A Year of Reading: What If? (book vs machine) - reviewed by Katherine Roberts

My reading has taken a slightly different path this autumn, thanks to the rapid development of Artificial Intelligence. Last month, Machines That Think (part of the New Scientist Expert series) rather blew my mind, despite being published 10 years ago. Since then, AIs have become more sophisticated and even more data hungry, resulting in a class action by authors against Anthropic in the US claiming their copyrights were violated. Similar questions are being asked by the music business, as highlighted in a recent Radio 3 programme The Artificial Composer. AI is also active in other creative industries, such as the art world and video games. It seems creative professionals the world over are in danger of being reduced to hobbyists, helping to train the very AI tools that will enable everyone else to become an instant author or composer, visual artist or video game creator.

How did this happen without us noticing? Let me take you back to fiction's favourite prompt: What If? and a book of the same name.

What If? by Randall Munroe

Subtitled 'Serious Scientific Answers to Absurd Hypothetical Questions', this book published a decade ago is both entertaining and rather weird. Some of the questions are the sort of thing your six-year-old might innocently ask, such as "If you call a random phone number, what are the chances the person who answered just sneezed?" That particular question is answered mathematically using probability calculations, including some amusing variations such as the chance of simultaneously calling someone else who is also calling random numbers looking for someone who has just sneezed... with me so far? Others are quite serious, such as the dangers of self-fertilisation in humans (yes it's possible, apparently), and if/when the bandwidth of the Internet will exceed that of FedEx, since it is (or was in 2014, when this book was published) faster to transfer large amounts of data by courier transported on a hard drive rather than sent online - although I'm not sure if that still applies now we have 5G, anyone know?

Luckily, the author Randall Munroe has a wicked sense of humour. If science and mathematics give up on the question with their hands in the air, then the reader is entertained with amusing illustrations and sly references to contemporary memes. This means you get a satisfying conclusion of sorts to nearly every question asked in the book, although a few of them, such as: "In Thor, the main character spins his hammer so fast that he creates a tornado... would this be possible in real life?" are simply given one-word answers (that one's a "no", in case you're wondering). I am still trying to get my head around some of the more absurd questions in this book, let alone Munroe's attempts at rigorous scientific answers. Although it's tempting to think an AI might have been used to generate them, these particular questions could only have been dreamt up by humans and were reportedly submitted by readers through the author's website. That's how things were done in the old days.

Which brings me to my promise in last month's post to look at creative AIs in more detail, starting with the What If Machine developed in 2014 by Teresa Llano of Goldsmiths University in London. This early AI was designed to produce Disney-like and Kafkaesque story ideas and did exactly what it says on the tin. If you're stuck for an idea for your next book, here are a couple generated by the What If Machine: "What if there was a little house that lost its door?" (Disney-like) "What if there was an old dog who couldn't run any more and rode a horse instead?" (Kafkaesque). I actually quite like that one for the basis of a children's story, although considering Machines That Think was published 10 years ago, those ideas have probably been done to death by now!

Story-telling AIs actually date back to 1973, when a specialist AI called Novel Writer created murder stories in a limited setting. This was followed in 1977 by Tale-Spin, which generated woodland stories in the style of Aesop's fables. 1981 brought Author, designed to model an author's mind (good luck with that, AI!) and generate plots such as those with happy endings. In 1983, Universes produced soap opera episodes with lots of characters and overlapping story lines without endings, happy or otherwise. But it wasn't until 2013 that an AI called Scheherazade could generate more wide-ranging stories about anything it could learn about on the internet... the age of data-driven AIs had arrived.

1973 also saw the first AIs designed to produce artwork. In the same year, an AI called Aaron created pictures that were good enough to be exhibited at London's Tate Modern gallery, while in 2004 a robot called Aikon could draw portraits in artist Patrick Tressant's style by mimicking his wrist movements and pen pressure. Just two years later, The Painting Fool was drawing portraits in different styles depending on its mood, and by 2010 another AI called Angelina was creating its/her? own video games.

Music did not escape. In 1987, composer David Cope developed Experiments in Musical Intelligence (EMI, or 'Emmy') to create music in the style of Beethoven, Chopin and Vivaldi. Emmy became sophisticated enough by 2010 to release its own debut album From Darkness, Light. Also in 2010, a Spanish AI called Ianus began to compose music without human guidance, which was played at Alan Turing's centenary concert in 2012. Four years later, Ianus released a pop album called 0music.

In the decade since the publication of Machines That Think, it's becoming clear that AIs not only think but also learn from us, and they learn frighteningly fast. OpenAI's ChatGPT is an example of a large language model that can write a chapter of your novel in less than a minute given a suitable prompt, although it will (probably) need some human assistance to make it publishable. Other large language AIs include Google's Gemini, Anthropic's Claude, and Microsoft's Copilot. You can design pretty much anything over at Canva (I've used Canva for cover design), which now includes an AI that will create artwork from a text prompt. You can do the same kind of thing with Adobe Firefly and OpenAI's DALL-E, among others. If you fancy composing some music, DeepAI has a tool that will get you there, while ElevenLabs can synthesise speech, and MetaGPT might help video game developers with coding.

Now we come to a sticky question: if a creative professional uses AI tools to help them develop their work, does that make the result somehow less desirable? The AI-created music featured in The Artificial Composer sounded pretty good to me, and I'd never have known it was not 100% human-composed. Some publishers are apparently writing into contracts that the author must not use AI in the creation of their books, and yet that is going to become increasingly difficult to guarantee when AIs are embedded in software such as Microsoft's Word. For first drafts, I definitely prefer to write creatively without AI assistance, because - let's be honest - that's the fun part! I also like to paint my own artwork for my book covers, before uploading it to Canva for the design part. However, all the AIs mentioned have free versions that could be useful during later drafts of a story, maybe for trying out various scenes from different characters' viewpoints, or in a different tense? You can do that sort of thing by hand, of course, but AI is much quicker and if it doesn't work then you can simply delete. Lately, I've heard some publishers are now exploring AI-generated artwork for book covers so possibly image generating AIs are improving enough to be considered professional... though I still can't forget the white horse with three front legs supplied for me a few months ago by a popular AI that shall be nameless :-)

AI-generated image of a hare leaping past a white horse.

*

Katherine Roberts is a human author who writes fantasy and historical fiction for young readers. 

Her most popular book for readers of all ages is I am the Great Horse. Originally published by Chicken House UK/Scholastic US, this is the story of Alexander the Great as you've never heard it before - straight from the horse's mouth.

I am the Great Horse
(correct number of legs)

In the spirit of using AI tools rather than letting AI use us, the first three chapters/hoofprints of this novel are available for free download in epub format, and - new for 2025 - as a budget audiobook narrated by an AI-generated voice. We chose a British accent for Bucephalas and would love to know what you think!

free ebook
NEW for 2025: Audible audiobook

You can find out more about Katherine's books on her website www.katherineroberts.co.uk