Tech Brew Ride Home - (BNS EP) Anyone In Control Of Big Tech? W/ David Auerbach

Episode Date: April 8, 2023

Check out David Auerbach's books: Meganets: How Digital Forces Beyond Our Control Commandeer Our Daily Lives and Inner Realities Bitwise: A Life in Code Learn more about your ad choices. Visit megaphone.fm/adchoices

Welcome everybody to the TechMeme Ride Home Experience. Today is Wednesday, April 5th. We are joined by David Auerbach. David Auerbach is an author. He's a technologist. He's worked at Microsoft back in the day on MSMess and Messenger. He's worked at Google, back on the web crawler. So we wanted to talk to David because personally I was in Greece in Athens recently.

And I was giving a talk for the economist. kind of about how do I really think about this current moment that we're in with regards to artificial intelligence and society and social media and all of these things. And I actually listened to an interview with a friend of mine, Daniel Muneham, with David. And a lot of the things that he was talking about related to his new book, Meganets, being really pertinent to our audience, to our interests, to what we've been talking about. And so we invited David on to come give us talk, or I guess to talk with us about some of these topics. And so with that, I guess, you know, welcome to show, David. Thanks for having you, Chris and Brian. Yes, welcome, welcome.

Chris, since I am not as familiar with David's work as you are because I didn't listen to that episode. You did not do your homework. I know. I'll see it off because I already know what I want to get into with him, but start it in the direction you want to go and I'll follow. Yeah, you know, I think the thing that's going to be most useful to understand, because I felt a lot of resonant when I was listening to David kind of like they'll tell the story and then also like thinking about my own journey, you know, with social media and technology. And so I kind of wanted to understand a little bit more just from the earlier days of your experience, one, working on MSN Messenger, which is one of the early chat applications, one of the first maybe examples of a social product or social application. And then going to work, of course, at Google in its hypergrowth era and period. And what that meant to you in terms of, like, one thing that I kind of want to understand, and I think this is true for both of us.

is like there wasn't necessarily a master plan to get to where we are now. And as a result, there was a lot of improv. And, you know, now looking back and you'd be like, oh, man, maybe we should have known differently. But I'm curious about, you know, your experience, especially like starting out with Microsoft. Yeah. You know, build a social product there. Oh, there was no master plan. Well, I mean, or at least any master plan that came up never came to fruition.

Microsoft had many master plans back then. And most of them, I worked on a product that was code named Hailstorm. Some people might remember it. Oh, Microsoft Hailstorm. Yeah, wasn't that really bad? It was never really anything. It never, it never, it was actually, I think the Microsoft developers conference, they announced it, but I don't think it ever came to fruition.

That's it. You got lucky, Chris, because that's a better answer than him saying, that was my baby. That's the dream I've been trying to return to for 20 years. I mean, you know, the idea, I mean, like with a lot of these things, the idea wasn't necessarily terrible, but I don't think Microsoft sort of was in the right mindset to build something like that they were still effectively client focused. You know, that was what that was what had happened in the 90s that they decided that they would bring the Internet to Windows instead of vice versa. And I was working on the instant messenger client, and that was one of the, um, That was one of their successes in the internet realm,

but a lot of it was that their user base came from Hotmail, which they'd purchased. And so it wasn't indicative of, you know, the longer-term strategy. So in effect, Microsoft was still operating as like a boxed software company at that point. There were companies that weren't, but Microsoft still had this idea of, that in some ways everything was going to be based around shipping CDs of this or that, even if the internet was an auxiliary. What about Bill Gates famous The Road Ahead, where he was like talking about the future of the internet?

I mean, was that just hyperbolead marketing speak? I mean, I think that the, let me see, when did that come out? When did the road ahead come out? Because there was, I feel like that was the book was published. No. Because at that time, I think the goal was still that Windows would be. be the heart of it. Yes, everything would be networked,

but it would still be Windows connecting you to everything. Ironically, I mean, you can see a little of that happening now with Android, with Android, and the fact that Android and iOS effectively are the bases that we're moving away from a content neutral browser and towards apps and walking down. Okay, so here, I looked up Hailstorm, and Halestorm was announced in 2001. Okay, 2001.

That sounds right. Yeah, that sounds right. Yeah, yeah, yeah. Yeah, it was right after September 11th. That's right. Yeah. Go ahead. Hold on.

And so what's interesting about Hailstorm was this is a dot-net framework, which is, again, sort of the Microsoft way of building applications. And it was described as a new breed of platform consisting of a set of XML-based web services and an underlying services architecture. So, you know, 2001 and how we used to talk about things. it goes on to talk about, oh, God, what is it? Okay, Halestorm employs the passport user authentication system.

