Trolls Gonna Troll

These once obscure Internet pests have gone mainstream, indeed geopolitical.  Pretty much everyone knows what an online ‘Troll’ is, and there is a lot of hand-wringing and suggested ‘solutions’.

But before we can ‘solve’ the problem, we need to understand it.  In particular, what do trolls do, and why?

Last year Hal Berghel and Daniel Berleant described “The Online Trolling Ecosystem”, classifying the variety of trolls they have seen [1].  As they point out, the internet was generally intended to increase access to information, with little concern for sources or authenticity.  In that sense, trolling is clearly a feature, not a bug, in the original concept of the internet.

Berghel and Berleant see the essence of trolling as misinformation, and, as we are all aware, this can be turned to try to influence public opinion.  It also turns out that creating an automated troll bot is easier than a believable chat bot.  Obviously, low quality speech is pretty automatic, while authentic communication is more complicated.

The most important part of the article is their “taxonomy of trolling”, which identifies 22 types of trolling.  There is quite a list, and quite a range.

• • Provocation trolling
  • Social-engineering trolling
  • Grooming trolling
  • Partisan trolling
  • Firehose trolling
  • Ad hominem trolling
  • Jam trolling
  • Sport trolling
  • Snag trolling
  • Nuisance trolling
  • Diversion trolling
  • False-flag trolling
  • Huckster trolling
  • Amplification/relay trolling
  • Rehearsal trolling
  • Proxy trolling
  • Faux-facts trolling
  • Insult trolling
  • PR trolling
  • Chaff trolling
  • Wheat trolling
  • Satire trolling

These categories are not necessarily mutually exclusive.  A troll might be motivated by both personal amusement and political ideology, for example.

The important point is that ‘troll’ isn’t a single or simple thing.  Some trolls are scammers (or just marketers), out for money.  Others are political actors, aiming to drive public opinion and gain power.  Others are just larking, amused by their ability to cause trouble and gain attention.

In short, you can’t “solve” trolling with a single approach.

Trolling is enabled by technology, especially cheap and easy network access and anonymity.  In recent years, we have seen the rise of global monopolies which are probably to big to control. In this sense, trolling is a parasite on the “legitimate” uses of the network, and consequently difficult to suppress.

“But perhaps it’s not as important to understand the psychology underlying trolling as it is to avoid being manipulated by it” ([1], p.43)

Most of us would like to see better authentication of sources:  it shouldn’t be easy to pretend to be a widely known source.  There is a lot of interest in various kinds of fact-checking and validation, which “work against the interests of trolls” (p.43)

However, for these methods “to be effective, the detection time must be near zero because the reaction time required to re-tweet, forward, and so on is negligible.” (P.43)  Trolling flashes around the world while fact checking is putting on her boots.

I don’t think that trolling can be eliminated.  However, this taxonomy is probably a useful guide for designers. With an understanding of different trolling, it should be possible to harden systems against some kinds of trolling. For example, for pay systems can be relatively resistant to the most trivial amusement trolling—it’s just not affordable.  And, of course, moderated forums can filter out certain kinds of junk, if there is a will to do so.

This might also be useful information for consumers.  I could imagine a rating system that tells you just how vulnerable a particular system is to different kinds of trolling.  Initially, this is easy to do:  everything is pretty much 100% vulnerable.  But it might be good to know that a particular news service is, for instance, low on ad hominem trolling, even if potentially high on PR trolling.

---

1. Hal Berghel and Daniel Berleant, The Online Trolling Ecosystem. Computer, 51 (08):44-51, 2018. http://doi.ieeecomputersociety.org/10.1109/MC.2018.3191256