When dealing with annoying people, there are at least two possible choices: ignore them or contain them. The former is a popular advice for combatting Internet trolls. The question is, shall we start to move slowly towards the containment tactic?

trollfaceImagine a group of people wandering in your neighborhood. They are not criminals, they are not dangerous at all. They just tend to waste everyone’s time. They are the ultimate masters of time sinks. Of course, we can (and we should) continue to ignore them (“don’t feed the Troll” mantra) until they are going away.

An alternative will be to invite all of them to a basement, get some pizza and other communal food, and ask a very smart robot (with top-of-the-line artifical intelligence) to keep them busy for as long as they care. The idea is very simple, the more time they spend in that contained environment, the less time they have to waste someone’s else time.

While such a robot is not available yet, the strategies can be still developed and validated in a smaller scope, perhaps as an online bot. In many cases, we need to borrow the elements from the trolls themselves. What follows is a random collection of some ideas, I’m sure these can be improved further.

Timing is always important. Tracking past responses makes it possible to deduce the likely schedule of the trolls. The bot needs to keep the trolls busy at their peak time.

Asymmetric effort ensures the effectiveness. Our bot should spend 5 seconds writing a reponse that keeps the trolls busy for half a day.

Provocation should be a recurring theme, the bot should ignore any baits from the trolls (wouldn’t it be easy?) and try to always get under their skin.

In our tech world, we have witnessed computers capable of understanding complex web pages, playing chess, winning a quiz show, and serving as a personal assistant. Will we see human trolls vs troll bots anytime soon?

  • s


  • Greg Baker

    Having found and so likes your “kinetics” series, brought me back to read more of your work, Ariya Hidayat. I truly love this post. To that end…

    There are some simple ‘rule-based’ and conversational AI packages one could use as a basis for such a bot. Also, grabbing a quick db of their internet foot-print and with it, building a 3rd or 4th order Markovian chain to echo their own language back at them via the AI’s conversational mode should do the trick, I would think. This should keep them busy for a long, long time, for as we all know, the think highly of their own words in print, no?