Comment by cosmic_cheese
2 days ago
I don’t think that bot mitigation is nearly the ordeal that the social media giants pose it to be, so long as the desired result is keeping such activity minimized (practical) and not entirely eliminated (impossible), especially in this era of increasingly capable lightweight language models.
Taking Xitter as an example, there are many tells that are visible even to readers with limited info that should be as plain as day to the platform owner. Many are barely even masked. The problem is that for ad supported social media, all incentives align with proliferation of bots, especially if they’re paying you to boost their reach. They’re doing all the hard work of genetically engineering perfectly engaging content for you; who cares about the deleterious effects they’re having on society?
This is why surveillance style adtech must be made into a massive political liability.
Bot detection in the case where the activity is not human like e.g. spam, is quite easy to detect. But the problem here is that a foreign influence bot is not like that at all. It is designed to act like a human. And the areas that foreign bots attack are also political issues that the domestic population cares about, so you can't really filter by content either.
You could still mitigate this by blocking accounts that use VPN IPs or requiring invasive IDV for accounts, but isn't this sort of thing exactly the type of authoritarian user monitoring that we would like to avoid?
I don't really get you point about removing surveillance style ad tech. Wouldn't that just make it even harder to detect foreign influence accounts?
Following the earlier example of Xitter, there are loads of accounts that are obvious bots or foreign propaganda outfits to anybody who's been using the site for a while, complete with a reasonably good litmus test: whenever you suspect a post is from one of these, check the posting account's location. For me the accounts that raise internal red flags are almost always posting from SE Asia, somewhere in Africa, Russia, etc.
That means their content and profile quirks follow clear patterns that are not that difficult to distinguish. Xitter just has no financial or legal incentive to do so.
> I don't really get you point about removing surveillance style ad tech. Wouldn't that just make it even harder to detect foreign influence accounts?
The difference is the invasiveness of the data gathered and who it's given to. For bot/troll/propaganda mitigation, invasiveness is comparably minimal (survelliance advertising isolates marketable qualities such as race, orientation, religion, marital status, purchasing habits, etc; none of these are necessary for mitigation) and gathered data never needs to leave the possession of the social network.