a magazine or a television show where the employees of the company produce the actual content. So there's expectations about you don't defame someone or you are trying to have an orientation towards fact and you have a fact checking kind of part of your company. There's an editorial system about what goes at the top of the show, what goes at the bottom of the show, what doesn't get put on the show at all. And all of those expectations about content, provision are what we expect of those different outlets. Now, Twitter will sometimes say, we are like the telephone company. We're just connecting people. Don't hold us responsible for hate speech, misinformation, disinformation, or whatever else is on the platform. And that's false because of the algorithmic curation. They are putting a thumb on the scale of what we see and also what we don't see and it's personalized. That's the important thing to recognize about any social network, YouTube, Facebook, TikTok, Twitter. The platform is defined by the algorithmic curation. That's its central function. Give us the experience that we get on the platform. And so inspecting that and scrutinizing that and understanding it is indeed essential to understanding the effect that Twitter has in the world or any other social network. And then getting more information about how they company makes decisions, how the algorithmic curation is programmed or coded to make decisions about upranking and downranking is essential to understanding the experience that anyone has on the platform. And that's where Elon Musk is right. Transparency in order to understand the value choices that are being made is indeed a reasonable expectation. But as I'm sure we're going to talk about 100% transparency would also be a mistake. There's a certain naive "tay" to saying you should open source the entire algorithmic curation system. - Can you say a little bit more about that? What happens if it is completely open sourced and everything is made available, even if that's possible? How does that pendulum swing all the way in the other direction end up creating more problems? Complete transparency is an invitation to adversaries to understand how to gain the algorithm for their own advantages. So if you're engaged in deliberate misinformation or disinformation, and now you have a completely open sourced algorithm to understand what gets upranked and downranked, the adversary can optimize their own content on the platform for maximum engagement, which is to say to amplify misinformation, disinformation, and hate speech even more widely on steroids than is currently the case. Now, Marron can explain some of the details about this because this is not just a social network phenomenon. Open source code of any kind leaves the company vulnerable to adversarial uses because the transparency of the code is what allows someone to gain the system. (soft music) Support for political ecology comes from Quince, which has quickly become my go to place for the fashion staples I always find myself reaching for first. Quince focuses on high quality wardrobe staples made with premium materials, like 100% Mongolian cashmere, organic cotton, and marina wool, all in timeless versatile styles designed to last. Their Mongolian cashmere sweaters start at just $60 and they are all the softness and quality you'd expect from a luxury brand but without the luxury price tag. My current favorite is the Mongolian cashmere Johnny Collar polo sweater in Navy, which has been perfect for this in-between weather and everything at Quince is priced 50 to 80% less than similar brands. So you're paying for high quality, not the brand markup. Even my highly choosy husband has become a fan because the quality of their products is more like what he's used to finding in European boutiques. I recently ordered him one of their excursion Italian leather sling bags as a gift. Only to find out he had actually already ordered one. So, well, now we match. Download the Quince app for app exclusive offers or go to Quince.com. Now available in Canada and the UK too. And when Quince asks where you heard about him, let them know it was politicology. You're now up before the sun with a full day ahead. And that's when AG1 earns its place. The last thing it needs is a nutrient gap making it harder. But AG1 next gen has 75 plus ingredients, vitamins, minerals, free and probiotics, superfoods, backed by four separate gold standard clinical trials. Go to drinkag1.com/ron. And for a limited time, get a free immune support bundle, including a bottle of omega-3 fish oil and a bottle of vitamin D3K2 with your first subscription order of AG1 next gen or AG1 pro. This podcast is supported by Anthropic, the Public Benefit Corporation behind Claude. Everyone's got a hard question about AI, andthropic was built to surface those questions and share what it finds along the way because there's hope in hard questions. Ask yours at clawd.ai/spotify and keep thinking. That's clawd.ai/spotify. Ron, I want to go back to the algorithms themselves. I think you've mentioned this. I think many people are beginning to understand at scale now that the goal of technology is to keep you engaged with whatever you're using, whether it's the device or the platform itself that engagement is highly prioritized in the way algorithms make decisions. So if the goal is to keep people on Twitter, what kinds of things would the algorithm be trained to look for? And maybe you can talk about this perception, you know, that the Twitter says, if the user is conservative, then shadow band the tweet, how accurate is that idea? Because I think that's what Elon Musk is tugging at. Well, I think there's a difference between just thinking about engagement and thinking about content moderation policies around things that would lead to shadow bands. If we do think about engagement, you know, from that standpoint, it's, what are the algorithms optimized to do? And so if an algorithm is basically trained on data, and it's trying to predict what are the things that are likely to engage you, what it has available to it