Here's Why I Don't Want My Book About the Metaverse Fed Into an LLM

Meta pirated millions of books to train its AI, the Atlantic recently reported, including mine. There's even a helpful look-up tool here to find if your favorite author is in Meta's training set, which is where I got the screenshot above.

I'm not sure how I'll protest, but it does bring up a related topic I've been thinking about lately:

Does it even make sense to feed my latest book into an LLM?

I’ve had friends / colleagues ask me something like that in recent years: “How about we upload your book into an LLM, so everyone can easily ask anything about it?”

For instance, as the head of an AI company put it to me, explaining how this can be done:

"I worked on a project where we hook near a petabyte of data to a LLM that just looped through doing research on authors of Gutenberg books. Then we turned the books into interactive books read by the authors own voice/style.” 

My immediate answer to the idea is on the order of: “Hmm, no that’s OK.” 

My more thoughtful, considerate answer is: “Fuck no I mean holy shit that’s not even how my book works. Jesus.”

In case you’re curious -- and the question keeps coming up, so I guess some people are  -- let me explain with more details and less cursing, specifically around Making a Metaverse That Matters:  

My book is a step-by-step argument and a set of stories enforcing that argument.

To take the structure of my specific book, Making a Metaverse That Matters, it’s why the very first page defines the Metaverse, and is followed by a lengthy introduction which tells the powerful stories of five people I’ve profiled over the years, whose lives have been transformed due to the unique aspects of the technology, as I define it. 

I put the definition up front to immediately tell potential readers that my book is not about a vague, overhyped concept, but an actual platform category that already exists in early form. While each one of those stories illustrates a key facet of my definition.

For instance, here’s the story I tell about Fran Serenade in the introduction:

Fran noticed an odd thing: she seemed to be gaining significant recovery of physical movement -- apparently, as a direct consequence of her activity in Second Life.

“As I watched [my avatar],” she told me back in 2013, “I could actually feel the movements within my body as if I were actually doing tai chi in my physical life (which is not possible for me).” 

Using a virtual world, in other words, seemed to abate her Parkinson’s symptoms.

… While the implications of this have yet to be studied to their furthest potential, they are likely to be profound -- especially in the face of a rapidly aging population around the globe. From what we can tell, they are made possible because they happened in an immersive virtual world. 

The story illustrates why I say it’s essential that the Metaverse be immersive and a virtual world. And hopefully the reader is inspired by how it's already helped people like Fran! 

All of that intentionality would likely be lost, if you just asked an LLM how my book defines the Metaverse. It would definitely be less inspiring.

Which takes me to another point:

My book is intended to be both inspiring and informative -- but only best succeeds at both if it’s read from beginning to end.

My personal favorite section of Making is the Epilogue, where I wrap up the stories of the people I write about throughout the book. I find these personal stories moving and inspiring, and hope the reader does too. But all of that is almost certainly missed, and definitely diluted, if someone was just to get them summarized by an LLM.

Matters only fully matters, in other words, if you’ve read it from beginning to end.

None of this is meant to suggest no book should be fed into an LLM/turned into a chatbot. That’s up to each author, for one thing; for another, not every book is intended to have a strong thesis or storyline, especially when it comes to technical / reference books. (My much more technical second book, Game Design Secrets, would probably work in an LLM, though it's pretty out of date.)

I also see quite a lot of potential in using LLMs to turn genre fiction books (especially fantasy and sci-fi) into explorable worlds -- in that case, the book becomes like a kind of game design document. (In fact, I’m involved in a project along those lines, but more on that later.)

So by all means, let’s experiment with books and LLMs integrated together -- as long as each author gets to decide that for themselves. Trouble is, companies like Meta are actively trying to take that choice out of our hands.

Comments

[sirhc desantis]

I think swearing is a good idea.
Keep doing it