So identity was even big back then. Oh, God, what was the old identity format? Oh, Samil, Samo. Anyways, to secure an individual's identity and information, both passport and the Halestorm services require affirmative consent and explicit opt-in by the user for the release of any personal information. And it talks about how this starts with a fundamental

assumption that the user owns and controls their personal information for only the user decide with whom they share their information and under what terms. Now, it's so mind-blowing to me to sort of go back and to think about some of these core concepts that were in flight back then. And to think about obviously where we've come and how Zuckerberg and Bill Gates are kind of buddies and there's a mentoring relationship and to see how these core ideas of platforms were established back then. I mean, that entire idea of identity was there. And it was key is that people are going to have these

correlating identities online and whoever's in charge of adjudicating them and handing them out stands to make a tremendous amount of money. Passport was one of those efforts. And Google was up front about trying to do it with with

us, you know, Google Plus. Identity service. And I write in my book Meganets, I talk about how that sort of online identity is happening. coalescing. You can see it in India right now with the government Adharr program, that, you know, as our, as the amount of information we have online multiplies, the wish to coalesce such things does grow. But is it going to be Microsoft or Google adjudicating it? Who is it going to be

adjudicating it? I don't know, but there is going to be that drive towards coalescing it. And that does pose certain risks, I think. So, yeah, when I, I mean, when I made the jump from Microsoft to Google. It was like stepping into another world because there was no, you know, nothing was being shipped in a box. This was a truly an internet service and was on a scale that Microsoft, yeah, that Microsoft hadn't come to terms with. I was like, oh, okay, these people really, you know, know, know how to run servers and build out server infrastructure in a far more advanced way, you know, well beyond anything. And Microsoft had to play catch up with a, at this point, they seemed to have done so. But it took them quite. quite a while, but making that jump, it was like, oh, okay, we're stepping from this sort of unidirectional world where you give software to people and what they do with it. Well, there's bug reports and all, but in effect, they're passive recipients. And that gradually broke down that Google, you know, at first it was pretty much web content publishers that had the most influence, their input is actually feeding into how the creation

of whatever it is that Google's doing. But it wasn't until you get to the social media age that you start getting that feedback loop to get tighter and tighter and tighter until, in effect, you get this, you get basically power is devolving onto the users and programmers and corporations and governments don't have the same level of control that they used to because this information is flying around and conditioning and tweaking the parameters of all these algorithms, not just AI. AI, I think is the most extreme version of this. But, you know, what I saw was the increasing influence that we all collectively had, well, none of us individually having a decisive say

over it. And that's in the book, that sort of system is what I call a megamet, which which is one in which, you know, there's a feedback loop between the users and the systems such that even the, such, such that there's no coordinated human organization that can keep up with what uncoordinated human activity is doing. And you don't need AI for it to get out of control in that way. I can exacerbate it because now you're putting it through, you know, a machine learning network that is very opaque in its, in its work. But if you have simple recommendation algorithms or, you know, much simpler mechanisms of any sort, you're still going to get to this point that you can't exert any fine-drain control over how the recommendation algorithms, feed algorithms, you name it, are being shaped. Let me, I want to come to the thesis of Meganets, but I can't resist pulling this anecdote in because we mentioned it, and it's my favorite anecdote from, oh, a book. that I wrote. But Bill Gates agreed to write the road ahead. It came out in November 1995 and the index of the hardcover edition had 68 references to the term information highway

and 46 references to the term internet and four references to the World Wide Web. About a year later in the paperback version, the internet suddenly took 169 references and the web suddenly had 59 mentioned. So from the paperback to the hardcover, he tried to act like he he had the vision all along, which is, you know, if we all had a good editor. Yeah, if we all had the chance, if we all had the chance to revise ourselves between the hardcover of the paperback, we look like geniuses. Okay. So the thesis of that's really funny. I was actually not aware of that. It is one of my favorite anecdotes. When I was, you know, when I was writing this book, I was very concerned with future-proofing. it future-proofing it as best as I could.

So we got into the debates of, oh, you need to talk about the metaverse. So every time I mention the metaverse, I put in a little sort of caveat of, or whatever it is that these things are being called in 10 years. It's funny about this. This is sort of relevant. Are you familiar with another book called Megynette? You know what? I'm not until after, not until it was way too late. I remember by Wilson Dessard. Yes, yes, exactly. I literally did not run across this book until like.

So what, like, in some ways, I don't mean to like tease this out like too, I don't know, to make it too big of a thing. But what I'm finding interesting is sort of like the return of some ideas and some concepts that have been around. the computing world for a while. So I was wrong, as you said, Brian, about when Bill Gates published the book. It was 95, not 99, 99 is when I graduated high school. But Megynette, the first version, well, not the first version, but the Dissert book was published in 97. And the subtitle of that book was the Global Communications Network that will collect everyone

on Earth. And it was about mobile phones and mobile phone penetration. So I find this to be very interesting, right? Because now, you know, 20-some-on years later, you're sort of coming back. and in some ways, revisiting and, like, building on the observation of what was happening then with just connecting people through cell phones. I should have called it giganets. Yeah, yeah.

Well, perhaps. Perhaps it'll be the sequel. Yeah, well, you know, Bill Gates can rewrite part of his book and the next edition. You know, when you have the paperback version, you're going to be the gigonet. In any case, I think what you're talking about, broadly speaking, with this concept of your Megonet, is the idea of the feedback loops getting faster and faster. And your transition from Microsoft from a model of boxed software where you were literally pressing CDs, shipping them out,

and every year you'd have a release cycle. This is how Adobe built its business. This is why Figma had the advantage. Adobe was still very much predicated on having releases, whereas when you go to Google, now you're in a world where the code is being updated hundreds of times a day, where release management is a whole new discipline in terms of, I think you can go beyond that, because the algorithm, they're auto-updated.