is data on things that have engaged you before. So it's more likely to show you things that are similar to what caused you to engage with the system in the past. That has the potential to do things like create filter bubbles that you get to see a lot more information that's similar to the information you saw before, or that the information you see comes from a smaller set of users from the people you're following, because they're the people you're more likely to engage with, because that's what the algorithm is learned based on your past behavior. So reinforces that behavior, because it's already gotten a strong signal that that's what you engage with. Now, at the same time, the folks at Twitter and other platforms want to try to keep the platform more or less if they can healthy. We can talk about whether or not they how well they actually do that relative to engagement, because if all you had, for example, on a platform with hate speech, chances are people would eventually get upset and lead the platform. So they do need to balance these different factors between how do we try to moderate content and how do we keep you engaged. There is more of an editorial decision process in that content moderation piece. I think that's part of what musk is trying to get at is how do we think about that content moderation. So when he talks about having transparency and the algorithm, the algorithm is only part of the story. Part of the story is also what are the content moderation guidelines that the platform follows. And that can open itself up to all kinds of issues around interpretation and where the platform wants to be. And you have to understand that also has second order effects around who actually stays on the platform. Let's then dig into content moderation, because I think Rob painted a really nice picture of how social media companies are neither dumb pipes, right, like the like the bell companies that we broke up a long time ago and made into public utilities. And they are not publishers either because they're not creating the content. They're somewhere in between. And I think this somewhere in between is is one of the biggest question marks not just from a technological standpoint, right. But also from a legal standpoint, because they want to have it both ways they want to be in between because they don't want to be subject to the rules and regulations that might govern either one of those to distribution models, right. So how do you think? How do you both think and maybe Ron, maybe you can lead off and Rob, I'd love to hear your thoughts. But what would be a responsible approach to content moderation? What do you think are some of the things and we can just talk about Twitter specifically, but feel free to open this up to Facebook as well or other platforms. But what do you think the role of these companies ought to be with regard to content moderation? How do you think they're failing and how do they need to fix their practices? That's a great question, Ron. I think it's a question our Congress is trying to grapple with currently. I mean, if you look at the historical foundations for what we have, we come back to the communications decency act section 230, which has been getting a lot of press recently, which is basically what grants platforms immunity from liability for the content that's actually published on their platforms. And back when the communications decency act was passed in the mid 90s, this was a reasonable thing in terms of trying to promote the internet. Now there is all kinds of grappling with how do we try to modify or potentially repeal CD8 230 to make these platforms more liable because there is this feeling that they are doing some kind of curation through their algorithmic amplification. So if you look at some of the bills that are in front of Congress right now, that's what the focus seems to be is the notion of algorithmic amplification and CD8 230 right now is written in a way that says if the platforms in good faith try to moderate content, that's what guarantees them immunity. So built into that already is a notion of platforms trying to do quote unquote the best they can and people debate what does that mean. Now I think part of the transparency, I think it's a little bit misguided to say we should just make the algorithms open sources a way to achieve transparency. There's other ways to achieve transparency like to say you have regular audits of algorithms for the kind of information that's being promoted by the algorithms, how it's being promoted, what it's political leaning is, there are some academic studies that have actually looked at this, but if there were opportunities to be able to do more in depth audits of the algorithm to actually understand what's being promoted. That gives us much greater clarity as to what users are actually seeing rather than people trying to dig into the code of an algorithm to understand what's going on that they won't be able to anyway because they don't have access to the data that that algorithms actually use to learn the patterns that it's making predictions based on. So I think we need to think about richer and more nuanced ways of doing that auditing of algorithms and that transparency algorithms rather than just saying the most transparent thing is making the code available because that's not actually transparent as people think. Yeah, and it's also not solving any problems either because the general audience is not going to be able to access that information if it's in code. So Rob, how do you think about this, the content moderation piece? I want to add to what Meron said because so I think and think of three specific kinds of approaches here that build upon the idea of an audit. So at the moment, any large social network has what it often calls a community standards policy or basically the principles by which it tries to moderate content on the platform. It will or will not allow pornography. Here's our policy about hate speech. Here's our policy about intellectual property violations or something like that. These are all typically quite