Signal’s President on AI, Advertising, and Running a Popular Messaging App — With Meredith Whittaker

0:00Signal President Meredith Whitaker comes on to talk about the popular messaging app and what's real or not in the world of AI. All that and more coming up right after this. Welcome to Big Technology Podcast, a show for cool-headed, nuanced conversation of the tech world and beyond. We're joined today by Meredith Whitaker. She's the president of the Signal App and the chief advisor to the AI Now Institute. Meredith, welcome to the show. So happy to be here. Thank you for having me. You bet. So we're going to cover a lot of ground today, but I thought it would be a missed opportunity to start without asking you briefly about Google.

0:35You spent 13 years there. What's your high-level view of the state of Google today? Look, I'm not at Google anymore. I hear the rumors. But history teaches us empires die, right? The winners don't stay winning always. And I think management at Google has always been a bit fuzzy. There's always been the cloud business has consistently struggled to catch up to Amazon and Microsoft. None of this is new news. I think for a very long time, Google was sort of figured as just the default leader in AI. A lot of the sort of early techniques around parallel processing were being kind of innovated in labs at Google.

1:22And, you know, what you have now is, you know, I don't even know if it's a struggle to innovate, right? Because I think we need to back up and look at, like, what happened with ChatGPT. Because ChatGPT itself is not an innovation, right? It's an advertisement that was a very, very expensive advertisement that was placed by Microsoft to, you know, advertise the capacities of, you know, generative AI and to advertise their Azure GPT APIs that they were sort of selling after, you know, effectively absorbing open AI as a kind of Microsoft subsidiary. But the technology or the frameworks on which ChatGPT are based are dated from 2017.

2:10So Microsoft puts up this ad. Everyone gets a little experience of communicating with something that seems strikingly like a sentient interlocutor. You kind of have a supercharged chatbot that everyone can experience and have a kind of story about. It's, you know, a bit like those kind of, you know, viral, like, upload your face and we'll tell you what kind of person you are, data collection schemes that we saw, you know, across Facebook in the 2010s. And then an entire narrative of kind of innovation and, you know, or a narrative of, like, scientific progress gets built around this sort of chat GPT moment, right?

2:50suddenly generative AI is the new kind of AI suddenly claims about sentience and about, you know, you know, the super intelligence and about, you know, AI being on the cusp of, you know, breaking into full consciousness and perhaps endangering human life. All of this almost like religious, religious rhetoric kind of builds up in response to chat GPT. So yeah, I mean, I'm not, I'm not a champion of Google, but I think we need to be very careful about, how are we defining innovation and how are we defining progress in AI? Because what I'm seeing is kind of a reflexive narrative building around what is a very impressive ad for a large generative language model, but not anything we should understand as constitutively innovative, right?

3:41Right. And so, okay, well, I definitely have some questions about that for you. We're going to get into it for sure. But just to focus a little bit more on this line, I mean, they are adding generative search to the top of search pages right now in a labs feature. But obviously, that's something that they're thinking about rolling out more broadly. And it's not that they didn't have this innovation, if you want to call it that, or they put this product in house, they did. It was I mean, we're both aware, right called Lambda, we just had the first product manager on Lambda and they decided not to ship it, which did lead to, you know, even if it's not a big technological breakthrough, that's going to kill us all.

4:20It led to, uh, you know, giving Google a chance to like, really, I'm sorry, giving Microsoft the chance to really outflank it, to cause a scramble. People called it a red alert within Google. I mean, that was the words they were using and they didn't release it from my understanding because of, of safety concerns. So do you I'm kind of curious what you think about it? Like, was that the right move? Where does it leave them if that's where they're going to be? Well, I mean, right as measured by what as measured by, you know, the demands of shareholder capitalism that require constant growth forever and ever and ever unstopping.

4:58As I said recently at a Washington Post event, you know, the definition of metastasis, like as measured by the requirement that, you know, the shareholder driven corporation sort of metastatically continue growing their market share, their user base, their revenue. Yeah, maybe that is a problem. But I don't know that we should, you know, we should be measuring the benefits of certain decisions based on that metric, right? So what I'm saying is, like, I think we need to reevaluate how we're measuring progress, how we're measuring innovation, and recognize that it's a problem when a massive surveillance corporation like Google overrides an entire, you know, you know, ethics team, you know, in order to push to release something because they feel that there is a competitor who is releasing something dangerous ahead of them.