That's new code is being rolled out at discreet, but very quick intervals. But moreover, it's not, you know, all the weights of the algorithms are changing in response to using feedback. So you literally are, you know, it's heraclytis, you can't walk in the same river twice, you know. And that wasn't fundamentally wasn't true 25 years ago, but it's true now that, you know, there's no easy game. guarantee that you're going to get the same output twice in a row. And that I think people underestimate what a profound thing that is to say that, because what we think of, so much of what we think of as how we control and test algorithms, starts with the assumption of being able to push a reset button that we can start with known

identical conditions and test in sort of an isolated, like, white box environment. And you can't do that's that's becoming more and more impossible. And I, from my perspective, that that that is as big a part of how we've lost control as anything else. That there's no reset button that you're just dealing with, dealing with this ongoing, the ongoing evolving systems whose value wise, not in their algorithms per se, though they contribute to it, but in their ongoing state. you know if you were to shut down Facebook and restart it from stretch you've just thrown away most of

the value that Facebook has so and and I think you know the the rise of deep learning machine learning whatever is an analogy and an extension of that because here you have these systems that

the algorithms are brilliant but you have to train them what are you training them on you're training them on the massive amount of data that we're now producing that we didn't produce even 20 years ago. But one of my favorite statistics is that we now produce more data every day than was produced in the entire history of humanity before the year 2000.

So that ties exactly into the question. The thesis of the book being in broad terms that people think, oh, these companies have control over their algorithms, or even their, fundamental businesses, but even in reality, they don't. They're just kind of riding a dragon that maybe they don't have control over. So my fundamental question is, is that the nature of the business that they're in, or is that a business model decision?

You know what I mean? Like, is the scale so large that no one would ever reasonably be expected to have their hand on the tiller, or are they deciding for whatever reason, maybe monetarily, whatever, to not have more control? I mean, primarily the former. Now, it's not to say that you can't do things, I get into this, that you can do things to try to arrest the degree of the loss of control.

Like when Facebook banned all political advertising in the run-up to the 2020 election. Well, that's actually, you know, you aren't going to eliminate misinformation on a case-by-case basis, But did that sort of like broad, fairly content neutral, coarse-grained intervention, that actually can do something. So if you want to look at it from the perspective of,

okay, could Facebook mitigate sheer chaos? Yes, there are options. However, there is nonetheless a degree of fine-grained control that you're just never going to get, period, simply because the size, the size, speed, and feedback these systems is just too bad, too fast. Sorry. If you want to go in and, you know, literally scan every piece of content that's going through Facebook or Twitter and eliminate the bad stuff for whatever definition of the bad stuff you want, that's not possible. It's not as though,

you know, when people are saying like stop getting people to be so mean on the internet, no, that's not possible. Now, the companies haven't exactly been upfront in admitting to that, you know, because I think they actually would rather be thought of as as greedy than as out of control. But I mean, there's stuff they let through that is they don't make any appreciable money off of. And that's just vile. And, you know, why? That's a loss of control. That's not.

That's not great. So, you know, they could be more honest. And I mean, you could also say, okay, well, um, Why don't we just shut everything down? I suppose a company could do they have that degree of control? That's not going to happen either, I think. But with regard to coarse grain mechanisms, yeah, they have some degree of control. But the myth that, oh, if only they wanted to, things could be nice.

And the way we used to remember them, no, that's not, that's not happening. One of the things that I thought was pretty interesting was you wrote in 2015 when you were describing machine learning, which, you know, given the timing, I thought was pretty early. And you wrote this for Slate. You wrote, and this sort of was describing the shift from, you know, somewhat like declarative deterministic programming where you should build an app. You press some buttons, you know, like you press the copy button and it copies. And then you press paste and paste. So beautiful. Right? Yes. To a world, which is somewhat more probabilistic. So you wrote, the question then is, why would one want to generate opaque and unpredictable networks rather than writing strict, effective programs oneself? And the answer, as Pedro Domingos told you, was that complete control over the details of the algorithm doesn't scale. And that there are three related aspects to machine learning that mitigate this problem. One, it uses probabilities rather than the true false binary. Now, of course, it sort of goes back to your book bitwise.

Two, humans accept a loss of control and precision over the details of the algorithm. And three, the algorithm is refined and modified to a feedback process. So why I think this is so relevant to revisit seven years later is because we are now in the enthralling era of scaled, machine learning, and artificial intelligence where everyone slowly but surely is going to realize that they are part of this feedback mechanism, that as you say, like this process of moving to a world of sort of probabilistic software is more like living in weather patterns. And what do we have? We have weather people that attempt to predict the weather and they could be right 50% of the time and keep their jobs. In a similar way,

we have these large scale AI systems that are, we attempt to predict their outputs. We attempt to improve them through modeling predictions that get better over time. But we also have to improve their performance through reinforcement mechanisms, whether human or virtual. Right. And yeah, I mean, at some point you're going to need AIs to assess the performance of AIs. And well, you can see where you can see where the circular dependency originates that at some point, it's like you're measuring yardsticks with yardsticks and how are you going to find what the right yardstick is? And no, that's a real, it's a real issue. This is why I think all those debates about machine bias and things, and that's the point a bit because we're going to have trouble just determining

whether something is biased. You can say, how are you going to audit, even at a black box level, how do you audit the performance? There are definitely ways to do it in controlled scenarios, but if everything's changing so fast, it's going to be a real problem. I think, but isn't like, I mean, to your commentary about like never stepping into the same river twice, isn't the nature of reality, and this is going to get metaphysical very quickly, so, we're warning to the listeners, but if reality is constantly and always changing, then one of the challenges with durable memory systems is that they attempt to record a steady state of reality in a certain moment of time and persist it eternally. But everything is changing