transparent and you get a sense then about the basic framework for content moderation. Although there's no external opportunity to verify whether or not that actually given policy is how the content is moderated on the platform. So you get a clear statement of the principles of the framework, but you have to take the company at its word that it's actually applying that framework. So step one would be we insist upon an external audit of the actual moderation practices to see whether they map on to the clearly stated framework or principles that any company has. No, no public imposition of a new framework. Just take the public statement of the company's own content moderation and then insist upon an audit to see if it lives up to its own stated framework. So step number one, step number two would be to take what what is a related idea, championed as it were as it happens by a colleague of ours at Stanford, a law professor named Nate personally to try to insist upon external access to the data held by a company about its content moderation practices and the effects those practices have. So that independent researchers can get a view about the effects of the platform. So, you know, famously, Francis Hagen, for instance, leaked a whole bunch of internal documents about Facebook and Instagram claiming from the inside of the company that Instagram was toxic in certain respects for teenage girls in particular. Now, the only way those studies can currently get done is by being hired by Facebook to work inside the company and one of the upsides of Nate personally's proposal of this a law that would require any large company to give independent access to researchers, not to the world, but to independent researchers to study whether or not there is a filter bubble effect, a polarizing effect, an extremism effect. What's the proportion of content that counts as disinformation or misinformation, et cetera, et cetera? We want to have as a public some general answer to these questions that don't depend only upon the people inside the company giving us those answers. Okay, so that's step number two. Step number three would be what I would call a kind of an audit that has the spirit of like an environmental impact assessment. So right now, of course, in many different places, if there's a company that wants to build a big new building on a plot of land, especially if it might be large enough to have some impact on the environment. Maybe there's an endangered species or maybe there's an impact on waterways or something before a permit will be given to build the building to a private company. an environmental impact assessment has to take place to get a sense of what would be the effect on the built on the landscape on the natural environment if this building were built. A third and more demanding step for social networks for Twitter would be let's take your proposed algorithmic curation across a whole variety of things of social concern political bias misinformation hate speech etc etc. And let's test out your proposed algorithmic curation system against all of these metrics and we get a kind of social impact assessment of your proposed your company's proposed algorithm curation. And that can then help guide some of the decisions on a public basis about what the likely impact was going to be that would be a more demanding step because it would require the company to submit to a public evaluation. The goodness or badness or the differential effects of its proposed content moderation but given the civic significance I mean so Elon Musk says the reason why I want to buy Twitter is because it's the digital public square and I have to rescue it by taking it private in order to make the public square work in the way it should. If in fact social networks are the public square well the public has an interest in knowing what's happening in the public square and having a kind of public impact assessment for the algorithm itself would be valuable and keep in mind all three of the proposals I just gave fall far short of complete openness and transparency. It breaks access to the algorithm out of the company itself and gives others some input and visibility into it but not complete open source transparency that anyone anywhere anytime can get a view about exactly how content moderation is working. I really like the researcher access proposal in particular and many people some people will be familiar with the fact that Facebook used to offer this kind of access to researchers once upon a time until it was until it ultimately led to the theft of the the entire graph that was then given to Cambridge Analytica and everyone knows what happened after that but but that seems really prudent and doable. The third proposal the audit to me feels the impact assessment essentially it feels like it would just lead to extraordinary controversy over the who does it and the methodology that is used and I can't imagine any sort of consensus coming out of. Of something like that that would have enough credibility to be then sort of adopted or deployed how do you envision that working given the increasing polarization. Everything that you just said Ron was true about environmental impact assessments when they were just introduced so you get commercially driven companies that want to build you know buildings plots of land they wanted they wanted to their developers. And the developer will say well this is just the private marketplace at work there's a contract that you know it gave us rightful title to the land we were proposing to build something and we have to get permits that's all fine and the introduction of the environmental impact assessment was to give something that was in the public interest. A countervailing force to the commercial interest at work in the developers come you know sort of community the same is true now with the social networks when you operate at scale affecting the information environments of millions of people it's not just the commercial interest of making money that should matter there are public interest or civic interest at stake. So