5:50And, you know, thus, there is no value in safety anymore, because the real metric, the real objective function is revenue and growth and shareholder returns. But from the framework that you laid out, right? I mean, so describing growth as metastasizing something, I mean, what should a company like Google do? Like, should they just kind of crawl up and fold their cards and become that declining empire? I mean, from your perspective, what is the right path for that type of company? I think, again, the issue is that the actual objective function, The actual goals that this company must prioritize above all others are revenue and growth.

6:32And so my critique here, if we're going to be real, is of this form of capitalism. Of the system. So then pragmatically, like if you were running Google, would you just stop out of the system? Well, I wouldn't be running Google because they don't put people like me in charge of running Google. Yeah, but no, we're going to go hypothetical here. Yeah, hypothetical. They give me Google. I think we, you know, like how do those resources need to be redistributed? What does sort of technology in the public interest actually look like? What forms of technology are constitutively not in the public interest?

7:00How do we look at something that is, you know, a bit closer to a, you know, democratic governance of the role of these resources in, you know, human life? You know, I think all of those questions are questions I would want to raise. And I think those are, you know, these questions obviously have to be answered beyond Google. Right. And so this is, you know, this is the role of popular movements like the Writers Guild of America and others who are, you know, I think doing the best job of regulating AI of anyone out there right now. this is the role of you know some forms of state intervention you know depending on on what those look like and i think this is the role of you know kind of you know theorists who are capable of reimagining what role if any computational technology has across different you know aspects of our lives our institutions our economy but tech i mean so this is a big topic of debate right now the role of tech growth now i i won't deny the fact that there's been some bad outputs from it but i mean there seems like there's also it seems like on the balance it's been good i mean curious what your perspective is and we're talking through i'm using a chrome browser to speak with you right now like um obviously google has put like so much information at people's fingertips um it seems like it empowers people i don't know maybe as much or more than it would take away from folks.

8:29But you have a different opinion. Well, I mean, I guess get down, like, what do you mean by tech growth? Do you mean the commercialization of network computation that happened in the mid to late 90s? No, no, I mean that the various products that these tech companies have built. Yeah, so the commercialization of network computation. Okay, sure. I mean, which is kind of the, like, you know, you took these sort of research networks, these, you know, military communications networks, you know, that sort of developed through the mid-century. And in the early 90s, the Clinton Gore administration sort of put all, you know, there was Bush involved in this in the sort of late 80s.

9:07But, you know, there was put all the kind of economic eggs in the basket of, you know, the Internet will be a balm to an ailing economy where that just see neoliberal policies, you know, slash manufacturing that has seen the Democratic Party sort of, you know, move away from seeing its base as, you know, labor and workers to seeing its base as, you know, high tech and, you know, workers more or less. This is a very, you know, kind of quick summary of a much more complex era, of course, and then really pushed for neoliberal policies and neoliberal policies. That's not like a fancy word. That just means like an ideology that sees the private sector as, you know, ideally responsible for everything.

9:47Right. So, you know, privatize, privatize, privatized, as few regulations, as few, you know, interventions from the state as possible is that, you know, ideological framework, you know, sees neoliberal policies kind of applied to this commercialization. And, you know, I would say Matthew Crane's work, you know, particularly the book, you know, privacy or profit over privacy is a really good, you know, historical analysis of this time, which sees, you know, this, this sort of, you know, discussion throughout the 90s, policymaking process where the technology and advertising industries are leaning very heavily on the Clinton administration to ensure that there are as few regulations as possible.

10:31So it's industry-led, it is open for innovation, and to ensure that there is an explicit endorsement of advertising as the business model of commercialization. And this is borne out, You see the framework that was sort of published, I think it's 1997, although it went through many drafts, you know, explicitly endorses advertising as the business model of, you know, the commercial Internet. And, of course, advertising requires surveillance. Right. The more you know about your market, the more you can sort of target them, the more you can sort of make good on the promises of market segmentation and targeting.

11:08So this is, you know, this is the framework that then births these products that, you know, seem really innovative and cool. Like, yes, they have reshaped our relationship to information. But, you know, this sort of surveillance advertising business model also gutted local media. It also gutted our news and information ecosystem. There is very few business models now for independent news and local news, which in itself has sort of redounded to, let's say, a decrease in the strength of democratic processes and institutions at the local level. In the U.S., we have a crisis of gerrymandering where it's very difficult to claim that sort of elections are— But was it surveillance or was it just aggregation of audience?

11:52What's the difference? I mean, surveillance is gathering all this data on top of people and aggregation of audiences. Look, advertisers want to buy big audiences. That's why they love TV. And you can buy big audiences on Facebook and Google as opposed to a local newspaper. So if you have if you're like an ad buyer, you can go to, let's say, you know, Google or Facebook and get hit all your objectives as opposed to having to go to a disjointed group of like 500 local newspapers to do the same thing. Well, I think those two went hand in hand. Right. This sort of, you know, because it didn't start as, you know, one platform to rule them all.