constantly. And so you need to have a system that in and of itself has the ability to constantly be kind of updating its model, adjusting its model. Even the use of the word bias, an entire neural network is based on a set of biases in the same. Exactly. Or an interpretation of reality. Exactly. So I guess where this leads me to is the change that is about to become common

in society in the world is one in which we have this new type of relationship, this emergent relationship that is co-modification. between our sense of self and our sense of the machine that we're working with. And that from one moment to the next, you know, I may be using my chat, you know, software or something and then it may have updated. And we, you know, we see this with auto-updating software now. You know, I am one of those strange holdouts that still wants to update all my apps on my phone. But for most people, their apps and their app experience is very much like the web. You know, they go there one day and some, you know, the features change or the search index is

updated. Something is different. Things have moved. Instagram is constantly moving things around, taking things. you know, A, B, testing, whatever. So even the fact of software is becoming one that is a lot more fungible and a lot more changeable. And so that creates a little bit of incoherence. Well, you know, it's funny. You bring this up and that, yeah, this was true at Google back 20 years ago with search with ranking.

I didn't work on ranking. Why? Because it was subject subject to exactly these sorts of feedback probabilistic mechanisms that made it very difficult. I wanted to build something that went from zero to 100 percent accuracy. And the ranking people are miserable trying to get from 70 percent to 71 percent. And they're having trouble just figuring out whether something's an improvement or not. So, I mean, those issues were seen in miniature then. Google, I think, was very fortunate in that they hit a gold mine. They hit the gold mine where they had this mechanism, this context in which people would tell

them exactly what they were looking for so they could sell ads on. exactly what they were looking for. I don't think there's been anything like that since then. I don't think Facebook found it. I don't know no one else has found a gold mine. Google did on the internet. It just, they were just, they just, they found it at the right time and they grabbed it.

But even then, you know, what, how, how the relevancy of search results, I see, you know, it was about like 70%. But that's okay. You don't need search results to be 70%. So the issue, And they were using machine learning like mechanisms. They were using Bayesnets and things like that.

Those AI things were in place, just not at the level that you see them now. And they weren't being used on analog data as you see them now. But those things were in plain. You were seeing that loss of control. Also because the webmasters would then change around all of their stuff to try to gain the end. And Google would have to go back and change it again. So there was also the users in terms of the webmasters, shaping how the ranking algorithm worked. But it still worked very well for a while.

One thing that's important about this observation and this point, about the dynamic between search and then webmasters and what became the industry of search engine optimization, is the same thing if you imagine like, you know, the bizarre, you know, thousands of years ago where, you know, being closest to like the entryway is going to help you have a better business. So in a similar way, someone comes to the market, they're looking for grapes or wine perhaps. And, you know, if you're closer to the entryway, then, of course, you'll do more business. So what I'm teasing out, though, is that there is this dynamic in the feedback loops that we're starting with Google. Essentially, there was this back and forth that seemed to be happening where Google would do one thing. And there was a whole panda kind of search algorithm change that affected a bunch of people, et cetera.

Webmasters learned how to design their pages to do keyword stuffing and all these things. Remember, Chris, I'm an entrepreneur who has scars for these words. Yes, that's true. Yes. You were on the battlefield of that era. But my point is, though, is that this is kind of evidence of those early meganets like forming, where there was kind of a command and response or market opportunity, you know, where there were incentives. And then that's, I cover that. And I cover that in the book, exactly, because that was when we see it forming.

And that was when I observed what I thought was if there was a single assumption, shared false assumption, I think tech companies, programmers, product managers, all of us, we all underestimated the degree of influence that users collectively would have on the system. It just didn't, it wasn't a way we didn't think of it in those terms. Why would we? It didn't make sense historically. The idea of making something and the users being passed. as the recipients was still there, even though, you know, we were now surviving on, these industries were based on content provided by the users.

And so I think, and I don't think that people even began to have an inkling until the era of social media, because at that point, it really starts to get out of control and you start to see actual, you know, side effects. I mean, we had a, oh, it's so, SEOs can be so annoying or whatever. But I think there was always the sense of like, whatever, we'll just, you know, there was no doubt about who was going to win. You know, the SEOs are annoying, but we're in charge. Well, as it turns out, yeah, the more, you know, the more of a voice people have and the faster these systems are working, the more you're giving up control. And I think that that is the main fallacy that we all have because I can't think of anyone that I knew who was talking. about user influence, collective user influence in that way.

And as I said, I mean, I was wary of it. I didn't like just the unpredictability of it, but I didn't, but at that time, I didn't see it as going so far in this direction. It, it occurs to me that again, like this is, to put it in like historical context, we've been talking about putting software in a shrink wrap box. and selling it on a shelf or whatever. But if you think of the 20th century and you create a product like a car, you sell that car to the end user, to borrow that term, and that's kind of it with your relationship with them.

I mean, you get feedback in terms of this model year sold better than that model year or whatever. The fact that in a digital economy, there is no sell it and it's done. And we were talking about, like, iterating different versions and such like that. But also, like, that's part of the nature of selling a digital product is that you have all of this feedback, so that it's almost like, as the producer of the product, you're overwhelmed by what you're learning about how it's used as well? Yeah. I mean, you become followers as much as leaders. And I think that I think this is true across the board.