that's why I have in mind here these these are inevitably about value tradeoffs how is it that we can give some standing to the private marketplace to make decisions on its own merits about how to make money and provide a service that people use. While also giving some place to public or civic values that sometimes undermine or come at the expense of pure commercial interest that's that's what's at stake with Twitter that's what's at stake with most of these tech companies it's it's it's an unique economist's language a really familiar situation there's a tension between rival values we have to do the best we can to balance them. And in the current environment there's just very little standing to the civic interest to the public interest and social networks so introducing an impact assessment would be not to govern the platform from the public side it's not public infrastructure but rather to say we should be considering some rival values in decision making about the public impact of these algorithms are these social networks. Yeah, I mean it makes a lot of sense and it and even at a very minimum of beginning a robust conversation in the public domain that not on that platform but about that platform with more understanding about what the implications would and could be and presumably you would have you know that process could be opened up to multiple parties or organizations or entities to participate in as opposed to a single regulatory body that's going to say here's what would happen for sure. Human rights organizations could weigh in you could get information from marginalized communities who are disproportionately affected in certain ways by the amount of hate speech on a particular platform. This is not just the government coming in and telling a company what to do. I'm really glad we drilled into that because I think that makes a lot of sense. Support for political she comes from quince which has quickly become my go to place for the fashion staples I always find myself reaching for first. Quince focuses on high quality wardrobe staples made with premium materials like 100% Mongolian cashmere organic cotton and marina wool all in timeless versatile styles designed to last their Mongolian cashmere sweaters start at just $60 and they are all the softness and quality to expect from a luxury brand but without the luxury price tag. My current favorite is the Mongolian cashmere Johnny color polo sweater in navy which has been perfect for this in between weather and everything at quince is price 50 to 80% less than similar brands. They work directly with ethical factories and cut out the middlemen so you're paying for high quality not the brand markup even my highly choosy husband has become a fan because the quality of the products is more like what he's used to finding in European boutiques. I recently ordered him one of their excursion Italian leather sling bags as a gift only to find out he had actually already ordered one so well now we match. Find your next fall favorites at quince download the quince app for app exclusive offers or go to quince.com get free shipping on your order and 365 day returns now available in Canada and the UK too. And when quince asks where you heard about him let them know it was politicology. It's the best way to support the show for politicology comes from a G1 October has a way of sneaking up on us shorter days darker mornings and a routine that suddenly feels off. You're now up before the sun with a full day ahead and that's when a G1 or in this place your body's already working hard to adjust to the season. And the last thing it needs is a nutrient gap making it harder. I'm headed out on a long road trip this month which means my diet is going to be shall we say less than optimal. But a G1 next gen has 75 plus ingredients vitamins minerals pre and probiotics super foods backed by four separate gold standard clinical trials so wherever we wake up I know that with one scoop cold water every morning before anything else my bases are covered. Go to drink AG1 dot com slash Ron and for a limited time get a free immune support bundle including a bottle of omega three fish oil and a bottle of vitamin D3 K2 with your first subscription order of AG1 next gen or AG1 pro that's drink AG1 dot com slash Ron. Everyone's got a hard question about AI your job your kids where it's going and thropic was built to surface those questions and share what it finds along the way because there's hope in hard questions ask yours at clod dot AI slash Spotify and keep thinking that's clod dot AI slash Spotify. Back to the users being able to have some greater control over how they're served content on social media platforms may run what would it mean and what could it look like if users had more actual tools. And I mean this from you know if we're not going to give them the code to understand exactly you know what's happening if they are able to read it in the first place what kind of tools would be useful to users and and how could it transform their experience and then therefore the you know the platform itself to have more control over what they're served. Yeah it's a great question and a dovetails really nicely actually on the notion of thinking about auditing algorithms and getting an impact assessment because one of the things that's often been touted is. Is there different ways that we could give users choices around how they see content and giving access to the code is actually not a good way of do that if you were to do that if you're open source Twitter the vast majority of Twitter users would not do anything they wouldn't look at the code they couldn't understand the code they wouldn't. Make any kind of changes in terms of the data they were actually seeing so what you actually need is something that's meaningful to the user and one of the ideas that's been proposed is could you have what's referred to as middleware technically but basically different kinds of algorithms the people. choose from. Now the question is how do you choose among an algorithm, especially if you're not a technical