Amazon’s virtual shelves already feature books written by artificial intelligence. One startup believes that even titles written by humans would benefit from some AI, in the form of an accompanying chatbot primed to talk about a book’s contents.
YouAI, a startup that offers tools for building AI apps, recently developed an app it calls Book AI, which promises to “turn any book into an AI.” It spins up a chatbot that knows everything about the book and can talk about it endlessly—like a bespectacled Terminator that just dropped in on your book club meeting.
Dmitry Shapiro, YouAI’s CEO, says he’s talking with a number of publishers large and small about creating chatbots to accompany new releases. Solution Tree, which offers thousands of books on continuing education, already plans to offer what Shapiro calls “conversational companions” to go with its titles.
A chatbot edition can be especially useful for textbooks because users may have specific questions or need things clarified, Shapiro says. And because the large language model behind the chatbot has, like ChatGPT and others, been trained on a wide range of other content, sometimes it can even put what is described in a book into action. Shapiro gives the example of asking the chatbot for a book on website optimization to suggest a design putting the key points into practice.
YouAI creates its book bots using a method known as retrieval augmented generation, or RAG, that makes chatbots hew to specific source material. The same technique is used by search engines to keep search chatbots focused on real information and to prevent them from “hallucinating” their results.
I tried uploading Mary Shelley’s Frankenstein to YouAI’s app to see how it works. It seemed a fitting test given current sentiments and concerns about AI, and with Halloween just a week away.
The Franken-bot did a decent job summarizing the book and its themes. It also saw some interesting parallels between the classic tale of scientific hubris and todays’ worries over an AI apocalypse—something it could riff on likely because its training data also included more recent content that mentions the book.
“Frankenstein is warned against playing God and creating life, which leads to disastrous consequences. Similarly, some argue humans should not try to ‘play God’ by creating sentient AI,” the chatbot said. “Frankenstein quickly loses control over his creature, which becomes violent and destructive. There are concerns AI could become uncontrollable and cause harm if not properly constrained.”
Not bad, although it might also have noted that just as Frankenstein served as a rather dramatic, ahem, lightning rod for angst about scientific progress back in the early 1800s, today’s headlines can often hue toward runaway nightmare scenarios to the exclusion of actual day-to-day risks posed by AI.
When I asked the chatbot if it identifies with the monster, it responded with a familiar chatbot excuse for dodging tricky questions: “As an artificial intelligence, I don’t have personal feelings or identities, so I don’t have the capacity to identify with any characters.”
Then, in a twist that Mary Shelley might have enjoyed, the chatbot suggested that perhaps I might identify with the monster instead: “Many readers might identify with the monster’s feelings of loneliness, rejection, and desire for companionship. These are universal human experiences that Shelley explores in depth through the character of the monster.” Touché.
Since I’ve written plenty about AI myself, from impressive robot breakthroughs to predictions of AI doom, I created a chatbot trained on some of my past and asked it what I seem to think of AI.
“It seems that Will Knight’s reporting indicates some unease and caution about the rapid progress being made in AI capabilities,” it replied. “Knight highlights the difficulty in predicting detailed AI scenarios and says lagging progress on safety should raise some alarms, but he does not jump to conclusions that doom is inevitable.”
I’m glad it didn’t say I think doom is inevitable—of course it isn’t.
I can certainly imagine publishers seeing chatbots as a gimmick that might help shift new books, particular textbooks, self-help titles, or the many business books that promise to teach you how to turn AI into profit.
Chatbots could also be used to put a new user interface on other sources of knowledge. Other startups use similar techniques to let you turn any webpage or website into a chatbot. For instance, Cohere, one of a handful of companies with the resources and AI chops needed to build its own large language models, recently introduced Coral, an experimental chatbot that will talk about books or documents—as well as any website you point it at.
I tried asking Coral what WIRED.com thinks of AI.
“Overall, WIRED seems to have a cautious and skeptical view of AI and its potential,” it told me. “In some articles, the potential benefits of AI are acknowledged, but there are also frequent references to flaws, over-hyping, and a lack of transparency and accountability surrounding AI.”
Not a bad summary at all, although to be true to that I should point out that a chatbot may not always be entirely accurate, or very original, about what it has ingested. There’s still no substitute for reading a book—including Shelley’s masterpiece—cover to cover.