I think that, you know, if you look at, I think Elon Musk has maintained influence in part because he actually jumps on currents that are already in play. And that we see that, you know, names become figureheads. I mean, albeit, you know, very rich and powerful figureheads, but that, you know, someone can just become the voice and figurehead of what is actually, a more decentralized or at least leaderless movement because you've got systems that are grouping similar people together who share similar interests who can then exert themselves in a coordinated way that wasn't possible before. And this can happen organically without anyone like driving it because all you need to do is to say, okay, look, give people more of what they want and introduce them to other people who want the same things. and that by itself is enough to sort of start the ball rolling. You want another fun anecdote, which you may or may not be aware, but when I had Matt Cuts on the Internet History podcast,

he said that one of the ways that they got intelligence from the SEOs is they would go to the SEO conferences and be very friendly and take them out for drinks or whatever. And he was like, after a few drinks, they would just tell you. They would just tell you what they're doing. Matt's a great guy. didn't know him well, but he had a fairly thankless job and did it really, really well. I don't know where he is now, actually.

Yeah, yeah, yeah, yeah, at least three or four years ago, yeah. Yeah, sorry. I think that's right. Matt, Matt, yeah. The White House. Really? Yeah, he is, he was the current head of the USDS, and that was as of 2018. Yeah, he was digital service.

Oh, that's really interesting. I had no idea. Well, he was really the point person on that thing, and he did an amazing job. He was really the fucks of it. He handled a lot of people getting very upset with him on a regular basis. I mean, to your point about Elon Musk, it is sort of like the kind of thing where now, it's not just like, again, walking in and out of the river and at the different river, now actually finding kind of like the currents and writing those currents and in such a way where you're kind of like like a pulpolar identity or personality such that there are many people, many actors in the system that are all kind of convening on you, as you said like the figurehead, that is drawing the ire of many different groups, both you know, positive, negative and whatever. And that balance seems to be the thing that kind of keeps people, I don't know, in the public eye or something.

Or another example was, you know, the GameStop stonk guy. Yeah, exactly. Do you remember his name? He testified in front of his name. But it seemed very clear he wasn't some mastermind. He happened to basically be in this right place at the right time and was a visible person talking about GameStop. But he, in fact, was representing a disparate group of people. He wasn't some, you know, demagogue who was typing people up.

He was one more YouTube cell up. And where is he gone? I don't know. These people, they arise and they... I could be wrong, but I believe he's suing Reddit because they... Yes, they shut it down, right? Or something or... But anyway, that could be wrong, yeah.

Listen, I do a daily news show, so I can't remember what I talk about... Yeah, three hours ago. Can I jump in real quick and then, Chris, back to you? One of the other things that I think you get into, is this idea that, you know, oh, what Facebook and Google need to do is just take all the misinformation and the bad stuff out. And then that'll fix everything. And especially as AI is coming, I'm curious your thoughts on that, to me, sounds like absolutely the horses are out of the barn. should we be thinking about not finding ways for average users or most users to not never see anything false or bad or whatever,

but find a way to identify what they're seeing for what it is XYZ. I'm sorry, I'm rambling. So go ahead and let me know your thoughts. Yeah, I mean, my sense is that even identifying it is tricky enough. and that because unless there's something that you can get, you know, 95% agreement on, and there's not a lot of that these days, then you're always going to have, then there's, it's very difficult to locate, um, where that authority should lie.

If it's in private companies, I don't think private companies even want that. It's nothing but a headache for them. If it's the government, well, they have tons of restrictions on exactly how they can adjudicate this themselves. You know, the, idea, the problem is that the myth of, you know, the section 230 of the communication decency act is that, um, is that there aren't any of these networks that are truly unfiltered anymore. Everybody's exerting some control. And you saw the Supreme Court hearings earlier this

year where it seemed like the, the justices were genuinely torn because they recognized that, that these weren't sort of just neutral carriers in the way that, say, a phone network was. But at the same time, they did seem to recognize, wait, how on earth are you going to make Google responsible for everything that's on their network? What the heck? They literally can't do it. So they're confronted. I think you're going to be confronting a lot of these problems legally because there's going to be basically total pressure not to accrue any liability. to any company of a big size.

They want to devolve responsibility onto individual users. But unfortunately, since individual users don't control the distribution of what they say, it becomes a paradox. So that's why my suggestion

is to say, look, you know, you're not going, it's just going to be whack-a-mole no matter what. If you're going to, if you want to address it, you're going to have to look at just things like basically slowing down viral, spread of everything not of specific types of content because identifying specific types of content in ways that a you know a super majority of the population won't find objectionable or wrong-headed is is not

feasible um now there are exceptions if you you know if there's some huge earthquake that okay no one's going to just hopefully not too many people are going to dispute that so you know the rawest of urgent is you can still sort of prioritize in that way. So wait. But anything else. So what you're describing is people have pejoratively called shadow banning, but you're not saying shadow banning for specific things. You're almost saying turning the temperature down on the virality of everything.

Yeah. Well, and I would say don't shadow. Well, if you shadow ban everyone, then nobody's shadow ban. Right, right. So what I would say is effectively the louder something gets, make it quiet or introduce some negative feedback and break up homogeneous groupings. In other words, there's this assumption that people, that you should give people what they want.