person? What you need is some sort of audit or assessment of that algorithm that says, look, this algorithm actually is much tighter on content moderation because you don't want to see hate speech. You don't want to see that kind of inflammatory material. This algorithm over here is much more lenient. Basically, the only things that it filters out are things that are deemed illegal, right? Things that are just patently offensive. If you have some description as to what kind of information the algorithms give you, and it's audited in some sense, so you actually get a sense for what have people found to actually make its way through these algorithms to be promoted by these algorithms, then you have a more informed choice that doesn't require you to have to be a computer programmer to understand it. What it just requires is that you're willing to actually take some time to assess what your own value trade-offs are and where you come down on those notions of free speech and content moderation, which are really the values that are at stake when we think about this. That's what we want to put in the hands of the user, and we need to give the meaningful choices that they can understand. We can actually do that through a combination of technology and these kind of audits so people can make informed choices about what they pick. What you're talking about is actually algorithmic choice, more so than transparency, which sounds fantastic. I mean, are there any beta tools that is anybody demoing something like this that people could get their hands on and say, you know, experience the kind of choice that they might want Twitter to give them? Well, so there's been a report that's actually, you know, came out of Stanford, we'll have several Stanford co-authors on it, that looked at this notion of middleware as a way to do content moderation. Part of the problem that comes up though is how do you get platforms to agree to this, right? How do you get Facebook and Twitter's of the world to do that? Because it's in their vested interest to have control over their own algorithms. Now, there's a few ways you can do that. One is you can say, well, possibly the market might take care of it, right? If enough consumers demand this and say that they want to move to a different platform unless they have choice, then maybe the platforms will respond. That's nice, but I think it's unlikely to actually happen because given how much polarization and how much argument we've already seen around content promotion on these different platforms and seen basically no response from the platforms in terms of providing more choice, I think it's unlikely that a market solution is going to give us the full answer. So if we have some form of market breakdown, the traditional role that we'd actually have is for government. Some kind of regulation that says what we want to have is not regulation as to what you see, but regulation around interoperability with different kinds of algorithms that can be plugged in, the consumers can have choice from. This actually provides a more robust marketplace where different companies can now come along and say, here's our version of how we do content moderation and how we do promotion. Here's the assessment report that actually shows you how well we stick to our principles and now you get to choose. So if we had that kind of choice and so all we're really asking for regulation to do is to regulate the ability to have that choice not to make the choice for us. I mean, what you're describing to me sounds like the best case scenario. I mean, for me as a user, right, to be able to choose the algorithm that I want to experience the content on a platform with. But I just can't imagine a world where the law is going to allow you to usurp the intellectual property of the corporation in order to do that. Is it have any hope? So here's the key. You don't actually have to take away the intellectual property of the corporation, right? The corporation can still have its own ranking algorithms. As a matter of fact, what you're doing is you're pushing folks like Facebook and Twitter to give you a greater variety of algorithms to choose from if there is a competitive marketplace of algorithms. So what you're doing is in the same way that we, you know, have laws years ago that said, you know, what we don't want anyone to have a monopolist to control over a particular industry. All we're saying here is we don't want someone to have monopolist to control over the choice of how you see information. So what we want to provide is regulation that provides choice not that takes away intellectual property or forces a particular decision. So I think there's actually good precedent on that in the past. Part of the problem is the most, you know, regulation around monopoly has to do with consumer harms in terms of costs in the marketplace. Here we need to think about a different way. We need to think about information costs to the user. But if we can reframe that debate, I do think we have a rich foundation to actually push forward. Okay. Rob, here's, here's my question to you about this solution. Because my first reaction is, man, that sounds great. It sounds really, really good. And you're right, Mirana, I think there is precedent for that, especially since you're creating more competition, right? But if you think about this, Rob, in terms of what has been happening to the media landscape since the days of network only broadcast news, right? Where you have, and I think I'm just reprising something that everyone will be familiar with, but just to set the table, right? You have two or three or four, maybe single sources of information that the public has access to that is very thoroughly vetted for fact. And then there's a discussion, right, about what to do about the facts. We've now moved from that, from, from that landscape to one in which it, we have so in from, so much information that are disposal, it's impossible to even have a conversation about anything because there is no consensus about fact, right? What does it mean if we enter a world where, from here, where we have all of this information and no tools to filter it, to a world where everybody gets to choose their own bubble to live in, which is essentially what would happen if you have a marketplace of competition between algorithms that end users can choose? That's right. You know, would there be positive social consequences, civic consequences, if consumer choice about their moderation preferences were far more significant than it is today? I want to answer that question directly, but just first, kind of complicate the framing you gave to the question. Okay. You said, you know, back in the heyday era of three broadcast networks, you know, the evening news with the three trusted men in America giving us our, our fact-based information, there was a kind of, you know, highly controlled nature of information provision. And now in the world of, you know, digital information provision and social networks, we have content providers galore and we've broken out of this highly gated system of information provision. Now, that conventional story has a lot of truth to it. But since we've been talking this entire time about algorithmic moderation, notice that in the current social network environment, we can point to three or four dominant players that moderate the information we get. There's Facebook, there's Twitter, there's YouTube, there's TikTok. And in so far as the core feature of these platforms is the proprietary algorithm that is moderating the content, it's not that much different than that old broadcast TV world in which the three trusted men of America gave you the news. Now, it's the four trusted, well, not so trusted algorithms that provide our information environments. Yes, thousands and thousands of content providers, unlike the old world. But all of the content filtered through a very tiny number of algorithms. So that's the complication to the frame. Very fair complication. I follow. All right, so what would be different if, in fact, consumers, users had far more control? I think you're right to express a deep concern. Will individual users choose civically beneficial feeds, so to speak? If I'm a user, do I want to see things that are on the opposite side of the political spectrum to the things I believe? Or are I going to say no, show me the things that I are already familiar to me? Am I going to surround myself with comfort and familiarity? Or am I going to choose to prompt myself with contesting points of view? That is a legitimate concern. And we might get more polarization, more filter bubbles, more extremism than in the current environment. Total consumer choice as it were is not always a salutary thing here. And all I want to say here is that this is not a question of principle or philosophy. This is an empirical question. So let's run the experiment and see what happens. And the relevant baseline is not to a world in which there's like some utopian view about what the ideal content moderation provision is. But rather, the relevant comparison is to what we have today in terms of polarization, extremism, misinformation, disinformation. Can we improve upon that through consumer choice? That's the relevant question. And let's run the experiment. Let's see what happens. I think that's a really good answer. It makes a lot of sense. I mean, you're right. It is completely empirical. And I'm now fascinated with this question. And I can't wait. So is anyone running these tests? What is the sort of frontier of this potential solution? There's growing momentum to try this out. But let me just say from, you know, to give you know, maximal charity as it were to the social networks position here. The concerns that they'll express about middleware or the concerns that they'll express about various forms of algorithmic transparency or giving access to data to independent researchers are going to be, hey, the European Union taught us and now the United States is super concerned about privacy. And the idea of opening up the potentially personal aspects of users experience in the platform to independent researchers or to a middleware provider goes against all these concerns that we've been battered about privacy for so long. And we've now tried to put in place in the GDPR, for example, very strong privacy protections for data and opening it up to other companies like middleware providers or opening it up to users and saying I can bring my social network with me to a different platform, even though the people in my network didn't consent to my data being transported over to some other platform is a violation of the privacy concerns. So that's going to be at least one of the positions of the social networks in their current mode takes to express concern about any of the things we've just been talking about. And that's not totally unreasonable. There do have to be ways to think seriously about protecting the privacy of people's data and environments into which we ask others, middleware providers, complete portability of a social network over to some other provider, et cetera, et cetera. In order to see whether or not we're getting the balance right, but let me end this just by saying Ron, like here we are, yet again, in the situation of value trade-offs or value tensions. There are multiple things we care about, privacy, an informationally healthy environment where there's competing points of view, various concerns about extremism or hate speech while also promoting freedom of expression. You can't maximize all of these different values at once. We have to make decisions that involve trade-offs. How should we make them? This is what democracy is good for. This is why we put in place democratic institutions which allow us to debate in the open how it is to resolve trade-offs. And if we don't like the current decision, well, in two years we vote to people out of office and we try with a new model. That's why this now a civic concern is let's confront the trade-offs honestly and openly and say there's no optimal solution here. There's just a balance between rival values. - Totally agree. Predicated on the