The Initiative for Digital Public Infrastructure

@ UMass Amherst

Reimagining the Internet

Tim Hwang, Subprime Attention Crisis

December 02, 2020

We’re delighted to welcome Tim Hwang to the podcast, author of the recently published Subprime Attention Crisis: Advertising and the Time Bomb at the Heart of the Internet and the brains a great number of eclectic, eccentric tech-related ventures.

He talks with us about unchecked fraud in the programmatic advertising industry and who he’s successfully managed to infuriate with his new book.

Tim is currently a researcher at Georgetown's Center for Security and Emerging Technology, organizer of ROFLCon, founding partner at Robot, Robot, and Hwang, core to the launch of the Awesome Foundation, curator of The Trade Journal Cooperative, among many, many other things.

Transcript

Ethan Zuckerman:

Welcome, everybody, to another episode of Reimagining The Internet. I'm Ethan Zuckerman from the kitchen that I've been trapped in for the last eight months. I have the wonderful opportunity to hang out today with my friend, Tim Hwang, who is hiding behind a microphone in a closet somewhere on the Upper West Side of Manhattan. This is, as you might imagine, yet another pandemic interview, but it's one that I'm particularly thrilled by. I could fill a podcast simply by reading the projects that Tim has been involved with in the decade or so that I've known him. So instead I'm just going to hit on some of my favorites.

Tim founded a law firm called Robot, Robot, and Hwang, until he got fired by his partners because once they were automated to do the work, they didn't need him anymore. He's one of the cofounders of ROFLCon, rolling on the floor laughing con, which was a conference of internet celebrities, which was really fun until we discovered that Keyboard Cat has an agent, which was so dispiriting that it's just kind of hard to laugh about these things anymore. He was my editor at the remarkable California Journal of Images and Mark Zuckerberg. That's probably my favorite academic paper of the last few years that I've written with his guidance, but really he is a writer, a very deep thinker about the internet and internet culture, a researcher. And right now he is the author of a really remarkable and important new book called Subprime Attention Crisis. It's just out with Farrar, Straus and Giroux.

And Tim, we ask everybody the same question here, and it's a pretty simple but massive two-part question. What's wrong with the internet, or maybe in this case, internet advertising? And what should we do about it?

Tim Hwang:

Cool. Well, Ethan, thanks for having me on the show. So in typical fashion, I think I'm going to attack the question and then I'll go with the question.

Ethan Zuckerman:

Absolutely.

Tim Hwang:

I mean, I think the hard part about your question is, how do we characterize the internet? It's such a big unknowable thing in some ways and it's very difficult to characterize, and almost any way of characterizing it is inevitably going to leave out the experiences of people. And so I think with that proviso in mind though, what is wrong with the internet? And I think one thing that is wrong with the internet is that some of the biggest companies that have really shaped the modern experience of the web are built around a financial engine of online advertising, specifically a kind of advertising known as programmatic advertising. And this core financial engine has produced a lot of bad impacts on the design of the web, but I think also may actually lead to a great deal of instability as we think about how financially robust is this kind of structure that we've created over the last 20 years? And where is it going to go for the next 20 years?

Ethan Zuckerman:

Let's talk about what distinguishes programmatic advertising from other types of advertising. So when I type into a search engine, I want to get my roof fixed in Pittsfield, Massachusetts. That's a different sort of advertising than what you're talking about, right?

Tim Hwang:

Yes, that's correct. Well, I should say at least, when you type in a term in a search engine and an ad is delivered to you, that is programmatic advertising. Now, I think the main distinction that people should know about though is when someone uses the word advertising, they usually think about Mad Men, right? These guys sitting in smoke-filled offices saying offensive things to one another. And I do think that one of the things that's often missed is that the modern-day structure of advertising is very different from this. It's a lot more sort of Wall Street than it is kind of Madison Ave, in the sense that-

Ethan Zuckerman:

A numbers game rather than a creative game.