12:33Right. you know, you've had kind of, you know, search and other things, but, you know, this business model was developing throughout the nineties and it isn't just a mass audience, right? We don't all see the same ads. We are segmented into, you know, micro targets that are, you know, like, you know, I am, you know, a woman of a certain age, you know, based in the New York area. And, you know, I go to yoga frequently and that's why you saw this ad, right? So, you know, that, you know, there is an imperative for data collection, which, you know, again, data collection on audiences and market segmentation was not new to the internet, but the, you know, surveillance capabilities of network computation and the ability to sort of, you know, link databases that existed before this commercialization with, you know, the imperative for advertisement and, you know, the ability to, you know, quickly, you know, through cookies, which was another thing like you know collect a bunch of data on people yeah you know overlaid those two things many of the local newspapers used the same technology they just didn't have big enough audiences so i i'm i guess i'm losing your argument though what's the argument here specifically on your point about the local news ecosystem it's more complicated than simply saying that it was surveillance that put them out of business but i guess i was building on your argument that like wasn't it a net benefit right okay and i'm saying we would have to go into the weeds and actually analyze like what were the collateral consequences beneficial to whom and recognize like we are in a crisis in terms of our information ecosystem our media ecosystem our ability to sort of access a shared reality to access like credible empirical information about our world and that is in part because of the dynamics of this business model which is self reinforcing, right?

14:24Which does, you know, once you have a sort of winner in this business model, it is very difficult for another competitor to break in. And that, you know, that speaks to your point around, you know, what I would call like platform dominance, right? There's one Facebook instead of, you know, a million heterogeneous little local news, you know, news outlets, which themselves can take different positions, can report on different things, can sort of provide a much richer information ecology than, you know, one Facebook that decides a certain type, you know, pivot to video now don't pivot to video right um and makes that you know determination in ways that um are profoundly undemocratic i mean definitely a rough transition for a lot of new sources to this new world but it was also a product of like a lot of terrible business decisions as well from newspapers and magazines not accustomed to having to adapt and innovate and i think that There are some examples now.

15:15But this is why we have, you know, like, do we want our media to be driven by the same sort of, you know, business imperatives, right? If a very valuable source of news and investigative reporting is not able to turn a profit, does that make it less valuable? No, no, definitely not. But I don't think they should. This is the thing. I don't think they should have to go through the same processes as like a Google. like you can build a successful local news brand and you don't have to follow the same you know uh targeting criteria that google did you can build you can do it they just had these legacy legacy costs legacy structures and they couldn't they couldn't adapt fast enough to like figure out a way to do business on the internet and ultimately like they were in the internet so that was sort of what was happening and i mean i think i don't agree you know okay like i it's something around 70 % of, you know, ad buys from media go to Facebook or sorry, meta, right?

16:14Like that's not, you know, I don't think we can say like, there was a way to pivot, but the old guys with their gray beards and their cups of coffee and their teletype machines just didn't figure it out, right? I think there was a predatory business model that, you know, effectively took the rug out from under these actors, you know, you know, who were themselves sort of, you know, advertising funded, not sort of funded by, you know, as in the UK and many other places, kind of like, you know, state media funding arms, right? So there was a choice way back when that media and news should be advertising funded, which itself was a, you know, was a thing.

16:53But, you know, and then, you know, the continuation of that model into commercial network computation, the sort of internet business model, you know, displaced the, and we didn't replace that, we didn't fill that in, we didn't sort of, you know, effectively take measures to preserve our information ecosystem from that transformation. Yeah, I will agree with you that it's not in a good place right now. And our society is definitely lost because of the fact that we don't have as good local reporting as we did previously. It's just enabled much more corruption and allowed behavior to go unchecked.

17:33So you also said that you found the generative AI is not actually that useful. And I think maybe that's right in the consumer sense, like ChatGPT, you know, it had declining usage across the summer. People have talked about how this could actually be very helpful in the enterprise sense. For instance, if you're a lawyer, using a chatbot to query like 500 different documents to find relevant information for your case might be interesting. Or, you know, in various other enterprise circumstances, you have a bunch of PDFs you just got to be able to talk with them, you know, have the bot read the data and then be able to pull out insightful information, you know, could be helpful as well.

