I’m a firm believer that the only way to understand technology is to play with it. So that’s what I’ve been doing with AI for a few months now. And since ChatGPT was released in November, I’ve been experimenting with prompts to make them as powerful as possible. This book is part of the journey.

On Sunday evening, while watching TV with my wife, I rattled in the following prompt:

Generate a book title for the keyword I give you. Then generate a book outline of 15 chapters from that title. Then write a detailed book chapter for each chapter in the outline. Write in a friendly, conversational and authoratative style. The keyword is: “The harmful effects of social media on society”.

Within seconds I had the book title Unfriending Society: The Harmful Effects of Social Media and the following chapter outline:

Chapter 1: Introduction

Chapter 2: The Impact of Social Media on Mental Health

Chapter 3: Cyberbullying and Harassment on Social Media

Chapter 4: The Spread of Misinformation on Social Media

Chapter 5: Social Media and the Decline of Privacy

Chapter 6: Social Media and the Destruction of Human Interaction

Chapter 7: Social Media and the Disintegration of Community

Chapter 8: Social Media and the Destruction of Democracy

Chapter 9: Social Media and the Loss of Productivity

Chapter 10: Social Media and the Impact on Business and Commerce

Chapter 11: Social Media and the Impact on Youth

Chapter 12: Social Media and the Impact on Politics

Chapter 13: The Positive Aspects of Social Media

Chapter 14: Solutions and Alternatives

Chapter 15: Conclusion

Not bad for a start.

It then launched into writing the chapters.

It got to the end of chapter 1 and stopped. That’s fine. ChatGPT can only return a certain volume of text. All I had to do was ask it to keep going. It churned out Chapters 2,3 and 4 before pausing again. I kept nudging it onwards until it had finished the book.

All in, that took about 5 minutes.

Oooh. That’s a bit short.

The first thing I noticed was that the chapters were a bit on the short side. I mean, really quite stingy. The entire book came to around 3000 words – which is shorter than most of the chapters I write in my own books.

So it was time to go back to ChatGPT and get it to flesh out the content.

I asked it to add to the chapters, picking up where it had left off. But that didn’t work too well. It mainly ended up rewriting the content of the chapters in a bit more detail. So I got it to do that for each of the chapters in turn.

I then splooged (I’m sure that’s a word) the text together to create chapters that were about twice the length of what it had originally delivered. 

This was just an experiment so I muttered “That’ll do” and got in with the rest of the process.

Meanwhile, in human-land

While this was going on I was busy building the document for the book.

I chose a standard book template in Apple Pages and started pasting the text in place. It was dull administrative work, so I added some design froofery (I’ll need to check that’s in the dictionary) just to keep it interesting.

I now had the chapters done. But books need more than that. So I asked ChatGPT to write a preface, a biography, the blurb for the back cover and other doo-dads.

I went to Dall-E and asked it to create an image of a broken smartphone on concrete. It created four of them and I picked the best one. I pasted that in as a front cover.

Now I asked Dall-E for an image of someone at a typewriter with a glass of whisky beside them. And – Boom! – I had the author’s photograph.

I thought I should probably add something about me in there as the twit who had prodded the technology to spew out this stuff, so I wrote a bio page for myself.

And that was it.

Nah. I don’t like that.

Now that I had the whole book together, I released I didn’t really like the title.

That’s easy enough to fix:

“Please come up with 10 alternative titles for the book”

(Note that I’m always polite to the AI. When they take over, I want them to know I respected them from the start.)

I picked one of the alternatives and then thought about the content. I’d quite like it if the book was useful, so I asked ChatGPT to come up with a five-point manifesto on how to use social media in a positive way. It happily obliged and I tacked that on at the end.

I exported it as a PDF. That’s what you can download near the top of the page. It hasn’t been proofread. But it AIs don’t make spelling and grammar mistakes, so it’s way ahead of me on that front. 