Tim Hwang:

Exactly. Right. It's extremely data-driven and it's largely facilitated through algorithms that compete with one another to deliver an ad to someone. And so when you go search for a term, say carpenter, on Google, essentially there's a split section, a second auction that takes place to deliver an ad to you at that point in time, and there is basically a huge infrastructure that allows for these auctions to happen in between the time at which you click on a link and the time in which the website loads. And these trades happen billions of times every single day, every time an ad is delivered online.

Ethan Zuckerman:

And so my local roofers are essentially participating at an auction and the prize is my attention.

Tim Hwang:

Yeah. The right to load something in front of your eyeballs, or at least attempt to.

Ethan Zuckerman:

Okay. And so, what's wrong with that? I wanted a roofer. Roofers are going to compete to pay for the right to put that ad in front of me. Shouldn't, under Western capitalism, doesn't that mean that the best roofer buys my attention and that everyone's happy.

Tim Hwang:

Sure. Right. And actually, the kind of allusion you're making there is a really good one, which is that there has been a lot of economic theory that has gone into designing these attention marketplaces. And a lot of it is, I think, very much tied to the sort of classic economic thinking which is, we create an extremely liquid market, people compete on price, and this should be a good way of clearing this inventory. It's an efficient way of doing that. Now, there's a couple of problems here, though. One of them is, that's unclear whether or not the ad at the other end actually works. And there's a lot in that statement, so let me unpack that a little bit.

There's the question of whether or not, if you see an ad, it is something that actually influences your behavior? There's not actually a whole lot of evidence to suggest that that might be the case. But also in many cases, an ad is delivered to someone who's not even real, so the prevalence of what's known as click fraud or ad fraud is huge in the economy. And so the ads might not even actually get to a real person, and in some cases, Google did this great study a number of years back where they said about 60% of ads that are delivered are never even seen because they appear in your browser, but below the fold. They're somewhat kind of weirdly placed so no one actually ever notices them. And so there's a lot of, I would say, interference between the purchase of an ad and its actual impact that people want, which is that someone would buy their service or buy their product.

Ethan Zuckerman:

But surely advertisers aren't idiots? So they're concerned about ad blockers that prevent users from seeing ads. Surely they know that some of these ads never actually got delivered. There have to be some effect to these or people would stop using them, correct? Or is that too much of an assumption?

Tim Hwang:

Well, I think it is too much of an assumption. It's actually really fun. My book is designed to be polemical and part of the success metric, if you will, of the book, was to piss off as many people as possible.

Ethan Zuckerman:

Always a worthy goal.

Tim Hwang:

Always a worthy goal, right? And I've heard from a number of ad tech people who have said, "Well, lots of people spend money on this, so isn't it proof that it works?" The title of the book is Subprime Attention Crisis, and very explicitly I'm trying to draw comparisons to other kinds of market bubbles that we've seen in the past. And one thing you learn looking at the history of market bubbles is that the fact that someone is spending a lot of money on something is not necessarily an indication that the market is actually functional.

Ethan Zuckerman:

Right. There was quite a robust market for very low quality distressed mortgages and for mortgage-backed securities that turned out at the end of the day to be largely worthless.

Tim Hwang:

Totally. Right. And I do think that there are a lot of perverse incentives in the ecosystem to keep this ball rolling. If you are an ad tech company, you're a Google or a Facebook or any of a large number of other companies that are in the ad tech space, you might very well want to really prove that this stuff, or show that this stuff, really works, and you have a lot of incentive to tout the benefits of this even if it doesn't. There's also extreme levels of opacity. We've built this incredibly large infrastructure for buying and selling attention online. The problem is, it's sometimes really difficult to understand why an ad is actually delivered to someone.