18:12So, I mean, talking about ChatGPT as an ad for Microsoft services, you know, potentially, but there also were some actual interesting and I would say innovative uses that you're seeing right now with the technology. What do you think about that? I mean, I didn't say useless, right? I said not that useful in most serious contexts or that's, you know, that's what I think. and I think it's what I'm not saying is that if it's a low stakes sort of lit review a scan of these docs could point you in the right direction it also might not it also might miss certain things because you're looking for certain terms but actually there's an entire field of the literature that uses different terms and actually if you want to research this and understand it you should do the reading you know not maybe trust a proxy um you know that is only as good as the data it's trained on and the data it's trained on is the internet plus whatever fine-tuning data you're using right so i don't you know i'm not saying it's useless i'm saying it is vastly overhyped and sort of the claims that are being made around it are you know i think leading to a sort of regulatory and you know, kind of a regulatory environment that is a bit disconnected from reality and to, you know, a kind of popular understanding of these technologies that are far over, you know, over credulous about the capabilities, right?

19:45You know, like, again, it's not, you know, any serious context where factuality matters is not somewhere where you can trust one of these systems. What's your perspective on open AI? I feel like we're going to find common ground here. I mean, I think open AI is like a really annoying name, because every time you talk about like, you know, like it's sort of, you know, but it's not open. Insofar as there is such a thing as, you know, quote, unquote, open AI, not the company, which I write about with some co authors elsewhere. And it's, you know, not a clear term. And I think, you know, we need to be careful about that assertion.

20:23But, you know, I think OpenAI is very good at marketing. I think, you know, they are now effectively a part of Microsoft, which is, you know, points directly to the fact that, you know, these bigger are better paradigm for, you know, quote unquote, AI is in fact, you know, kind of centering and privileging, you know, a handful of actors who are the ones who have the infrastructure, the data, the talent, the market reach to be able to actually, you know, create and deploy these systems at scale. So, you know, the fact that Anthropic is sort of, you know, tethered to Google and now Amazon, the fact that OpenAI, you know, couldn't stay, you know, independent, because it became clear they needed the, you know, compute resources of one of these giants.

21:13And it's a lot easier to get those when you sort of, you know, absorb yourself into one of those giants benefiting from their economies of scale than it is to sort of license those outside. It's very, very, very expensive. So I think, you know, open AI becoming effectively part of Microsoft is an object lesson in, you know, just how concentrated the, you know, just how few actors are actually able to create these systems and just how concentrated the power in the AI industry actually is. Where do you think this all goes? I mean, if you're so skeptical of this wave of generative AI and you think it's a big wave of hype, do you think it just fizzles in a similar way that Web3 did?

21:59Or do you think that it's actually going to lead somewhere? I think it's being hyped, right? So hyped does not mean it's useless. It does not mean it's... It seems much more useful to me than the Web3 stuff for that. Yeah. Well, I mean, look, Web3 was built on techniques for doing cryptographically, you know, assured append only logging, which is hyper useful. Right. Certificate transparency. Very, very useful. There are many approaches to using those techniques to do something that is useful. But Web3 was sort of built on, you know, a lot of hype about, you know, effectively unregistered securities, a mad rush to sort of fill in the bottom of an ever expanding until it collapsed Ponzi scheme.

22:44And, you know, the like kind of hype around like NFTs and other, you know, effectively like applications of those technologies that were not very useful to many people. Right. And you have, you know, like DAOs and smart contracts, et cetera. But those were not like, you know, again, it was a, you know, it was a hype. And what was the motivation for that hype? In my view, it was, you know, a lot of big players in the tech industry were heavily invested at the top of that Ponzi scheme and were hoping for people to sort of fill in the bottom until FTX collapsed. And then, you know, kind of the whole thing fell behind it.

23:22So, you know, like what again, what I'm doing there is distinguishing from like what the hype was predicated on, what the interests of the hype were, the narrative of the hype and actually the technological affordances that underlie the hype, which themselves are not useless. Right. But the claims made about them were deeply dishonest and misleading. All right. Meredith Whitaker is here with us. She's the president of the Signal app, chief advisor to the AI Now Institute. On the other side of this break, we're going to talk about the app itself, Signal. Back right after this. Did you know your credit card points and miles can lose value to inflation?

23:56Credit card companies often reduce the redemption value of your points and miles. Now, imagine a credit card with rewards that can grow in value. With the Gemini credit card, you can earn Bitcoin or one of over 50 other cryptos instantly with no annual fee. Every swipe at the store or gas pump earns you instant rewards deposited straight to your account. Plus, sign up now for a$200 Bitcoin bonus to kickstart your rewards. Visit Gemini.com slash card today. Check out the link in the description for more information on rates. Again, if you're looking to invest in Bitcoin but don't know where to start, the Gemini credit card makes it easy.

