Specifically, I referenced two proposals for privacy-preserving ad personalization, FLoC (https://github.com/jkarlin/floc) and TURTLE-DOV (https://github.com/michaelkleber/turtledove), and their response was that privacy wasn't the issue, the effects of personalization are harmful in themselves.
For privacy I do agree that you can't just do it with legislation and you also need technology. But if you think the issue is personalization itself, regardless of how privately it is implemented, then legislation is the critical playing field. Blocking personalization on the open web through technical means while allowing it in walled gardens means you (a) still have most of the claimed harms of personalization but also (b) massively disadvantage the open web.
Reading through those links on FLoC and TURTLE-DOV, those are interesting concepts. My worry is that even though the ad selection is done in the browser, it will still leak significant information. For image/video ads, it would require bloating the amount of bandwidth used to serve the ads, or would leak information about which ad was selected by the client.
The two are more or less intertwined. "Personalization" leaks information about people. It has also allowed advertisers to promote material targeting (or eliminating) otherwise protected groups.
People have been outed for being pregnant and for being gay, by outside observers noticing what promotions they were getting.
Unfortunately back when I thought of this I wasn't aware of browser fingerprinting or all the other nefarious things the advertising industry is doing. Fingerprinting turns both FLoC and my idea into tracking vectors, so that's out of the question. The authors acknowledge this problem in the readme.
Turtledove is a longer spec and I haven't had the time to read it thoroughly, but while the idea is good, I see some potential problems - first off, this still involves executing JS to do the "ad auction" on the browser which could have security implications (they explicitly mention that JS should be prevented from talking to the outside, leaks would be vulnerabilities in this case, and I don't trust the ad industry to be nice and not attempt to exploit them); there's also the issue that I just don't want my device to expend processing power and expose me to potential security vulnerabilities for ads; I want ads to be served as static images from the website without the whole bidding thing. From an advertiser's perspective, I'm not sure how fetching the potential ads in advance and moving the auction on the client device would work - my understanding is that auctions are in real-time. I'm also not sure whether exposing all the ads to the public (even the ones that ultimately don't get chosen at the auction) and the auction details would reveal information that advertisers consider sensitive.
Finally, if you look at my other post here: https://news.ycombinator.com/item?id=22680559 you'll see that I have other problems with online and targeted ads besides the privacy and security aspect.
Unlike print and TV ads, there little-to-none human review, no accountability (Google isn't liable if they serve me a scam ad if I fall for it), and targeting allows to target scams to only the people most likely to fall for them while hiding these ads from the savvy people who would recognise the scam and report it. At the core, this isn't a tech problem but a law problem, but short of a law it can be solved by tech if we heavily block the current model of ads (targeted and served from a central ad network) so it becomes unprofitable and publishers & advertisers have no choice but to host static ads on their own domain, thus becoming liable if they host malware or something illegal (this would hopefully encourage human review).
It was critical to election tampering in 2016, it's been used for racism, sexism, and every imaginable sort of scam.
The only way personalized advertising would be remotely ethical is if it were explicitly opt-in. I notice that's not a thing on your list of steps Google is planning on doing. No surprise.