And then I think the final thing is actually, there's also empirical problems, which is, a lot of the academic literature on this topic of, do online ads work, land at the conclusion which is, they might work but the effect is so small that you have to run such a big experiment that in practical terms no one ever does. And so I think there's lots and lots of layers that prevent you both from knowing whether or not ads work and a lot of incentives for people to keep pushing the narrative that it in fact does work.

Ethan Zuckerman:

I absolutely take your point about the incentive, and I absolutely take the point that much of the work in this field is either sponsored by the ad tech companies or is only possible because of data coming from the ad tech companies. Certainly anecdotally, I have heard people on Facebook, for instance, talk about the fact that they flooded the market with Facebook ads and they got an uptick in sales. Is that not an existence proof that advertising works, even if it doesn't work necessarily as it says on the box? At least it does have some sort of an effect.

Tim Hwang:

Yeah, I think that's right. And to be clear, I think the claim of the book isn't necessarily that advertising categorically is ineffectual. Certainly we can point to examples in which it does. I think the question is, does that constitute... Is that the norm or is actually the outlier that we're pointing to? And we don't know because in many cases the ad tech companies have refused to give you the data that you would need to try to make those types of assessments.

Ethan Zuckerman:

Sure. And for instance, if we're at a situation where 60% of ads that are sold are never actually seen on a user's browser, and if we think about questions like click fraud, you might actually be dealing with a situation where an even smaller percentage are actually being delivered. Tim, talk about how click fraud works. Who is profiting when robots... I mean, I assume, distant relatives of your former legal colleagues are out there trying to make money by clicking ads on the web.

Tim Hwang:

Yeah, they're definitely in the same extended family, I think that's for sure. So, the way click fraud works is pretty straightforward and easy to understand. It's basically an advertiser goes to buy attention in this attention marketplace, and that attention marketplace says, "Oh, here's someone on an iPhone just about to open this app." Or, "Oh, here's someone on a laptop just about to look at this website." And basically the advertisers' algorithms are competing to buy the right to show things to those devices. Now what's interesting is in many cases, those devices may not actually be real, and the way the click fraud scam works is, you, the advertiser, purchases the ad, pays the person that's providing access to this eyeball, but it turns out that that person is just inventing all of the attention that they say they have. And so, rather than being delivered to a real person, it's either delivered to either a robot, or it's sometimes delivered to someone who doesn't care.

So you do have these incidents of these click farms where people actually just sit around and clicking on ads. They're not there to be interested in the products. They're mostly there to drive up the click-through rate so that they can pull the money from these advertisers. Ad fraud is a known problem in these programmatic ad markets and there has been work to try to deal with the problem. The perverse thing that I think is working against those initiatives though, is that in some ways this fake attention actually makes these marketplaces look more valuable than they actually are.

Ethan Zuckerman:

Sure.

Tim Hwang:

These marketplaces-

Ethan Zuckerman:

They get a cut of it.

Tim Hwang:

Totally. Right. And they take a percentage. And so, again, I think they have incentives to do just enough that their advertisers don't pull out of the system, but certainly not anything that would be structural to try to really resolve some of these issues. I think the other piece that I'll just mention really quickly is, I think that the bad guys, if you will, are really nimble. They're very clever at getting around these things, and in some cases what they are actually doing is they're spoofing like The New York Times. They're actually pretending to be known publications and the fact that that's a persistent problem, they haven't been able to stamp that out, I think just shows some of the problems with trying to resolve these issues.

Ethan Zuckerman:

Interesting. So they're pretending that the ad they're showing up on The New York Times? Is that the theory?

Tim Hwang:

That's correct. Right. So the Guardian actually I think filed a lawsuit against an ad tech company a few years back, in part because they discovered that people were selling ad inventory on the Guardian website that they didn't themselves even offer. It was like video ads on the Guardian which is totally made up inventory, and those scams can net millions and millions of dollars. And so, yeah, I think that's a big issue, for sure.