assumption that we have elections going forward into years. - Fair enough. - Fair enough. - That is a morbid tongue-in-teach joke. Mayeron, I do have a question about you on this privacy front, right? Is it possible to run this kind of experiment or to move in this direction while also ensuring the privacy of the users? For example, can the data be anonymized so that it can be worked with by researchers in this way? Or is it required to give up the personal data of individual platform users to third parties in order to run this kind of experiment? - Well, so there's different kinds of trade-offs involved here. One is to try to anonymize the data. To be honest, anonymization historically has not worked very well. If someone really wants to de-anonymize for portions of data, they've often been successful doing that. So the question is what kinds of safeguards do you put in? This is not just about social networks. We deal with this kind of information all the time. Our credit card transactions are healthier information, right? We have all kinds of places where if bad actors get involved, bad things could happen, but we think we put in place reasonable safeguards relative to the value that we get out of having, for example, a healthcare system. And so it's the same thing here with social networks. We can put in particular safeguards that make certain kinds of data available that create safeguards around the way and middleware provider could plug in to an existing data platform. So they may not be able to see individual data at a transparent way, but their algorithm would be able to have access to that if it's actually chosen to be part of the platform. The user would have to have consents to opt in to provide particular kinds of data to the platform. And you would have a liability regime that says if someone abuses the rights that they're granted under these data platforms, they have legal liability. So can we ensure privacy? No, we can never ensure privacy. But what we can do is we can put a series of safeguards around it to give consumers choice and the provide legal liability if someone tries to do something wrong with the data to get to a place that we're comfortable making that data available to be able to run this experiment. Got it. OK, I would like to bookmark a conversation potentially in the future for another trip to Stanford about privacy specifically and potentially privacy and cryptography. But Rob, before we go, I have one more question for you, which is this concept of the public square, right? There's been a lot of talk about how or whether the public square is now privately owned. And if Twitter is the private square, what does it mean that a single person wants to control it? If social media platforms are sort of collectively where public square debates are happening, then what does that mean for civil society? And I guess I'm wondering to what degree do you buy that argument? Do you buy into the, well, the public square is now virtual and it's largely privately held? And if you do subscribe to that, what does that mean for social discourse? And if you don't subscribe to it, why? Yeah, it's a fantastic question. And it seems to me it's the kind of question we all should be confronting because in an age of enormous wealth inequality, many of the most important providers of news and information are owned privately and sometimes are being taken private by people either commercially minded or philanthropically minded. So of course I have a mind here, Jeff Bezos, buying the Washington Post or Elon Musk's decision that in order to save the digital town square of Twitter, it has to be taken private out of the public marketplace of Wall Street trading. And there are lots of examples to point to here about philanthropists either as patrons or as donors or as commercially minded individuals taking private the providers of our public information. So I don't think there's a simple answer to this by which I mean it's not that the town square has to be publicly governed all the time. As many people including Elon Musk will point out, the New York Times is in fact a kind of family-owned newspaper for hundreds of years has not been, it's not PBS, it's not the BBC which gets public dollars and has some loose oversight by Congress. The same is true for lots of other kinds of things. There are Supreme Court decisions that have shown that private shopping malls, the traditional shopping mall that you go to, while you can't show up and protest inside the store. The hallways are functionally public squares and so there's been oversight about allowing freedom of expression even though it's privately owned land. I teach at Stanford University which is a private nonprofit and yet laws about freedom of expression apply to Stanford University called the Leonard Law in California so that whatever the First Amendment allows at Berkeley, a public university also applies to the private university that is Stanford. All of this is to say it's a complicated landscape. It's certainly possible that private ownership of various aspects of the public square is possible and delivers good results and it's also possible that it delivers bad results. And so a lot will depend upon not just what the law allows but what kinds of norms or kind of public spirit a private owner brings to whatever it is that is owned. The New York Times family, the Salisburgers, claim that they have been oriented toward the public interest defined in their own way rather than maximizing profit at all costs. That's part of the difference here is that if Elon Musk to go all the way back to where we started with Twitter, if Elon Musk wants Twitter to become the legitimate digital town square that it's meant to be in his view, he will have to demonstrate that he's oriented towards the public's interest rather than describing this as an opportunity for his own personal views now to be installed in the platform itself. He alone can rescue Twitter from its current problems. That seems to me like just an obvious self-refutation. It's like a reductio ad absurdum. In order to