Most of us take months or years to write a book. Entrepreneur Angelique Schouten has written several books by hand, and has now embraced the power of ChatGPT in her latest work Rising AI Tech Demystified. The whole project took her just under a week to finish.
How did you come up with the idea to write a book using ChatGPT?
All my life I have been fascinated by technology. That started very early with a game console or Meccano toys and later it became an interest in cloud and software. Last year I received an invitation to join the AI Innovation Circle from Endeit Capital and Ai.nl, and we went to Seattle and the Bay Area. This was a great experience with over nineteen companies and forty speakers who showered me with information and new trends and applications.
Previously I was the author of the book Money Money Mind and after we visited OpenAI I was inspired to write a new book, but as an experiment together with the co-author ChatGPT. From that came Rising AI Tech Demystified. The co-author means that I wanted at least 50 percent of the content written by the tool, such as the words and structure. In addition, I wanted to publish it within a week. So those were kind of my requirements to write the book.
Where do you start when you start working with such an algorithm?
I was working with one of the early versions of ChatGPT and it had some limitations compared to the current version, but I got a long way nonetheless. I started with a simple initial prompt: write a book index. To my surprise, the tool starts typing and a whole structure comes out with several chapters. With that I already had the entire outline of my book.
Then I tried to add more depth with naming paragraphs and setting up the framework. This is actually exactly the way an author would do it, but now the tool did it for me. That was the easiest part of writing and having it written.
I wanted to write about my two business fascinations: AI & Fintech. I then asked the tool to come up with a topic based on this information that is also related to AI and machine learning and how it comes together. That turned out to be about crypto and trading algorithms. This laid the foundation for the book.
What are you surprised about, both positively and negatively?
What goes incredibly well is writing the outline, so that you have the entire structure within a day. That normally takes months. Monkey Money Mind took us eight months. Each part of the process of writing that book took about the same amount of time. Three-quarters of the time we spent writing the structure, and the rest of the time we edited and edited the text.
With ChatGPT this is reversed, because I was writing 25 percent of the time, and editing and editing the rest of the time. That degree of speed is very good. What also goes well is actually the application of creativity. It is not my strength to write a film script or write haiku, but because that AI tool and the ChatGPT are so accessible, you become more and more creative. For example, I asked for a Star Wars script, and for a Metallica song. Then you know that your creativity will explode thanks to this tool.
What is less good is the variation in writing style. Every piece of text that is delivered has the same structure and the same words. For example, a word that came up a lot in ChatGPT’s answer was ‘overall’. If you don’t figure it out, every third paragraph will contain the conclusion ‘overall…’. That means you have to edit and rewrite as a person, but then I wouldn’t reach my goal of 50 percent written by the tool. You want to stay original, but it still feels mechanical.
The second thing that didn’t go well was the representation, so the sources. The underlying data is simply based on existing sources and there is a bias in that. For example, I asked: Give me the best books written on this subject. Then it spoons up a list of authors, but they were all men. I then had to adjust the prompt to specifically ask for a list of only female authors, but the answer was not 100% accurate. I also saw that fact and fiction were regularly mixed up, especially when the algorithm names figures such as turnover figures or stock prices.
How did you deal with this discovery?
So that’s where the reversal of that 75 and 25 percent comes back. What I did then was check all the numbers that are in the final book. My goal was also that this would be an experiment, not just a book in which everything had to be 100 percent correct in terms of content.
I also deliberately did not check a number of things, because I wanted to see it as a co-author. So in some places I also saw that it was wrong, but I did market it because I want to show that such a tool is not flawless, just like people are not flawless. It comes across as very convincing, but just as people don’t always write down the truth in books, the tool doesn’t do that either. Then again comes the question of whether it is human or mechanical. My final advice is to check all numbers, dates and facts if you are not using it as an experiment.
What were the reactions when people heard that you had written a book within a week?
The reactions were very diverse. On the one hand, people who are concerned about the underlying data and, in particular, bias within the model. In addition, there was a concern about is misinformation that I talked about earlier. The information is very convincing and then it is easy to accept that ‘truth’.
Others mainly asked for advice on how to write such a book. Many have never written a book and were interested in how I did in collaboration with the tool. It can lower the threshold for people who have never written a book or who have difficulty with it. In addition, your creativity is stimulated and you can address and manage weaknesses. It is, as it were, democratizing content and then putting your opinion and vision on paper.
What is your advice to people who want to start this too?
I would read a lot about prompts first, because you go from copywriting to promptwriting. I wouldn’t start writing all of your manuscript too early, because you’re going to get better at writing prompts. So experiment with the prompts first. You can also find many websites on the internet where you see examples of prompts. So do that first. This also improves your output.
Second point is that you have to pay attention to the psychology. If you see ChatGPT as a co-author, you’re making a compromise. If you’re writing with someone, you also have to make compromises in terms of style and content. If you’re doing it with a tool, you have to make the same kind of compromises, or else you’re writing the book all by yourself. Finally, I would combine human creativity and AI, because you are the one who gives the instructions. At one point I found the lyrics to be a bit boring, so I prompt that it needs to rewrite the lyrics in a more tantalizing way.
In other words, don’t immediately grab one output, but make sure you have different variants. I also included that in the book. My original version of the prompt, and then version one, two, and sometimes as many as three of ChatGPT’s output. That’s why I also included a Metallica song in the book, because I could never have come up with such an output myself. That combination of human creativity and AI is incredibly cool to work with. cool and very nice people to work with.
What is your image of OpenAI now that you have also visited it physically?
It is surrounded with mystique from the outside. I think we were 20 minutes late, because there was no sign outside, the entrance was hidden and we had to wade through all that security. My view of the company, also based on the presenters, is that the mission they have to really make their AI developments better and open up can be felt and tasted everywhere. I found it very special how the people who worked there and also how they talked about it.
All the talent that has been gathered around the mission and I think that many other business models, but also startups, will benefit from that. It’s not just that Microsoft has invested so much money in OpenAI. However, there is also the flip side of the story: how dominant will this company become and who will compete against them or at least ensure that there is diversity in the underlying models that are developed? But my impression is very positive, especially when you look at how groundbreaking they are not only in the model but also in the UX and accessibility of making it available to people.
Can we expect another AI book from Angelique Schouten in 2023?
There will certainly be another book, but I’m not going to fly it as an experiment anymore. In other words, I’m going to take more time and edit more with fewer compromises, because I’ll see it as a tool and not as a co-author.