Ethan Zuckerman:

And of course there's also ways in which programmatic advertising works that can be deeply unpredictable, even when it's not scammy. Could you talk briefly about Sleeping Giants and their work on adding Breitbart?

Tim Hwang:

Overall, a lot of their work is connected to an issue that's known in the ad industry as a brand safety, and this is a good euphemism that people use to refer to the fact that this kind of ecosystem is so automated and sprawling that sometimes you're a brand that has an ad that lands up next to the video of a white supremacist, or something like that. And I use that as an example in Subprime Attention Crisis to make the argument that this ecosystem is actually quite opaque, because it can be very difficult to know why your ad ends up next to this type of content. And actually efforts at controlling it, like the kind of efforts of the biggest brands in the world to try to prevent this from happening hasn't been able to really stem the problem. In fact you have activist groups like Sleeping Giants being able to surface lots of examples in which, effectively, there's some really terrible stuff on the internet being underwritten essentially through this ad system, even without the willingness of the advertiser. They sort of end up there in a number of cases.

Ethan Zuckerman:

I found myself learning a little bit more about Sleeping Giants when I found out that ads for MIT distance learning courses were airing on Breitbart.

Tim Hwang:

Oh, my God.

Ethan Zuckerman:

I had to find the academic department at MIT and who was doing the ad buys and help them understand that what they were buying was inventory that was running across many different sites, but that Breitbart was one of them. And of course they hadn't consciously made that choice. One of the things that I love about this is the term brand safety. I was involved with building some of the earliest web ad tech in 1996 and 1997. Our problem was, since we were hosting personal homepages and we weren't very good at controlling the contents of that, we would routinely have an advertiser like Ford Motor company find their ad on a page filled with gay pornography and they would cancel a contract with us.

Ethan Zuckerman:

My boss gave me the job of figuring out how to stop that and my solution, of course, was the pop-up ad so that the ad didn't actually end up on the same page as the gay porn because we were so bad at stomping out-

Tim Hwang:

Sure, yeah.

Ethan Zuckerman:

So there's a lot of bad things that you can accomplish in the name of brand safety. Now, you're not worried about destroying the internet through pop-up ads. You're much more worried that this opaque and poorly understood market might be on the verge of collapse. Explain why that is, and explain what the consequences of that collapse might look like.

Tim Hwang:

Sure. I would put it this way. You have a market which is highly opaque. You have a bunch of self-interested actors that are pushing the value of the marketplace. At the same time, you have a lot of dynamics that are actively eroding, I would say, the "actual value" of the ad. That's the rise of ad blocking, ad fraud, like we've talked about. And also the fact that it just appears that, for a lot of big classes of online ads, people are just paying attention to them less and less.

Ethan Zuckerman:

Okay.

Tim Hwang:

And in my mind that is the ingredients that add up to a market bubble, which is, you can't see what's going on. People keep telling you that it's great, but actually, in fact, it's not great. And when the perceptions of markets snap to reality, the resulting panic can be effectively what a market bubble is. That's how the 2008 crisis occurred. And so the claim here is like, "Okay, so what is the trigger that causes a panic whereby advertisers say, 'Wait a minute, what are we spending millions and millions of dollars on?'" And you're already starting to see these kinds of screens, like, well, head-scratcher moments. Procter & Gamble two years ago said, "We're going to stop $200 million of digital ad spending." The end results to their bottom line was zero. Nothing happened.

And I do think that those types of events can trigger a looming sense that maybe this market is maybe not as great as we think it is. One of the things that I'm thinking about is like, "Does that set up the kind of grounds for a crisis? And what would a crisis like this look like?" Well, I think part of it is obviously the kind of companies that are facilitating this ad buying and exchanging, who are losing a lot of money. But of course I'm not so concerned that Mark Zuckerberg has like a billion less dollars. That doesn't really matter to me. What I'm worrying more about is that there's a whole ecosystem of media creators and communities online and social media platforms, platforms that facilitate all of this communication in the public sphere that are dependent on this financial reactor, this engine, to keep going and to stay solvent.