rescue the digital town square, I have to privatize it and I alone can save it is just a patently ridiculous thing to say. - Although in both of those instances with both the Salisburgs and Musk, we're still relying on their assurances. - Correct. - Right. That's right. (birds chirping) - Support for Politicology comes from Quince, which has quickly become my go-to place for the fashion staples I always find myself reaching for first. Quince focuses on high-quality wardrobe staples made with premium materials like 100% Mongolian cashmere, organic cotton and merino wool, all in timeless versatile styles designed to last. Their Mongolian cashmere sweaters start at just $60 and they are all the softness and quality you'd expect from a luxury brand, but without the luxury price tag. My current favorite is the Mongolian cashmere Johnny Collar Polo sweater in Navy, which has been perfect for them. in between whether and everything at quince is priced 50 to 80 percent less than similar brands. Only to find out he had actually already ordered one so well now we match. Get free shipping on your order and 365 day returns now available in Canada and the UK too. Before for politicology comes from AG1. Shorter days, darker mornings and a routine that suddenly feels off. You're now up before the sun with a full day ahead and that's when AG1 are in this place. Your body is already working hard to adjust to the season. I'm headed out on a long road trip this month which means my diet is going to be, shall we say, less than optimal. But AG1 next gen has 75 plus ingredients, vitamins, minerals, pre and probiotics, superfoods, backed by four separate gold standard clinical trials. So wherever we wake up, I know that with one scoop, cold water every morning, before anything else, my bases are covered. Go to drink AG1.com slash Ron. That's drinkag1.com slash Ron. Your job, your kids, where it's going. All right. I just have one lightning-ground question for you before I let you go, which is is democracy up to the task and the information age? Yeah. We are seeing the guardrails of democracy stretched possibly to their limits. And you know, the news of the primaries of the past couple weeks in which election deniers have been, you know, put on to the general ballot for governors, governorships as well as ordinary seats in Congress. Strike me as really, you know, huge warning signals for democracy. To put it differently, like, step number one of any democracy from any standpoint of democratic theory, democracy depends upon the idea that there are shared expectations about competition for public office and that if you lose, you recalibrate, you reorganize for the next election rather than try to change the rules of the election to make it seem more likely for you to win. So, democracy's most elemental feature is a peaceful transition of power in which simply asserting one's power to stay in power is the thing to avoid. And we are dangerously in the, you know, in the zone of the brute assertion of power to reorient elections so that minority interests can stay in positions of power. And I have to say that the next three or four years in American democracy are going to be the severest test we've seen yet. And there's good reason for hand-ringing and the work of journalists and of ordinary people to speak to each other across partisan lines is the most essential thing before us. In that spirit, Rob, I hope we can do more of this, maybe another conversation in person when I come to California, but it's been great to visit with you both and thank you for making the time today. Thanks, Frank. Thank you. Thank you to everyone at home and on the go for listening. Podcasts tend to grow based on word of mouth. So, if you want to help more people discover a political ecology, you can share this episode or one of your favorites with your friend group, your family or your colleagues. If you have questions about anything we've talked about, you can reach us as always at [email protected]. And even when we can't respond, we do read everything you send us. Whether it's an episode idea, a guest recommendation, or a simple note about how the show has impacted you. And we'd love to hear from you. I'm Ron Steslow. I'll see you in the next episode. Ryan Reynolds here from Mint Mobile. I don't know if you knew this, but anyone can get the same premium wireless for $15 a month plan that I've been enjoying. It's not just for celebrities, so do like I did and have one of your assistance assistants switch you to Mint Mobile today. I'm told it's super easy to do at MintMobile.com/switch. Intro rate first three months only, then full price plan options available, taxes and fees extra, default terms at MintMobile.com. We just moved into our new house and I didn't realize just how close we were to the neighbors. That's when my friend told me about Blinds.com. Customers who shop at Blinds.com experience excellent customer service from consultation to installation. Side effects include increased privacy and better sleep. Talk to your partner today and see if Blinds.com is right for you. Visit Blinds.com to order your free samples. When you're building a company, you learn that trust is what closes deals. You may have the best product, but buyers won't sign without proof that your product is secure, especially if it has AI. Vanta gets you compliant fast with Sock 2 ISO 2701 and more and keeps you secure with continuous monitoring. Hyperbound, a YC backed startup used Vanta to unlock the Fortune 100 in under a year. Snowfire got audit ready in a month and closed a multi-million dollar deal. Trust matters, get started at Vanta.com. When it comes to my policy, Farmers gives it to me straight, so there's no surprises. An insurance provider with a conscience. We're a choir, but yes, we have that too. Straight forward coverage you can literally understand. That's what you get when you have farmers, bump, bump, bump, bump, bump, bump, bump. Learn more at Farmers.com. Underwritten by Farmers Truck or Fire Insurance Exchanges or Affiliate, products not available in every state.