How to Write a Book Using Artificial Intelligence (AI)
Curtis Carmichael, MS IT, CSM?, SSM?
Senior Technical Business Analyst with UI and Backend Experience, Certified Scrum Master (CSM?), Certified SAFe? 6 Scrum Master (SSM?)
Technically speaking, you are not writing a book, nor are you considered its author. However, you can take credit for being the book's editor while crediting the AI model for the content. The caveat is that at the moment, the US Copyright Office only allows humans to be considered authors (an AI model cannot receive copyright protection). So, subject to your own legal counsel's approval, one can generally put their own name as the author when seeking copyright, and perhaps include appropriate disclaimers on the book cover, in the book's credits section, and possibly also on the copyright and limitation of liability/disclaimer pages. You should, of course, be adept with the book's content, especially if it is a work of nonfiction in your field. It would be embarrassing to use a book as the basis for a course or highlight it on your resume without being able to answer questions or speak intelligently about its content.
Once published, do not hesitate to market the book as yours, with the appropriate disclosures mentioned earlier. Even before AI, course instructors, teachers, and professors would regularly use prefabricated course content for their courses, provided the content disclosed the author. AI-generated content is no different.
Now, let's discuss the specifics of how you can create a book using AI. High-level, these are the step-by-step instructions for the ChatGPT AI model 4 (create an account at https://chat.openai.com/ -- you will need a paid account to access model 4, which is well worth the $20/month cost).
With each step, you will need to get used to 'talking' to your computer via prompts. If you do not like the output, feel free to provide constructive feedback to the AI model using 'thumbs down' and specify what you were looking for, even if it means repeating certain points. Even the latest ChatGPT AI model (4) is not perfect, but it is far better than other models at the time of this article. Be sure to select model 4 as the model from the drop-down/select menu when you create a new chat. Note that the latest model currently has a time cap for content production (25 messages every 3 hours), and each output is limited to 2048 characters. Ask ChatGPT to indicate when they are done with the output by typing <endcontent>, for example.
For step 3, you will need to provide specific instructions if AI does not generate what you are looking for. For a scholarly book, you will likely want references. You can ask for things such as "use AMA style and use in-text citations with page numbers" and "include references with dates within 5 years of today's date." You will likely need to refine the output with other instructions. Once you have done that, you can use that for every chapter and prompt AI with constructive feedback when necessary. Model 4 is very good at retaining context but not necessarily following instructions without occasional reminders (e.g., in-text citations). It might be worth asking AI to use only the latest publication, as you would want to avoid listing the same book three times with multiple publication dates in the references section of your book.
Of course, for a nonfiction book, references are typically not present. Your main focus would be on the table of contents to ensure that the plot and any subplots, environmental descriptions, and character development are as you expect before expanding on each point. Sometimes model 4 will attempt to jump ahead on other chapters. Do not hesitate to click on the stop generation option and provide constructive feedback at that point.
Unfortunately, with the UI-based AI prompts for model 4, your quota of 25 messages every 3 hours will quickly run out. Retaining the current chat will help you with subsequent continuation of the book production rather than creating another ChatGPT account (by the way, this requires a non-VoIP phone number, meaning it typically needs to be tied to a physical cell phone). A potential workaround would be to request API access and then keep a close eye on the output for cost reasons. The latter requires some technical expertise but is not out of reach with the amount of open-source documentation and examples to assist publishers.
For the book cover graphic, I highly suggest using DALL-E 2 (https://openai.com/product/dall-e-2 ). There are reasonable image generation credit bundles offered. DALL-E 2 has an edge over Midjourney, as it is cost-friendly and your images are not broadcasted within public Discord channels. When typing prompts for DALL-E 2, I suggest using abstract images for your book cover, as those tend to look the best at the moment. Avoid trying to generate photorealistic images of people, as DALL-E 2 does not produce realistic images of facial features at the moment. Also, avoid using text in images, as the text output does not look great. Perhaps this will be improved with a later iteration of DALL-E. DALL-E 2 is recommended over stock images from a cost perspective, as you will not need to pay extra for an expanded license, which would be required for a digital book download or a mass-published physical book.
If you liked my article, please share. I do not make any monetary gain from writing articles, and I enjoy helping others learn about new technology.
I am a Senior Technical Business Analyst with a large multi-national software company. I have over 10 years of experience as a Business Analyst, along with Scrum Master experience.
Please feel free to contact me for any questions or to explore new opportunities in AI or business analysis.
Curtis Carmichael, [email protected], 603-769-8012