So all these algorithms are designed to give people things that stoke engagements, things that get what they want. That in itself is a lot more problematic than people seem to think, because giving people what they want is not always the greatest idea. And it's just that was a way to sort of kick the can down the road. Well, we're not going to decide what you want. We're just going to give what you want. Unfortunately, I think we're seeing that there are some emergent properties when you do that that lead to nastiness.

Like, if you give people what they want and what they want is something really is stuff that a lot of other people think is bad or whatever, you've just exacerbated it. So I think this raises like a very interesting generational or like generational ex societal or cultural question, you know, which is to the degree to which these things are kind of, you know, bad for us, both individually and collectively. It's almost like an information nutrition problem. And, you know, we've had to learn and teach ourselves over time what is like sort of a balanced and healthy approach to consuming calories. and a distribution of calories and proteins and vitamins and all the rent in order to maintain something of a healthy existence. In a similar way, I think the question of digital health or, what was the word I used? Anyways, let me all in those lines. Hygiene. Thank you. I call it data hygiene. There is a question as to the degree to which we need to teach people about these systems,

about how they are affected by these systems and also how they affect the systems themselves. And yet, I guess, you know, and I'm curious if you touched on this in the book, like, what are interventions that can actually happen at scale where you change the fitness function of humanity to be handling or to be living within the meganet? That's the, that's the last chapter of the book, which is what you can do. Okay, great. That is because I think that these megonets are, you can't, you can't educate people enough because it's like saying, it's like educating people to, recycle and use less fuel. It's not and doesn't make enough of an impact. You need some

sort of it's going to be a both ends. I mean, because we now like look, we are, this is the 2023 with first year that people are able to use something like chat QPT in all aspects of their educational experience. However, education now is changing in terms of its function in society and culture because you can just ask chat chbtee to do 90% of the work or more for you. And as you know, and as I know, the cleverness of people is such that they will find ways to use and incorporate these tools in very subtle ways. I remember you told a story about MSN Messenger and AOLN's to Messenger and how you sort of had forced or adversarial interoperability. That is an example of the type of cleverness that humans will invent to get around what are the types of censorship or kind of prohibitions that may be invented.

I mean, prohibition itself was something that people found their way around. So what, like, in your view, how do we prepare people to like be able to take advantage of these tools will not also usurping, to use your word, their own sense of self-sovereign. Yeah, I mean, I think we're doomed on that one. But like, the thing is, no, no, no, we're looking. And thank you for coming out. We're gradually outsourcing more and more of our fall, whatever, to, you know, to. to computers and that's just the way that things are going you aren't going to you aren't going so the self has already I think changed in alterably I think that you know this

idea you know the existentialist idea of oh you're a blank slate you can reinvent yourself at any time well no you now got a huge digital paper trail a mile long everybody knows what you've been so in other words my own identity has already been getting locked down in ways that it never was locked down before. So I think we are looking at a fundamental change of identity. I think we're looking at an identity that's much more quantified. People are defining themselves much more with labels, labels upon labels upon labels. So, you know, and you can argue whether that's good or bad. I'm not a fan of it, but whatever, it's here. My concern is more in terms of, I think, preventing sort of those

viral, harmful viral explosions in which there is a sudden shock or destabilization of the system that causes massive damage, you know, akin to that of like the 2008 financial crisis and high frequency. As you say SBB. Yeah. Like, is there an equivalent in the information system such that an SBB-style information trauma would not have cascading effects? Right.

And I mean, you're seeing that happen with cryptocurrency because, you know, whatever we've cryptocurrency is nothing compared to what we're going to see because getting more and more integrated into the global financial system. And if you look at, you know, at FTX, you look at it's like, how did these people convince anyone that they were serious in retrospect? And part of that is because there's so much information that validating information, and making sober assessments of, oh, is this insane or not? Is this offshore crypto company sane or not? I suspect it was done because they had the right pedigree,

and that was basically the shorthand. Well, they had all the labels that would identify them as being credible or worthy of being believed or believable in the network archaology. Yeah. And as a result, they stayed, yeah. Whereas, you know, a few decades ago, you wouldn't have just needed that, you would have also had to come into a room and shooze with people. Right.

It wasn't necessarily, and I mean, I'm not saying that that was perfect at all. It led to nepotism of a different sort. But what we're seeing now is that you can actually do this remotely, and it opens the door to people's BS detectors not firing because everything... They have been trained on these types of bacteria. Totally. Right? Brian?

Yeah. 15 degrees to the right or left a little bit. We have thousands of listeners that work at big tech companies, tech platforms that I don't think any of them will be surprised at your assertion that no one's really in control, you know? I know. I know. I know. That's what I start off the book. I say, you know, there's this conversation I have where people is like, why don't you just fix that? And all the engineers I know are like, I guess maybe this is unfair, but my question was going to be a listener like that that believes either in their company or in their original, why did I get into tech? I want to make a dent in the universe. If you're working at these platforms, do you have any advice for those folks who want to make them better? I mean, yeah, I think that what you want to look at is what interventions actually have been effective. And I mean, to some extent, you know, even the people on the ground, I'd say this from experience. Often engineers are being told to do impossible things by executives. So they may also feel.

I don't have advice on how to change that, I'm afraid. But the advice I give at the end of the book or some suggestions is to look at these non-targeted mechanisms of making recommendation algorithms, you know, met with them and breaking up homogeneous collagulations of social groupings and content groupings. and in fact, not requiring a greater degree of data validation. There's so much data that goes through that often is not accurate. Facebook thought I was black for a number of years. They may still. I didn't tell them that. I don't know where they got that from.

