The real question is how many of those bots are spambots and bots attempting to pose as real people, as opposed to bots that are open about the fact that they are bots? There's plenty of top-shelf novelty accounts on Twitter that are also bots.
That's one reason why a no bot policy wouldn't work.
A bot posting articles from your favourite news site is actually useful. A bot which prints out Trump's tweets, burns them and then tweets out the video [@burnedyourtweet] is mildly amusing (and some pretty smart engineering). A bot which automatically replies to every tweet by $personality with a meme is usually an 'orrible troll, but occasionally amusing and often owned by someone who's very convinced the First Amendment applies to Twitter. A retweet bot which does a plausible impression of an angry Trumper might be indistinguishable from the real thing without investigating its network, and I don't think eliminating the real thing would be a great move for Twitter to make.
The other reason is that Twitter's review process is more or less completely random in its responses (I'm reminded of the girl that reported literally the same abusive tweet directed at her from three troll accounts, and received a different response for each one)
This is why Reddit style voting is super useful. Twitter suffers from a noise problem as a side-effect of focusing so much on newness and sorting by date (I know they've made recent changes here).
This community driven ranking seems like the best solution to low quality bots posting. A modding driven centralized editing process comes with plenty of issues, as we've seen on Reddit. And yes Bot's provide plenty of value, far outweighing some niche political accounts making the rounds.
Not to mention, a single person creating thousands of bot accounts is easy, you need those thousand other bots so your one primary bot has enough followers/likes. But in the press that will be counted as 1000s instead of 1. Not to mention getting those bots to get a meaningful audience is very challenging.
I have plenty of fake 'bot' followers, and 99.9% of their followers are other bot. It's largely a bunch of bots tweeting to other bots.
The main risk is hashtags being gamed and people buying voting rings/followers... which has been a problem on the internet forever.
The open bots have an incentive to make one bot for each usecase, one person trying to make spam bots has an incentive to create as many bots as possible. I have no numbers on the subject, but I would be shocked if it were anywhere close to fifty percent.
Frankly, I'd expect the percentage to be significantly less than 50%. Determining good from bad bots without a manual check is an interesting problem, though, and it's not cool on Twitter's part if they ban legit bots that people find useful/entertaining on top of the bad actors.
> A researcher finds 1,600 bots tweeting extremist posts in U.S. elections also spread anti-Macron sentiment in France
A second quote from the article shows a much larger bot population.
> Research from the University of Southern California and Indiana University shows that 9 to 15 percent of active Twitter accounts are bots.