Tim Hwang:

And so I do think that a sustained downturn in some of these programmatic ad markets really has a big human consequence. And I think, again, you can look at the COVID-related downturn in the media as a sort of hint at what might be, because it's clear that a lot of major media entities don't really have the wherewithal to stand through a month or two of downturn. And so I do think that the cost there and the harm there is what I'm most worried about.

Ethan Zuckerman:

So if we imagine other companies looking at what Procter & Gamble did with that experiment and essentially saying, "Well, what if we cut a significant amount of our online spending, our online brand spending?" Your feeling is that it might have a net zero result on the bottom line of those advertisers, that what it might end up doing is triggering a very serious shift for a company like a Facebook or a Google, but that the most vulnerable actors in the ecosystem might be something like The Atlantic or Slate, or someone who's doing high quality journalism and is highly dependent on advertising as part of the revenue. Is that the ripple effect that you're mostly-

Tim Hwang:

Yeah, that is the ripple effect, and in part it's because programmatic advertising as a model has become something that there's an active effort to extend it into all different parts of the economy. For example, there's an effort right now to make podcast advertising based on the programmatic ad model. And so I do think that the implementation of this business model in lots of different places means that the ripple effects also could impact lots of people that we don't expect.

Ethan Zuckerman:

What's the solution? If we are dealing with a subprime attention crisis, in the banking crisis a number of regulatory changes were made. Elizabeth Warren spun up evidently quite effective consumer fraud protection agency, banks were subject to stress tests, and it's a whole lot harder to get a loan than it used to be. What are the things that we could do to respond to the subprime attention crisis?

Tim Hwang:

Sure. I do think that there is a need maybe on really two places. The first one is that I do think that the power of the federal government becomes very useful here in compelling transparency from largely the ad tech duopoly, your Facebook and Googles of the world. And the thing I'm responding to there was, you may remember a few years back Facebook was really encouraging what was known as the pivot to video, that, our data suggests that everybody is watching Facebook video. There's more and more people that are going to be looking at this, and we're going to tweak our newsfeed to really promote video, so fire all your journalists and hire a bunch of video producers. And later it turned out, and in future litigation, there's been claims that basically that they inflated those numbers on the order of 60 to 80%.

Tim Hwang:

And the problem is that the advertisers don't have any leverage to get third-party auditors to go in and be like, "Are the numbers that you're telling me about this ecosystem actually real or not?" And I do think that there's room there for saying, "Okay, what is the level of transparency we want out of these markets? And do we have to compel that disclosure?" So I think that's the first one. A second one that I have a vision of, that the government might be involved in is, again, to take a very, very wonky analogy, but in the capital markets there's something that's known as the Security Act of 1933. And essentially what the Securities Act says is, if you're going to offer stocks on the marketplace, we require you to give us certain types of information about that product.

If it turns out to be wrong later, you can be liable for it. And I do think that those types of incentives might be really powerful in a market like ads where there's just so much noise and there's basically no consequences for introducing bad data into the system. And so, again, the idea is not necessarily to try to figure out what ads are good or what ads will be effective, but in the very least for people to know what they're getting themselves into when they do buy an ad in these marketplaces. Now, I think the worry here is that the market, if you implemented these transparency measures, might be much smaller than it is right now, but I think it would also be a lot more stable in some ways if we didn't have all of the junk, like ad fraud, that's really inflating the scale of the market.

Ethan Zuckerman:

You mentioned earlier on in this interview that you were hoping to piss off as many people as possible. How's that going?

Tim Hwang:

It's good. People are annoyed. And what's interesting is I would say that they're annoyed on both sides of the debate, which is maybe an intriguing thing and maybe it's a little orthogonal to what you're asking about. But on one hand you have ad tech people who are like, "This guy's an idiot. He knows nothing." I was joking with someone earlier today. It's almost like debating with someone in a national security establishment because they're like, "Oh, we know ads work, but we can't tell you anything that would prove that [crosstalk 00:22:53]."

