Introducing Chris Rodley / The Art of Bots

Tue 12 Apr 2016

by Sarah Blaszczok

Chris Rodley is one of the creators of Magic Realism Bot, a story-telling Twitter bot, featured in The Art of Bots. We interviewed him to find out more about his work and how real-time data is being used in digital writing…

Tell us about your practice – do you see yourself as an artist, programmer or writer?

All three, in a way, but it’s my sister and co-creator Ali who wrote the code for Magic Realism Bot. I started out wanting to be a conventional writer, and toiled at this for quite a few years before moving towards work at the intersection of writing and digital technology. This usually means working with code to some extent (although there are some great prefab platforms now like ‘Twine’ which the non-programmatic are doing fantastic things with). As well as literature and code, digital writing usually also means engaging with an extra-literary dimension – for example, graphic design, images, animation, interactivity, live art, projections, installations. That’s where the art comes in. I think we are actually seeing a really interesting convergence between writing, programming and visual art. I’m actually not sure that the words exist yet for this type of practice, which is why myself and other bot artists sometimes use vague self-descriptions like “I make stuff on the Internet”.

Magical Realism Bot is one of our favourite story telling bots and has a popular following Twitter. Can you tell us more about the bot? How does it generate content?

The bot generates stories using the same basic principle as the games Mad Libs or Cards Against Humanity. First, it randomly chooses one of well over a hundred templates we have created. These templates provide the basic syntax of the story, but are not populated yet with key words.

A template could be as simple and general as SOMEONE or SOMETHING is DOING SOMETHING to SOMEONE or SOMETHING ELSE; or it could be much more complex, with lots and lots of “moving parts” that vary enormously. Once the template is selected, it is populated with words from our pretty large vocabulary list, which is categorised and tagged into particular types of people, places, things and actions: for example, rulers (like queens or presidents), animals, gemstones, books, magical things, things that are long, things to do with space, transitive verbs like kissing and killing, and intransitive verbs like singing and sleeping. There’s also clean-up code that fixes things like grammar, pluralisation, choosing between a or an, and so on.

One principle we try to follow is to get as much variation as possible with each template; ideally, stories from the same template should look mostly or totally different. One way we do this is by making the vocab as flexible as possible, and employing as many categories of things as we can at all times. Probably our two most fruitful and frequently used categories are simply “concrete things” (like clocks or swans) and “abstract things” (like love or capitalism). The variability means that even we are surprised by the results a lot of the time, despite knowing everything that’s in the database.

In terms of authorship – do you feel you are the creator of the Magic Realism Bot stories, or is the bot the author?

It’s both, really. Many people seem to ascribe full authorship to the Twitter bot, which I don’t mind, but I find a bit unusual, because I know how long it took to get it tweeting coherently!

Actually, I think, it’s not unusual now in digital writing to see authorship split between the primary creator, automated processes or bots, and increasingly other users on the Internet.

Think of how many of our everyday interactions now rely on predictive text; Google’s Inbox is often strikingly good at writing my email responses for me, though I usually edit them first. Or look at how much of the stuff we post online is based on content that was originally made by other people on the Internet, piped to us by Facebook or Google’s opaque algorithms, which we then build upon.

I think we need to get used to this posthuman, as it were, notion of authorship—both in the literary sense and in terms of quotidian communication.

The Magical Realism Bot currently exists as a Twitter bot. You are developing a version specifically for The Art of Bots, can you explain more about what the difference will be and how this relates to your art practice?

The iteration of Magic Realism Bot that will be shown at The Art of Bots will be different in a couple of important ways. Most importantly, you’ll see the miniature stories being constructed piece by piece—the code will propose and discard potential vocabulary choices before settling on the final outcome. This will help to illuminate its machinic nature—sometimes on Twitter we are accused of writing the tweets ourselves or at least editing them, which of course we don’t (apart from anything else, I’m way too lazy to write 7000+ tweets).

I guess the other difference is that you’ll be seeing the stories en masse, rather than one by one, which is how people encounter them on Twitter, where they simply pop into people’s feeds every two hours. So you’ll hopefully get a sense of the bot’s innate character beyond any individual piece of text.

