Tom Whitwell

It took Twitter four hours to shut down my modernist architecture bot. Why don’t they shut down the antisemites?

An ingenious and inoffensive online toy got the attention of Twitter's 'moderation police/ Why aren't they so hot on the real offenders?

January 13, 2017 10:08

Twitter is struggling. Struggling to recruit new users, struggling to grow advertising revenue and even struggling to find a buyer with big enough pockets to cover their losses - half a billion dollars last year. 

One reason buyers are wary is that the social network has become a haven for extremism, misogyny and – in particular - anti-semitic abuse.

The ADL reported last October that they had identified 2.6 million tweets including anti-semitic language between August 2015 and July 2016. 

In November, Twitter announced new guidelines against "conduct that targets people on the basis of race, ethnicity... religious affiliation." 

Twitter's defence has always been that policing their platform is difficult. But this week I learned just how powerful - and fast - Twitter's moderation tools can be.

In certain circumstances. 

Over a couple of evenings this week I wrote a simple Twitter script. The script, or ‘bot’ in Twitter parlance, would compose and post tweets without any further interaction from me.

It was based on page 93 of Le Corbusier’s book The Modulor. The page shows what he calls a panel exercise; take a square, and divide it into smaller shapes according to various rules, to see what looks aesthetically pleasing.

My script automates the panel exercise, creating 48 differently divided squares each time it runs.

I turned it into a bot that replies to anyone mentioning ‘Corbusier’ on Twitter (there are a few every hour) with a freshly-generated image and a plausible Corbusier quote generated by a Markov chain.

I turned it on at about 6pm on Wednesday evening.

It got a few ‘likes’ and ‘follows.’ A frair few Twitter users seemed to enjoy playing with it.

It had a brief conversation with an professor of urban planning about the Bauhaus and Indian urban planning in the 1950s.

Then at around 10pm it stopped working after sending 49 messages.

I was confused, but by reading the error message quickly learned that Twitter’s systems, had noticed it infringed this line in their Automation Rules and Best Practices: “Sending automated replies based on keyword searches is not permitted.”

With hindsight, this rule makes complete sense; a Twitter roamed by countless reply-bots would quickly become unbearable.

Twitter were responsive and helpful — I got an automated reply very quickly that made sense. I’ve changed how the bot works, so it will now only tweet when it is directly tweeted at.

I’m optimistic that Twitter will at some point rescind their ban.

Obviously my freedom to speak to random modernist architecture enthusiasts has been infringed, but that’s understandable.

So, Twitter have reasonably efficient and effective systems for dealing with straightforward infringements of their Terms of Service.

But not all infringements are equal

Here is another section from The Twitter Rules:

Hateful conduct: You may not promote violence against or directly attack or threaten other people on the basis of race, ethnicity, national origin, sexual orientation, gender, gender identity, religious affiliation, age, disability, or disease. We also do not allow accounts whose primary purpose is inciting harm towards others on the basis of these categories.

These words are hard to square with those 2.6m tweets that the ADL found, or when you look at KKK chief David Duke’s horrific account. He joined Twitter in September 2009. so they've had seven years to catch up. 

Why does this happen?

Any business has to look after customers. Twitter’s customers are advertisers; they’re they people paying for the service, so it’s important that the advertising system works for them.

Automated replies based on keyword searches subvert the advertising system.

If I was a media agency promoting a Le Corbusier exhibition, @corbusierbot might be a nice way to engage interested people. With a few lines of Python, I could bypass Twitter’s ad system and cost them revenue.

However, David Duke’s hate stream is popular, with 27k followers (don’t go and look at those accounts), so eyeballs can be sold to advertisers like… Disney.

Hopefully, one day Twitter will find a way to block hate speech as effectively as they block modernist architecture bots.

Tom Whitwell is a digital media consultant and 'reformed journalist.'

January 13, 2017 10:08

Want more from the JC?

To continue reading, we just need a few details...

Want more from
the JC?

To continue reading, we just
need a few details...

Get the best news and views from across the Jewish world Get subscriber-only offers from our partners Subscribe to get access to our e-paper and archive