If there was more of an emphasis on even validating, you know, of elective third party data that would do something for it. But you wouldn't even have to validate it. You could just poison it and say, okay, look, we're not, we're not going to take that. We're not going to go on the assumption of this data is correct or we're just going to intentionally corrupt it because we can't validate it anyway, so mix things up. I know TikTok was actually trying to show more heterogeneous content to people after a lot of like pro anorexia videos are getting shown and recommend.

That's the sort of thing that I think can work is that you're not looking at stamping that stuff out. You just don't want, you just want people to be taken out of a narrative bunker in which everything they see as the same set of shared assumptions. You're kind of talking about introducing some decay, perhaps, into the system, some way to get people out of their kind of information. cul-de-sac. I think that's I think that will help a lot yeah along with the sort of the negative feedback mechanisms to decrease viral spread and prevent anything from from from blowing up too big too quickly you know yeah I mean one of the things that experiment experimenting with those sorts of what I would suggest to people who are working at these companies I mean the thing is

these companies themselves are frequently huge entities. So the will to do so, it often has to come from even one company isn't enough. You need to get coordination between multiple companies to agree to do things. That could happen. So in effect, I guess even simpler than that, I would say start thinking about things from more systematic perspective. rather than looking at individual actors. Because the individual actors, you can't shape them that much. I remember someone suggesting many years ago,

oh, shaming people on Twitter with pop-ups to, like, prevent them from acting too negative. And I was like, yeah, sure, that'll work. They have that. It did show that there was some. It's one of those sort of like, I don't know, it's not attunement, but it's like, eventually you become sort of

in near to them and like if you want to call them to fuck with like you will and it's just like too bad. I get them sometimes and I'm like no actually that's what I really want to say you know. Yeah or you express yourself in ways that don't trigger it's like China's been trying to do this for a long time and with very success.

The thing is that it's like oh or I think I quote this in the book my favorite headline this was from 2016 it's time for the elites to rise up against the ignorant masses. The mere fact that you have to write an article and publish it with that title shows that it's not going to happen. One thing that I want to bring up as we as a rapper is, you know, you wrote about Wittgenstein. And I don't think we've ever brought up Wittgenstein on this podcast before.

And I will admit that I ever... That's where the yardstick analogy is from, actually. I actually already brought them up, yeah. I feel like over the next period of time, I'm going to need to actually do a little bit more digging, because a lot of his philosophy around language and words and the experience of words is so pertinent to this moment that we're in, right? This large language model kind of epoch that is upon us is exactly one in which,

as you were talking about labels, and labels being either self-asserted or applied to other people determines in some ways like, who you are to the system and what recommendations you receive. And so the degree to which there is not a label that is sufficient to describe you or your interests or perhaps the negative space, which are negative labels, which are things you haven't experienced yet, could be an opportunity to essentially create, you know, sort of a backlog of like, here are all the ideas that are still out there that you haven't yet yet, and, you know, right, like it's sort of like negative space recommendation system.

Well, I think we all go around, assume it, you know, with this idea that things are more ground that our language, linguistic expression, even our concepts are more grounded than they actually are. And if you look back at, if you look back at the history, it's concepts involved at a, at an incredible. And as we started, Bill Gates was like, oh, that's not the words I really mean. I mean me. Yeah, exactly. Yeah. Or I was walking by a house, an apartment complex yesterday.

And I was like, oh, it was weird. And it used to be a hospital, but it looked like a little like fortress with with circular, like a bunch of circular cylinders connected together. And I looked it up and it's like, oh, it was created because at the time, the thinking was that corners spread disease. That's why they had these round. No, this is really there isn't really. There's like, holy, holy. That's why they've got these round cylinders, cylindrical things. It was like, So there's so much that will get revised going forward, and we're expecting machines to grasp and track that evolution. It's going to be inexact. And yeah, you know, that's Vittgenstein's contribution is to say that language is that is the yardstick that you're measuring with another yardstick, that you don't have something to appeal to and say, okay, did I say that right or not? And in effect, we're all doing a bit of legislation and saying, okay, this is how I think language should be used every time we speak.

And now we're doing that, we're telling that to chat GPT. Right. And by speaking, chat GPT is legislating to us as well. And that's the new. I think the thing, you know, just again, to close this out, what are the things that was really, that really stood out to me and why I think the concept of Megynet's, appealed to me was, of course, with my relationship to the hashtag, it has created kind of flocking behavior within people where previously, like, there weren't, you know, words that existed that captured a certain, you know, idea or experience. And that upon doing, upon sort of like

the invention in the flow of language, you know, whether it's like hashtag Me Too or hashtag Black Lives Matter, these were words that didn't have purchased previously. And then through the back mechanisms and the speed of those feedback mechanisms, now you can have broad participation by a set of people that previously perhaps were not franchised. And so... Oh, yeah. Yeah. Yeah.

