Twitter’s Conservative Ban

Recently Twitter has been accused of targeting conservative users of it’s platform and censoring them. But let’s dig into what happened from a technological standpoint…

The algorithm that Twitter used to tell what is a bot vs a normal user is most likely a neuro network. Neuro networks are unique in that often times not even the programmers of the neuro network knows how it actually works. The neuro network essentially learns how to tell the difference between a bot and a human by taking tests. The longer the test the more intelligent the neuro network can become. The neuro network slowly becomes better and better at telling the difference between a normal user vs a bot.

If the programmer is a teacher, a neuro network is a student. The teacher could create a complex lesson plan that is intended to teach the student as well as possible and the student does learn reasonably well. But the teacher does not have full control over the student and how it learns. In the end the student learns for themselves and the teacher only has a vague high level of control in how the student learns. Now ask the teacher if they have an understanding of how the student’s brain thinks. This is similar to the level of understanding that a programmer eventually has of how the neuro network thinks.

Now then what went wrong at Twitter? Twitter apparently has built a neuro network that is good at telling the difference between a normal user vs a bot within an acceptable level of false positives but then failed to notice that these false positives contained a higher than normal percentage of conservative users. The most likely reason for this scenario is that the people who created the test that the neuro network was trained upon couldn’t tell the difference between bots and conservative users also. In the end what happened here was the neuro network learned to profile conservative users. And now the question is, is there someone to blame? And if there is who would it be?

So is it the responsibility of the programmer for not creating a good enough test to train the neuro network against? Is it the responsibility of the bot creators for creating bots that resemble conservative Twitter users? Is it the responsibility of the bots for emulating the conservative Twitter users? Is it the responsibility of the conservative Twitter users for having behaviors too similar to the bots? Is it the responsibility of the neuro network for not being able to tell the difference?

But for now I will sit back and enjoy watching conservatives, a group of mostly cisgender, heterosexual, white or male fret over their first instances of getting profiled.