I'd start by not approving all the scrapers and spammers and scammers. I'd avoid sites that get the majority of their traffic from reddit/etc, if they don't have natural traffic but they're consistently featured on social media sites they're probably spammers.
Once you take all the trashy people and their trashy sites out of the equation I think it would become a lot easier to detect fraud.
I would focus on:
1) high quality publishers
2) more interesting formats ... people used to talk about banner blindness, these days AdSense has to be pretty much shoved between sentences to be seen
3) actual support
Something probably like The Deck but more open.
Right now advertising is interrupt-driven. Could one do the opposite? What you are suggesting seems like a frontal attack and that may not be the best tactic.
What is the goal of advertising anyway?
For the ad-buyer, the goal seems to be increasing sales.
For the consumer, the goal is ... what?
When it comes to websites anything that diverts my attention from the page that I came to look at is going to interrupt me so I don't know if that's really a good criteria.
Take the easy case of a consumer ready to buy something this minute. Say, they want a colorful winter jacket for men, and they are willing to pay anything for it.
Can they find it with advertising? Can they find it with search?
One goal of advertising for the consumer is that it helps them make a decision. Which of these X jackets should I consider?
Now, is being interrupted the best way to make a decision?
On the other hand, if a site gets 99% of their visitors from search engines, it makes you wonder about the quality of the site
Another thing is that advertisers are tied to the same medium for long periods of time. It would be great to have a platform that could switch an ad (banner ad in this case) from one portal to another, depending on price, date and web traffic conditions.
Really, I'd like the original Google ads back again, you know, the text-only ones. Though perhaps not run by Google any more.