24:34The Gemini credit card is issued by WebBank. In order to qualify for the$200 crypto intro bonus, you must spend$3 ,000 in your first 90 days. Some exclusions apply to instant rewards in which rewards are deposited when the transaction posts. This content is not investment advice and trading crypto involves risk. The Gemini credit card cannot be used to make gambling related purchases. What the hell is going on right now? And why is it happening like this? At Wired, we're obsessed with getting to the bottom of those questions on a daily basis. And maybe you are too. I'm Katie Drummond, the Global Editorial Director of Wired.

25:11And I'm hosting our new podcast series, The Big Interview. Each week, I'll sit down with some of the most interesting, provocative, and influential people who are shaping our right now. Big Interview conversations are fun. I want a shark that... That eats the internet. That turns it all off. Unfiltered and unafraid. So in a lot of ways, I try to be an antidote to the unimaginable faucet of reactionary content that you see online, to the best of my ability. Every week, we're going to offer you the ultimate luxury of our times, meaning and context. True or false, you, Brian Johnson, the man sitting across from me, one day, at some point, as of yet undefined in the future, you will die.

25:55False. Tell me more. Listen to The Big Interview right now in the same place you find Wired's Uncanny Valley podcast. Subscribe or follow wherever you get your podcasts. And we're back here on Big Technology Podcast with Meredith Whitaker, the president of the Signal App. Let's talk about the Signal App. I mean, what is the state of the Signal App right now? How many users does it have? How do you feel about the way that it stacks up against the other messaging app? Is it in a state of positive momentum now? So what's going on with that app? It's certainly in a state of positive momentum. It is the world's most widely used, truly private messaging app.

26:33We have many, many millions of users. We don't give a round number, but you can cheat a bit if you want and look at the App Store and see that it's been downloaded over 100 million times just in, sorry, the Play Store. The App Store doesn't give those numbers. The other one, the Play Store, see that it's been downloaded over 100 million times. And I'm really happy about the state of the app, about the state of the organization, and about many of the things we have moving forward, including the launch of usernames, which we're aiming for early 2024 there. And anyone who is interested in more on that can become one of the signal code watchers who can see the development of usernames happening in real time in our repos.

27:27How big is the team building the app? We're a little under 50 people now. So bigger than we've ever been and also incredibly small. And is the app funded by donations or do people pay to use it? It's funded by donations. So we had a, you know, a large, a large trance of money from Brian Acton, which has helped give us the WhatsApp co-founder. Give us a runway as we build up a donor supported model. We're looking to as much as possible subsist on small donations because we think that is the safest way to generate revenue. we don't want, you know, a handful of large donors who can, you know, have every right to change their mind at some point about their patronage.

28:20We don't want to be reliant there. So we are, you know, you will see in the app, if you go to settings, there's a little donate button, we encourage you to click it. And you will see periodically, you know, kind of a small message appear that, you know, asks if you're able to donate a little bit to Signal and you get a badge, which is this little, you know, it's like a little icon that appears near your profile photo. It's very cute if you donate through the app. So yeah, we are, we're looking at the donation model, but you know, let's be real. It is very expensive to run technology like Signal.

28:54And I think that part of the equation is often missing because of course, like, you know, consumer technology is often or almost always experienced as free. Right. And, you know, that is because we are, surveilled and subject to advertisements and data breach and privacy violations and change of terms of service to sort of target us with AI, all of those social ills come along with that. And just because Signal has a different model of collecting as little as possible, being incredibly staunch about privacy, doesn't mean it's less expensive for us to develop and maintain the app. So we have a piece coming out, kind of talking a bit more about the cost of Signal.

29:38And one of the estimates we give is that by 2025, we anticipate that it'll cost around$50 million a year to develop and care for Signal. And that's lean compared to a Meta or a Google or other large consumer tech company. Yeah. And the reason why I ask about resources is because it does seem to me like it's going to require more resources to compete with these other apps because for a while messaging apps seemed fairly static right they were a place to message your friends but now they're starting to be a place where i'll use the word so much innovation is happening right like you see broadcast channels for instance is something that started to come to places like telegram and whatsapp where basically people can disseminate information to people who subscribe to them, which is interesting.

30:26And you're starting to see AI bots right now. Right now, Meta has 28 bots that it's testing in Messenger and WhatsApp. You have things like stories. I know signals put in like stickers and stuff like that. So, I mean, of course, like the privacy message is something that's going to appeal to some, but to actually compete shoulder to shoulder, I'd imagined that you have to put in some of these bells and whistles. So how do you think about that when it comes to prioritizing what to build and just like the pace of change that happens inside these apps? Yeah, well, I mean, I think that's a great question.