cathrowan

https://blog.zgp.org/attribution-cartel-update/

Previously: a Terminator ending for Google “Privacy Sandbox”?

Things seen pretty quiet over at the attribution cartel. It’s not secret like the lysine thing was—they have a GitHub repository at w3c/attribution: Attribution API and everything. They’re still going, as far as I can tell, but not getting a lot of ink.

The attribution cartel is learning from the failure of Google’s “Privacy Sandbox”. Where Google’s 2019-2025 project was noisy, the attribution cartel is quiet. Where Google tried to do a full ad stack in one browser, the attribution cartel is doing one key piece of the stack across multiple browsers. So far, it seems to be working. My best guess as to why is, as they say, all this. The companies involved have so flooded the zone that dumping one more load can happen without attracting much attention. With the whole “AI” infringement situation, the fraud and malware crisis on search and social, and the economy-wide trust collapse in general, it’s not a surprise that this thing is moving but not raising much of a stink.

So what’s so bad about the attribution cartel?

Fraud part 1: Designing attribution tracking without fraud protection wastes advertisers’ budgets and deprives legit sites of revenue.

Fraud part 2: What’s worse is that fraud creates incentives to “front-run” conversions, so adds more incentives for risky data practices. For example, a future fraud operation would be able to “win” attribution by running a covert speech-to-text app using smart TV microphones to identify households already about to buy. Turning on the mic isn’t part of the “Attribution” proposal but “Attribution” provides the incentive and cash-out mechanism to give more players a reason to do it. (More: The Hidden Dangers Of Privacy-Preserving Attribution – And A Smarter Solution)

Boundary testing probably best describes the Google “Privacy Sandbox” saga and is happening again. On August 22, 2019, Google pitched “Privacy Sandbox” as an alternative to fingerprinting. The deal was if people accept some in-browser adtech, then Google won’t have to fingerprint them. After the project had gotten some momentum, though, Google changed their policy to allow (and for practical purposes, encourage) fingerprinting. The “Privacy Sandbox” project may have been completely sincere from the point of view of the individual Googlers advocating for it in public—but it was part of a program of centralization of power.

There is a lot of online advice about how to handle boundary testing. See Why People Push Boundaries & How to Protect Your Peace by Brooke Sprowl. In general, it’s good to clearly communicate the boundary and consequences for violating it as early as possible. Thanks to early and creative development of spam filters, the Internet of the late 1990s and early 2000s did a much better job with setting norms on email spam than today’s Internet is doing on modern Big Tech schemes. But there’s still time to get out in front of this one. On the optimistic timeline for the attribution cartel, widely adopted privacy tools are already set up to block it before the browser origin trials start.

Breaks Right to Know: A common feature of privacy laws is right to know (RtK). Sometimes called a data subject access right. Policy makers and Big Tech have different interests here.

• Big Tech wants to profile you and deliver deceptive, personalized ads and offers. (they have to, because growth stocks) This goes for everything from small-time personalized/surveillance pricing up to fraud and large-scale political misinformation.

• The drafters of privacy laws typically want to expose deception and discrimination, so they give us a variety of RtK options. Although RtK can be time-consuming, it’s effective as part of a research program or for some kinds of lawsuits..

The attribution cartel recognizes that a well-designed RtK will expose a lot of illegal or actionable practices. So the “Attribution” proposal hides discrimination by moving measurement into complex ML systems. Right now the “Attribution” proposal covers only measurement, not ad placement. But it measures high-discrimination and low-discrimination placement choices on an equal basis.

For example, say an employer has disability and language prejudice in the hiring process, and uses two demand-side platforms to run ads for seasonal employment. One DSP uses a conventional ad placement methodology. The other DSP uses ML to avoid running ads to people with disabilities and speakers of certain languages. The “Attribution” proposal lets that discrimination ML hide in the same way that it lets a front-running attribution fraud operation hide. If someone suspects that they aren’t getting certain job or housing ads, the conventional adtech system gives them a thread to pull on, and sometimes they can make a case out of it. Adding “privacy-preserving” math to attribution tracking obfuscates the evidence, allowing Big Tech to keep serving the advertisers who prefer to discriminate.

Forum shopping: Attribution cartel members chose to work in a forum where Big Tech is present, but advertising experts and smaller competitors are not. Current best practices on how to measure advertising results (check out Geo RCT Guide from Central Control) and discussions of attribution requirements at the IAB Tech Lab’s Project Rearc don’t match up very well with the capabilities that the attribution cartel offers. That’s not surprising—Central Control is working for their clients, as are many IABTL members, and the attribution cartel is working for the attribution cartel.

Platform oligopoly is built in. Key features of the proposed system assume a power imbalance between large platform and small advertiser. One example is the so-called “privacy budget.” If an advertiser makes a mistake on one query, they can be locked out of future queries regarding the same ads. A large monopoly or oligopoly company—the kind of firm that can fail to provide user support and make users go to the Attorney General for account issues—can enforce this kind of one-sided requirement on a smaller advertiser, but in an environment where advertiser, publisher, and intermediary companies are of similar size, that doesn’t work. The power imbalance built into the “Attribution” proposal would lock us in to a centralized, oligarchic Internet future.

Attribution reporting criteria don’t match market norms. When people participate in markets, they want trusted counterparties to have more information, while protecting information from being disclosed to parties they don’t trust. For example, you probably want your local newspaper to have more information on you than some random site that your uncle sends you a link to, or a creepy social site you click on a link to. The attribution cartel doesn’t match this expectation, so facilitates commodification of ad contexts.

User research: Jereth et al. found similar perceived privacy violations for a browser-based advertising system as for conventional third-party cookies. and Martin et al. found misalignment between privacy-enhancing techniques and the privacy that people expect. Some users found conventional cookie tracking more acceptable than the more complex system. [F]or improving services across contexts, consumers judged the use of raw data as more appropriate compared to using inferences based on that same raw data.

Environmental impact: Protests against data centers in the USA are somehow one thing that a divided nation can agree on. The attribution cartel is proposing burning a lot more cycles than conventional adtech/martech—not to get privacy as experienced by people, but in order to add some mathematical properties that (see user research above) people don’t want. (W3C already had a big argument over proof of work for identifiers—W3C overrules Google, Mozilla’s objections to identifiers—and the attribution cartel is setting things up for another round of sustainability debate.)

Single country dependency. At a time when more and more IT decision makers are looking for alternatives outside the USA (France to ditch Windows for Linux to reduce reliance on US tech—that’s a big project.) Play La Marseillaise. Play it! the attribution cartel is going the other direction.

The attribution cartel ate somebody’s pet cat in Ohio! Just seeing if you were paying attention. (Just watch, this is going to be the one fact from this page that makes it into the “AI Overview.” Oops.)

“Pivot to Video” again…really? Laura Hazard Owen at Nieman Journalism Lab said it best. News publishers’ pivot to video was driven largely by a belief that if Facebook was seeing users, in massive numbers, shift to video from text, the trend must be real for news video too — even if people within those publishers doubted the trend based on their own experiences… Read the whole thing.

https://blog.zgp.org/attribution-cartel-update/