Scam bot operators will just use stolen credit cards - or even easier, iTunes gift cards that they get from the victims of their scams - to pay to "prove" that they aren't bot accounts. For the fake followers/interaction bot "services" it increases the cost of operating, but I doubt they spin up a bunch of new accounts for every client - that $1 per account can probably be spread out pretty thin. I don't see this solving the bot problem any more than prioritizing paid account replies did (it didn't work at all for that).
The context of the quoted section of the article is about what an acceptable false positive rate would be, not about what situations drunk driving would be acceptable.