In a less techinical point of view, ads are probably a net negative for society. People are buying things they don't need, spending time they could be doing other things and just having a worse experience they could have were not for content farms and other practices incentivised by ads. I do see the value of ads bringing services to people who wouldn't have money to pay for them otherwise, it's income distributions of sorts, but I think we can do better.
Anyway, interesting read about ip blindness. As long as the CDNs and proxies are not controlled by the same company that owns the ad networks, then it could work out. Though it's hard to find the correct incentives for the right people to own the right parts of the network. Another alternative would be something like onion, which is more distributed (although quite wasteful of resources).
Yes, there are a lot of user identifying bits in an IP address. Chrome has two proposals: https://github.com/bslassey/ip-blindness I'm not sure what other browsers are thinking?
> That's considering the browser doesn't leak out any other potentially identifying information.
Which they definitely do. All the browsers are working on figuring out how to thwart fingerprinting, and it's really hard. I am glad, at least, that we were able to get Google Ads to publicly commit to not fingerprinting.
> when you click on the ad, they know one interest about you and, if you clicked in ads from other campaigns they run, they may reconstruct your profile well
Yes, when people click on ads in Turtledove the advertiser does learn something. This is a huge improvement to the status quo where advertisers learn things just by bidding, or an intermediate stage where advertisers learn things when they win an auction -- users don't click on ads very often, so the amount of information leaked this way is very low.
Exactly how much information the advertiser is able to learn on a click is still very much up in the air, so if you have views on this you might consider participating on the repo?