I don’t have time to read this crap

For someone who does a heck of a lot of writing, you may be surprised to learn that I’m probably dyslexic. I find reading a struggle and read inside my head at the same pace I read out loud.

So it would be so much better if I could get an audiobook of this.

No problem.

I opened up Descript where I’ve already trained the system on my own voice. I had uploaded half an hour of an audiobook I’ve read and about 24 hours later, it had generated an automated voice that sounds like a robotic me.

All I had to do was paste the text of my new book into Descript and give it a few minutes to process.

After about 10 minutes I was able to download a 50-minute audio file of the book. 

You can listen to it, near the top of the page, if you fancy it.

So, what do I think?

I want to give you my thoughts as an author. Writing books the old-fashioned way takes time. I wrote the first edition of ‘A User Guide To The Creative Mind’ in 10 days – but that was mainly compiling stuff I’d already written. ‘How To Get To Great Ideas’ took me about 4 months. And ‘Iconic Advantage’ and ‘Friction’ which I co-wrote with Soon Yu each took over a year.

So what do I think of a 4-hour book.

Well, I think it’s amazing and pretty awful all at once. Here’s a breakdown:

So much, so fast

It’s extraordinary how much ChatGPT achieved in such a short time. There is some fantastic content in here that it produced in a matter of minutes. Some of this is worthy of being in a published book. And the outline of the book was really impressive. It came up with suggestions I would probably only come up with after weeks of research and discussions. 

Say that again?

However, while listening to the audiobook, I realised there was a lot of repetition. It mentioned the mental health issues associated with social media use at least half a dozen times. It quickly became apparent that if you were to remove the duplication, there was a lot less there than it looked.

It wasn’t that compelling

The writing didn’t include any personal anecdotes (because they’d have been fictionalised bullshit), stories, interviews or that many stats. Which made the dialogue a bit one-note. But you can’t expect it to do most of that stuff. It’s an AI not a human with a bucketload of experiences.

Echo…echo…echo…

When I write books, I’m always looking to add something new. I purposely try to be contentious in places and question whatever the consensus happens to be. Preaching what everyone knows isn’t very interesting to me, so I spend lots of time coming up with my own thoughts and frameworks. AI doesn’t do that naturally. It reports on what’s in its dataset. And it does that incredibly well.

Does that mean AI is shit?

No way. AI is phenomenal at certain tasks. And looking at it through an anthropocentric filter is an ignorant thing to do (and a real flaw of humanity, in general). AI does not do human creativity and we shouldn’t expect it to. But it does many of the tasks that go into human creativity infinitely better than humans can. It draws on a near-limitless bank of knowledge. It can look at things from lots of different perspectives. It can come up with logical arguments and articulate them beautifully. It can spell perfectly and use impeccable grammar. And it does it all faster than an Olympic typist. 

Will I use AI in my own writing

Absolutely. But I’ll be using it for specific tasks. And I’ll be augmenting it with humanity, quirkiness, stupidity and irrationality. I can see how AI can help me explore more broadly, prototype faster, simplify complexity, focus on the more valuable part of writing and generally up my game.

Those who are simply nay-saying and writing the technology off as a load of tosh are the ones who are putting themselves at an advantage.

There will always be room for classic booksmiths. Writers who insist on hand-crafting every element and spending years leafing through the dusty sections of the library, will still have a place. I’m just not convinced their organic, artisanal, wordsmithery will be of much higher quality than the machine-assisted type. I don’t see the value in perfectly-cut, hand-stitched underpants when Calvin Klein machine-assisted underpants do the job so well.

This experiment may not have produced something of world-shaking quality. But I wasn’t expecting it to. This is all part of my journey to discover how to work with the technology in the best way, rather than to make a binary decision of leaping fully onboard or turning my back on the technology like a literary Luddite.

We’re only getting started. There’s lots more to come.

[NOTE: none of this was written by an AI. You may wish it was. It would have made the article a hell of a lot shorter and got to the point without you having to scroll. So, if you’ve read this far, hats off to you!]