They act as attractors. Exactly. And I mean, I think... I mean, so this is my own philosophical bias, which is that, you know, I don't think language works the same online as it does in person. That I think there's something fundamentally different about talking to people in that sort of disconnected way, as opposed to speaking to them in an embodied way. an embodied world because basically you're taking away that much more grounding of okay like oh

i'm pointing to this and i'm seeing the same language and i think the way that gets compensated for is that you fold yourself into a micro community that speaks the same language and these terms you can identify it's not just the terms with which you identify yourself it's also a whole vocabulary you know if you're talking about wokeness that's very different than if you're talking about intersectionality the mere use of those terms is enough to be as inclusive or exclusive. Well, it's also interesting because everyone's like, well, you would never say that to me if I was in front of you, but you'll comment and say terrible things to me on my YouTube post or whatever. But what you're saying is like the actual functionality of the language is different. Yes.

Yeah. I think so. Yeah. I think that if you look at it, like it can't make meaning in the way that it's, it's different. does in real life, or at least it does so. It's much more. Like with memeification and the shorthand that that is how, not to use memes, but like that's almost a grammar that is not used.

It's used. We quote from movie lines and stuff to each other in person at a bar watching the football game or whatever, but it's not used as much as it is online for that. Yeah. So a way to think about language as not operating system is like there's a bootstrapping method by which you kind of arrive at saying, what is the sort of assumptions or primitives or, um, yeah, like, I guess like language modules that we can take for granted. And memes for movies only work if you are in the same operating

system culture because you've seen the same movies, right? When they're, right, and they're from Indian culture, Japanese culture, they don't make any sense to me because I haven't seen the movies and I have no idea what they're talking about. So that vernacular is locked on me because I'm in the wrong operating system. Yeah, and now you're getting, and I think the thing is that, you know, there's no overriding mass culture to the extent that there was 40 years ago, even when I was growing up. No one watches all, even, even, I mean, I don't think you could even get something like the end of the Sopranos anymore where everybody was complaining about it. Like, you have it. It's still, like, it's still, like, there's still a concept.

I love succession. But it's a very, it's a particular demographic. Yeah, and I was like momentarily in like the left of us niche. and we were all sharing means. But also, there's another thing. Think about the time shifting of it, where everybody watched MASH at the same time. Everyone now knows the wire or breaking bad, not everyone,

but not everyone did in the same time scale as well. So there was a language for people that were on the wire or breaking bad from day one, that for years, people, that was a language that they didn't understand. Yeah. And now I don't, I think it's fracturing even more that, I mean, the wire, I think that's still going to have a much bigger, has a much bigger resonance than Succession will. And I love Succession, so I'm not going to complain about it. But, you know, but I, one of the nice things about sort of like dropping into different subcultures is, yeah, you'll stumble on stuff that seems to be that, that no one in. your, you know, peer group has ever heard of. I, someone brought up the Russian video game pathologic last night.

And I was like, holy crap, I happened to have played that, but I can't say that I've talked to anyone else who has. It was really, it was just like, what are the channels? And that's the thing is that there's, the sheer volume of it is such that it's hard to make something for a mass audience that, because there's that much less of a shared collective knowledge. Well, okay. So last thought, though, like when you say mass audience,

I mean, if there was, I don't know if it was 2 million or 70 million people that watched the last episode of MASH, but these days, it was like 80 million. Relatively speaking, 80 million, like, what is a mass audience? If you can get out there to 80 million today, if that's a relatively modest audience, it's still as big as the greatest mass audience of the previous epoch. So is it just a matter that we are living with just a different scale of access and communication?

I think so. I think it's a accounting weather. The bar of content creation has dropped more, too. I mean, soon enough, people will literally be, you know, I could, you know, I'll be able to make a new Humphrey Bogart movie in 20 years. Right. With, with, with whatever machine learners. 20 years. I was going to say, like, take 20 days. Yeah.

With a runway an hour. So, I mean, maybe sooner. But that's the thing is that the bar of content creation and content distribution. And obviously there are people who are trying to keep control of that, but it is slipping gradually. I think everyone will agree with that. And I think that it's accelerating. Since interest rates went up, there's less money being fed into VC-backed firm. And so there's that much less attempts to control sort of, you know, what's popular and what's big.

And that's so succession, you know, however much you like it, is that idea of a prestige show, I think, is somewhat dying out as well. Simply because, you know, I might be able to find something that I like more. And so not enough. There's never a critical mass for anything. And let's not even get started about there was a time when everyone saw the same movie that wasn't a superhero movie. Okay. We should wrap.

I'm going to do the plugs and then Chris will close us out. David Auerbach, the book that we've been talking about mostly is Meganets, how Digital Forces Beyond Our Control, Commodeer, Our Daily Lives, and Interested. realities he has a previous book called bitwise a life and code check those both out and i'm going to thank you personally and now chris is going to thank you yeah well again david thank you so much for coming on the show thank you that's great i really appreciate it um where where shall people go to find more about you either follow you on some social media platform or let's see uh well if you search on Meganets, you'll turn up.

Oh, this is the best thing. I search for mega. That's probably the best thing. You'll find, no, no. If you search for megan. I have a, I have a, I have a, you'll get our stack, you'll get our stack,

and then stack. My name. And I'm on Twitter as our Bach Keller, which is a GERTA reference, which no one ever gets, but it's a reference of those. so I should probably change that, but whatever.

And I'm in the other usual spots, too. But reading the book, hey, that's probably the best thing to do. And for the coders out there, yes, I made it bitwise so that people would know that I had some idea what I was talking about. Your own little self-label there. All right, well, David, thank you very much for coming on the show. Appreciate it. Thank you so much for having you.