The final difference is visual: we’re presenting them in a way that’s more sympathetic to the spirit of magic realism, as a kind of looking-glass of the kind a Jorge Luis Borges character might own. One limitation of Twitter is of course that you have more or less no control over how text is displayed.

You often refer to the writings of Borges in your work. What is it about his style or the content of his stories that you are most interested in or inspired by?

Jorge Luis Borges – the great Argentinian magic realist author – absolutely inspired the work. I have long been a fan of his writing, particularly his ideas as opposed to the execution of them—although that’s also brilliant—in fact I remember I used to write down scores of ideas for stories in a Borgesian style. By that I mean stories that stretch concepts or structures to their logical endpoint; that embrace infinity. For example, a library with every conceivable book in it, a map that is as detailed as the territory it represents, an author who copies another novel verbatim, a crystal ball in which you can see everything simultaneously.

It occurred to me that you could possibly automate this process, applying the same transformations to other things like zoos and shopping malls and ATM machines and pizzas. So that’s what we tried to do, and in fact, the project was initially called BorgesBot. After a while, we realised that the bot had its own particular tales that it composed really well—that it wanted to write, maybe?—which were not really like Borges but were interesting anyway. Stuff that was more visual and romantic—say, the Moon falling in love with a golden owl—whereas Borges was quite dry, austere and academic.

Also, we found it hard to recreate Borges’ metatextual style: he would often have stories nested within historical accounts to provide verisimilitude, which is hard to do in a 140-character tweet. To be honest, we’ve now even widened the bot beyond strict magic realism; it occasionally encompasses fantasy, sci-fi, horror, fairy tales, mythology, and so on.

What do these auto-generated stories say about the role of the ‘imagination’ (historically, the author’s) in storytelling?

One thing I like about Magic Realism Bot is that it shows—well, sometimes, anyway—that machines can actually produce content that not only merely mimics human creativity in a somewhat clunky way, but actually writes new types of ideas that humans wouldn’t necessarily have thought of. I notice this sometimes happens when the bot writes stories that include abstract concepts placed into a context where you’d usually expect concrete things. It also happens in stories which are about something being constituted by something else—there was one recently about lightning bolts made of cockroaches which wigged a lot of people out.

I’m really interested in this distinctive robot imaginary and how it will develop. Currently, I think, many botmakers are focused more on the task of training bots to be as human as possible, rather than making them ‘machiney’. Because humans are already pretty good writers, it’s a tough task to make a bot that’s as good or better than one. But what we can focus on is making a bot that’s good at being a bot.

Can you tell us more about how Magic Realism Bot relates to the field of data-driven literature – an emerging genre that you have previously written about? 

Data-driven literature was a term we proposed for a diverse range of writing projects that myself and another collaborator Andrew Burrell, a new media artist, had been observing for a couple of years. We used it specifically to refer to electronic literature and text art projects which make use of the emerging sources of big (or at least moderately big) data—most notably Twitter, but also Google, Wikipedia, Tumblr, Flickr, the Project Gutenberg database, the New York Times archive, and so on.

Many popular Twitter bots do draw on these large repositories of data, including Microsoft’s Tay, which utilises the Twitter archive, and Darius Kazemi’s Two Headlines, which remixes news headlines from Google. As the disastrous debut of Tay showed, working with large datasets can be tricky—live data is messy, unpredictable, and often actually just pretty nasty. Our work Magic Realism Bot uses its own corpus and so doesn’t fit into this category of data-driven writing; while we could have drawn upon large online word lists or Wikipedia, we wanted to be able to tightly control the vocabulary that was being outputted (we want grandfather clock, say, but not a Swatch watch on our list of machines). In our next project, though, we aim to take the basic technique we used in Magic Realism Bot and expand it with real-time data from Twitter and the wider web. Watch this space!


A physical manifestation of Magic Realism Bot, by Chris Rodley and Ali Rodley, is on display at The Art of Bots, 15 & 16 April at Somerset House, London.