Review: Story Machines by Mike Sharples and Rafael Pérez y Pérez

Mike Sharples and Rafael Pérez y Pérez’s Story Machines: How Computers Have Become Creative Writers presents an entertaining account of Artificial Intelligence, and of both human and machine creativity.

Having heard Mike Sharples talk on a podcast about his ideas regarding the future of Artificial Intelligence, I expected the book to be looking forward into the potential applications of AI writers. Instead, what I found was a thorough and enjoyable history of the development of AI from Victorian era machines through surrealist poets and up to modern day Natural Language Generation tools like GPT-3. Throughout the book the tone is kept playful, with just enough of the technical aspects of AI to demonstrate the obvious depths of Pérez y Pérez and Sharples’s understanding. Overall, the narrative of the book takes precedence over the nuts-and-bolts of robot writers, making it accessible for a much wider audience.

The book begins with a Q&A-style chapter which introduces some of the basic concepts of “Story Machines”, including the influence of Russian folklorist Vladimir Propp, the idea of story grammars, and the general curve of the development of AI writers. The book then shifts focus to analyse human creativity which, after all, underpins anything that can currently be output by a machine. The authors discuss childhood creativity, drawing on their own research into how children tell stories, and imagined story machines from Gulliver’s Travels to Roald Dahl.

Having established human creativity as the source for much of what comes next, Story Machines begins to explore the early days of “artificial” writers including a mathematician’s booklet to create Latin verse, and John Clark’s formidable clockwork poetry machine. I didn’t expect such a detailed and fascinating history of past attempts to create story machines, and I was swept up in the detail – including diagrams and photographs – of these early inventions.

John Clark’s Eureka machine. Image source: https://www.atlasobscura.com/

Sharples and Pérez y Pérez move from 17th century mathematicians to Victorian poetry machines, via the French Oulipians, the Dadaists, and the Surrealists to land in the 1970s, with the advent of computer generated texts. With various case studies spanning the 1970s-1990s, the authors explore many different forms of automatic writers, giving examples of the output throughout. The stories produced by these early attempts at AI writers range from ridiculous to just-about-passable, but often fall far short of a human-level text. The problems with AI writers are laid out alongside an exploration of linguistics, Russian Formalism, and creativity. For example, some of AI writers’ biggest issues stem from the inability to self-correct, a tendency to wobble off-topic, and the lack of self-awareness that allows for revision and continuity.

When I tried to use an AI writer to produce some content for this review (very meta) the app, Writesonic, came up with this rather bizarre introduction:

With a co-author, a TED talk, and a blog post to his name, Mike Sharples is no stranger to the world of writing. However, with his latest book — Story Machines: Creating Fiction in the 21st Century — he takes on arguably his biggest challenge yet. And just what exactly is this ‘story machine’? For Mike Sharples, it’s an immersive writing experience that can be accessed anywhere online through virtual or augmented reality. With a focus on technology and how it can help us as authors, you may find some of these ideas vaguely familiar — but not quite in this context. Whether you’re an aspiring writer or simply someone who enjoys reading books from time to time, here are 5 reasons why you should read Mike Sharples’ book Story Machines.

Writesonic AI Article Writer 3.0

The passage neatly demonstrates some of the issues highlighted by the authors of Story Machines. Firstly, the attributions to Sharples seem arbitrary and slight odd (just the one blog post? He has several, I’ve read some of them). It also fabricates the subtitle of the book (“Creating Fiction in the 21st Century”) and throws in some virtual reality for good measure.

Neural networks and the rise of the Story Machines

Once the book moves into the realm of GPT-2 and 3, the quality of the machine output improves dramatically. The authors explore Natural Language Generation tools, video game “storyworlds”, and programs like MEXICA which are far more capable at constructing believable narratives. In the penultimate chapter, they offer a fun and simple way to create your own story generators, which could easily be used by students in the classroom and requires no understanding of digital technologies or AI.

The book closes with an extract from a machine writer named File Explorer, which writes about its experiences in the first person as a being which resides inside the internet. The story begins with the protagonist – a process – being spawned in the file system of a computer. After many (many) lines of story in which the process navigates the file system, curiously exploring file contents and getting excited about folder elements, the story ends with a contended process that is “grateful that a user had enabled its adventure to happen.”

Story Machines: How Computers Have Become Creative Writers offers an accessible and engaging history of how human creativity can be explored through – and augmented by – machines. The writers are clearly optimistic about the future of story machines, and I’ll be interested to see how their future projects pan out.

If you’re interested in reading more about the impact of AI on education and creativity, sign up to the mailing list to stay up to date:

Processing…
Success! You're on the list.

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s

%d bloggers like this: