Markov, Trump and Counter Radicalization
Gary Short gave a very quick 10 minute lunchtime talk at Developer Developer Developer 12 called “Markov, Trump and Counter Radicalization”
He began by saying the topic is not as frightening as it first sounds:
Radicalization is persuasion from moderate to radical views.
Counter radicalization is persuasion away from radical views to more moderate views.
There are a couple of risks involved with attempting counter radicalization which Gary recommends we read up on:
But this is not so much a psychology talk as it is about machine learning. The basic idea is to use known good texts, focus groups, known good text, testimony from de-radicalized individuals.
We use Artificial Intelligence to generate a Markov chain
Given n words, return me the next w words which have the highest probability of following
We create social media bots to post the generated text and then analyse the responses to those messages to see if the person’s views have become more moderate. Then this information is fed back into the program to help it generate more effective moderating messages in the future.
Gary has written a “TweetLikeTrump” project. It uses Trumps speeches before he was made president as the input that the program uses. The generated text has same statistical fingerprint as Donald Trump.
The TweetLikeTrump application is on GitHub. It’s open source C# project and uses the GPLv3 licence.
10 minutes was not really enough to do justice to this topic so I hope that Gary has the opportunity to do a longer talk on it soon. In the meantime check out the code and have fun!
Although the project is a bit of fun and kind of silly, there is also a serious side to it and with enough further development the technology has the potential to help our society to become safer in future.
If you are interested in learning more about changing people’s opinions for the better without using machine learning, and good film that I watched recently is called Accidental Courtesy.
