13 Jul 2024 Search

(update 25 Jul 2024: add links to FTC surveillance pricing story)

Every advertising event is full of thought leader insights about privacy-enhancing technologies (PETs) for ad personalization and reporting systems. Somehow Big Tech, adtech, and martech are all fired up about projects for reimplementing personalized/surveillance advertising, but this time with a bunch of complicated math added in a way that makes it hard to identify or track an individual.

In the real world, though, individualized tracking is not the top advertising privacy problem, and it might not even be in the top five. Arielle Garcia, director of intelligence for Check My Ads, said, Privacy-enhancing tech doesn’t make creepy and disruptive ads less creepy or disruptive in the eyes of the average user. And the user research backs that up. Jereth et al. find that perceived privacy violations for a browser-based system that does not target people individually are similar to the perceived violations for conventional third-party cookies. Co-author Klaus M. Miller presented the research at FTC PrivacyCon (PDF):

So keeping your data safer on your device seems to help in terms of consumer perceptions, but it doesn’t make any difference whether the firm is targeting the consumer at the individual or group level in the perceived privacy perceptions.

This might not make sense if you compare just the information passed through PETs to what is possible to do with a third-party cookie. It’s fewer total bits of information, so the users should be getting more privacy and like it better, right? Not so fast. The real problems that people complain about when they raise privacy concerns are more about information asymmetry.

People tend not to want to participate in markets in ways that give their counterparties too much information. Imagine going on vacation and visiting an old-fashioned rug market—and showing every rug merchant your budget, shopping list, and departure date. Information imbalances are not how sustainable markets work. And that’s just legit markets. The situation is even worse on today’s scam culture Internet, where the old fake it til you make it has been replaced with make it, then fake it even harder. The pervasive risks of scam culture are mostly group-level personalization risks and not the consequences of being individually identified.

Some PETs just measure ad performance and don’t personalize. But when the same measurement system covers both harmful and win-win ad placements, it creates incentives for advertisers to chase the best possible metrics while creating the least possible value for the user. In a sustainable system, the user, or some party that the user trusts to provide an ad-supported resource, would be a data gatekeeper to keep data from leaking into the negative-sum corners of the Internet. People choose to provide their information to parties they trust—they don’t broadcast info about themselves to trusted and untrusted parties on the same terms. If browsers offered a privacy-preserving way to share corporate information with web users who visit the company site, would the corporate IT department leave it turned on?

how things got this far

PETs started to go mainstream in the advertising scene back in 2019 when Google announced the Privacy Sandbox project. (Apple was first with wide release of a PET ad system, Private Click Measurement in 2021.) At the time, some people picked up on PETs as an alternative to platform dystopia, a way for independent retailers and publishers on the web to make a compromise and continue to do business without a hard dependency on fixing the Internet’s deeper monopolization, discrimination, and fraud problems. But in the about five years that people have been been arguing about web ad PETs, most recently over Google’s ongoing effort to make their PETs acceptable to the Competition and Markets Authority (CMA) in the UK, a big PET win is looking less and less likely. After all the work that has gone into PETs, not only do big risks remain, but—and this is super awkward—the companies working on PETs are still the same Big Tech companies that PETs were supposed to…protect us from?

Meanwhile, in the years since the PET trend began, mainstream privacy laws have gone off in a different direction, and focused not on obfuscation and math, but on data minimization and on enabling people to find out how companies use data. Data subject access rights, or “right to know,” are a key component of modern privacy laws and make it possible for web users and for organizations working on their behalf to detect patterns of illegal discrimination. In a 2019 settlement, Facebook agreed to stop using age, gender and ZIP code for housing, employment and credit ads. Reporters were later able to track the company’s progress. Privacy researchers and advocates continue to benefit from “right to know” processes today. A PET environment, however, limits access to data, making algorithmic discrimination hard to detect.

Much recent privacy news would have been as bad or worse in a PET environment. Grindr users aren’t suing over privacy because they were individually identified, they are suing because they were identified by HIV status, a fact common to a large group of people. Reviews of Google’s Privacy Sandbox, have, for quite a while, anticipated regulator concerns over algorithmic discrimination and transparency. The W3C TAG, in a review of a Privacy Sandbox” sub-project, wrote,

The Topics API as proposed puts the browser in a position of sharing information about the user, derived from their browsing history, with any site that can call the API. This is done in such a way that the user has no fine-grained control over what is revealed, and in what context, or to which parties. It also seems likely that a user would struggle to understand what is even happening; data is gathered and sent behind the scenes, quite opaquely.

Under privacy laws, users have the right to access not just their raw data, but the inferences made from it—a capability that will be difficult to retrofit into PETs. Google posted a FAQ stating,

Chrome can and will take steps to avoid topics that might be sensitive (i.e. race, sexual orientation, religion, etc.). However, it is still possible that websites calling the API may combine or correlate topics with other signals to infer sensitive information, outside of intended use. Chrome will continue to investigate methods for reducing this risk.

No results have been posted from this investigation so far. Someone will probably get a Best Paper award at a conference for solving algorithmic discrimination in a PET-based ad system…eventually. Until that happens, PETs will struggle to meet basic transparency requirements in more and more jurisdictions. The data obfuscation problems introduced by PETs will also create compliance challenges in the area of competition policy. In a recent Google report (PDF) to the CMA, the company acknowledges that compliance with the Digital Services Act (DSA) by “Privacy Sandbox” publishers and advertisers remains an unsolved problem.

alternatives

So what do we do instead? PETs may continue to be valuable in fields like software telemetry, where the end result is all users receiving the same bug fixes, not different treatment of different users. But ads are a harder problem. First of all, PETs are not a shortcut to solving some important platform trust issues.

I see that your industry sold my parents a retirement scam, put malware in my search results, and showed fake repair shops on a map of my neighborhood…but I totally trust your answer to this complicated math problem. — no one ever

The hard part about running any Internet service is the moderation (including ad review if it’s an ad-supported service). At some point Big Tech management needs to get over its dogmatic union-busting and sign fair contracts with the moderators and all the other people who do the high-skill human side of their operations. You can’t fix a reputation problem with math.

As far as how to make the ads work, it’s possible to get rid of third-party cookies and other problematic identifiers like mobile ad IDs without also taking on an open-ended research project. Instead of starting from what data flows are acceptable to today’s platforms/adtech/martech and trying to stick enough math on them to make them acceptable to users, start with how people choose to participate in markets and automate it. The Lindy Effect is a real thing in marketing. Ideas get tested out all the time, but the practices that tend to persist are the ones that have endured many generations of marketers, companies, and technologies. So the history of marketing practices that people have been willing to accept in the past is probably the best guide here. Realistically, markets will always have high and low reputation sellers, and the advertising most likely to persist will be whatever can pay for itself in (ad-supported resources + economic signal) > (cost in attention + resources + risk).

Finding something successful and sustainable and adapting it to the web is good for more total ad revenue over time—even if in any one transaction you get to capture less of it. A browser is a user agent, which means it does what users would do for themselves if they had time. People like to buy stuff, and prefer to buy better stuff. How can browsers help?

More: Sunday Internet optimism