[choppaface]

Part of the plan but not part of Google’s consumer-oriented hype for it. Google has intentionally declined to state the fingerprint-joining and dollar impact to third parties because it would detract from the (already specious) message that FLoC improves privacy. Perhaps during some anti-trust hearings in the future we’ll see internal emails that detail how much $$ Google estimated that FLoC would take away from other advertisers as a function of what level of fingerprint joining they succeed at.

FLoC is NOT proven to be incentive-compatible with consumers and how they value their privacy. The only guarantee is that users are (on average) harder to distinguish within a cohort. Google absolutely studied the possible economic consequences of FLoC prior to announcement and they are hiding that study. Either because the results are crappy, the Google Ads employees are less competent than they were a decade ago, or both.

[Karrot_Kream]

> FLoC is NOT proven to be incentive-compatible with consumers and how they value their privacy. The only guarantee is that users are (on average) harder to distinguish within a cohort. Google absolutely studied the possible economic consequences of FLoC prior to announcement and they are hiding that study. Either because the results are crappy, the Google Ads employees are less competent than they were a decade ago, or both.

Could you walk through your ideas of the incentives here? I'm curious because while I like the idea of FLoC in general, Google is the last company I trust with this. Moreover there are a lot of details (such as cohort sizes) that could have the potential to mask identity and align incentives, but has been underspecified by Google so far.

[zwaps]

Normally, incentive compatibility is a concept from game theory and, more significantly, mechanism design (most known for auction theory).

In simple terms, it states that the mechanism that (say) Google offers to the consumer is such that the consumer finds it optimal to act according to their true preferences. In so called direct mechanisms, this means that the consumer would prefer to reveal their true type (information about themselves), rather than trying to fake being someone else. If this is true, it is much simpler to build mechanisms that are robust and achieve good outcomes.

For example, Hotel booking websites are typically not incentive compatible if they offer different prices to different users. Some user who is known to accept high prices will be offered higher prices in the future. However, if this tracking is imperfect, the price mechanism is not incentive compatible. The consumer would prefer to pretend to have a lower willingness to pay, and would thus be offered the same hotel for a lower price.

On the other hand, bundled goods (like cars accessory packages) may be incentive compatible, because even though each package is somewhat imperfect for each customer, they still prefer the "their" package over another one that is less costly. That way, the car company can maximize profits.

In mechanism design (for example, in the design of auctions), incentive compatibility is usually a requirement to derive mechanisms to that maximize the objective function. This is so, because there are of course infinitely many complex mechanisms one could propose. But, under some conditions, incentive compatibility ensures that the "optimal" mechanism will have a simple structure.

Finally, there are substantive reasons: If a customer has to act against their true preferences, the system is likely unstable and gameable. And also, people might walk away.

Some people also consider the participation constraint a part of incentive compatibility. That is, the consumer must at least gain as much as not participating at all.

In the present case, the participation constraint is violated. The consumer gains nothing by being targeted and, if asked, would rather opt out. In that case, Google can design whatever they want, an informed customer will reject it.

Google and Ad firms know this, which is why it is virtually impossible to truly opt out. The participation constraint is knocked away by deceptive practices and dark patterns.

[choppaface]

Right! This is the notion of incentive compatibility I'm trying to draw upon. One problem here is that not all of the incentives and objective functions are visible. That said, Google I'm certain has approved FLoC for launch based upon a mechanism, and so far I really see no evidence that that mechanism can actually be favorable to the wider public (not just "privacy geeks"). Would love to see Google's actual research that was provided to the launch committee, but I might as well ask for access to Area 51.

For the majority of users, I do believe tracking has some upshot, at least for reducing ad spam and ad fraud. It's really hard to prove this without a counterfactual study, but if you compare Facebook ads with the crap on news sites today, I think that situation is illustrative of a place where Facebook's improved targeting saves the user from a lot of spam / fraud. And I'd argue that outcome is a result of Facebook having a monopoly over their own targeting and identifiers.

[choppaface]

As a business owner who wants to post online ads, I want the providers to be as competitive as possible on price and quality. As a user, if I'm going to have ads targeted to me, I'm hoping they don't blink, slow down my computer, or are the utter trash you see on today's poor news sites. As as user, I want competition to cause ad targeters to filter out ad spam-- it hurts the properties where they place things.

Any new ad provider (one that does not have 20 years of user data like Google, or 15 years like Facebook) will not be able to fingerprint as well with FLoC as they could with cookies. Existing FANG competitors like Verizon, AT&T, Comcast etc have lots of historical data offer and even some FLoC-like products but if Google's FLoC replaces cookies then it's harder to derive value from, say, Verizon's FLoC-- simply because a cookie is a more effective fingerprint than a Google FLoC.

So FLoC is anti-competitive and just further supplants Google's monopoly. Does that really impact consumer privacy? Well, if the ad market is less competitive, the level of privacy that the market can support will be less consistent. There will be people like Apple who make so much money off hardware that for ads they're fine with the unobtrusive choice of selling Google default search engine rights (versus, say, the loads of crapware that can be found in some Android distributions). But there will be lots of smaller ad players who will get more desperate for ad spam and more shady ways to target you with what little data they can get. For smaller properties, perhaps Google FLoC is so useless for targeting and lift that everybody switches to requiring a sign-up with a phone number SMS confirmation.

_Can_ FLoC be incentive compatible? What are the actual likely economic consequences? Google knows this (or they think they know), hence FLoC got launch approval. But Google won't share those details with you. They want you to think FLoC is just great for privacy. This is how Sundar addresses his position that "we need to work hard on user trust."