Ethan Zuckerman:

If you only could see what we see.

Tim Hwang:

Exactly, you would agree with us. And that's totally a bit of rhetoric that I have seen. What's also really interesting is a response to the book, I would say even tech critics have not been happy with the book. And they've been not happy with the book in some ways because it subverts the narrative that this advertising infrastructure is a kind of mind control ray that has the ability to shape society in a very discreet manner. And so I guess everybody's sort of been pissed off, for what it's worth.

Ethan Zuckerman:

Let's explore that for half a second because we haven't talked about privacy. We haven't talked about surveillance, capitalism, Shoshana Zuboff's name hasn't come up yet. Do you think that advertising, as it works right now, do you think that these large marketplaces for personally identifiable information, for behavioral data, for psychographic data, do you see a social harm associated with it? Or do you just think they don't work?

Tim Hwang:

I do think there is a social harm. I just think our critique needs to rest on something more than Mark Zuckerberg has the mind control ray, because I just don't think that's supported by the facts. But certainly you can still resist a huge surveillance infrastructure that we've built on privacy grounds. Yeah, I just don't think... The account of us having to believe this sort of industry's claims about this isn't necessary for us to resist the business model of the web.

Ethan Zuckerman:

So let's hope that this, of course, hugely listened-to podcast changes the minds of all advertisers out there. They pull enormous amounts from programmatic advertising, ripple effects, rack the industry, a new Biden administration, and of course, regulates transparency in the market, Google, Facebook, and others shrink dramatically in their market power. How do we create and deliver content on the web? How do The Atlantics and any number of small creative publications remain solvent? Once you've blown up the financial model of the web, Tim, how are you going to fix it?

Tim Hwang:

Yeah. I think this is obviously the huge question and the hard question, and I do think that... Let me frame it up and then I'll make an attempt at answering it. I do you think that part of the ethical problem with programmatic advertising is that it really is like corporate jet fuel. It's very difficult to find a similar model that just creates so massive of a return in such a short period of time. Of course, one of the reasons why it's been so popular is for that reason. Now the problem, though, is if you believe that a big chunk of that is fraud, what are we going to do? What's the Indiana Jones thing where we switch one thing for the other? The problem is we don't have a whole lot of other models that could guarantee that scale of growth that that's at that speed.

And so in some ways, probably implicitly what you're saying is, "Okay, to change the business model of the web is also to shrink the size of the pie." And then it becomes really a big ethical question, which is, "Okay, what level of shrinkage of that pie are we comfortable with?" And are we willing to do that even in a world where we say, "Okay, there's a number of well-meaning or well-deserving journalists that are funded by the secret system who will be out of work if you do this." And I think it's hard and this is one of the reasons, and this is my attempt to answer the question is... There's two things. One of them is, I really believe in what I call like a controlled demolition of this model. There's some people who are my friends who will say, "Who cares? Let's just blow this whole thing up."

But I think the human cost is too great to really blow up the bubble in this way. And so what you want instead to do is find ways of slowly stepping it down over time and allowing alternative models to grow and form. I think the other one is that I think our model of the internet in the future has to still include ads because I do think the model does have its benefits, right? Like accessibility to services. That's an indisputable benefit. And so I do think that really what I most hope for is an internet that's not just built on a monoculture, but instead has more of a diverse ecosystem of different business models. And the final point here is I do really think that advertising, in its dominance, has kind of smothered out this experimentation that we desperately need.

Ethan Zuckerman:

Tim, thank you so much both for being with us and for putting forward a book that really is challenging how we think about the space. I'm really thrilled that it's out there and I'm always thrilled by the incredibly diverse and powerful work that you are doing.

Tim Hwang:

Yeah. Thanks for having me on the show.