Reference News Network, April 24 — The Times (UK) website published an article on April 22 titled “AI Tutor Bots Will Spark a Textbook Revolution,” by Danny Fortson. Excerpts from the article follow:
Xu Daohui has been given the cold shoulder for several weeks. Previously, the Michigan State University professor and founder of AI startup SuperFocus had arranged a meeting with the CEO of one of the world’s leading textbook publishers.
SuperFocus developed a chatbot trained on one of the publisher’s best-selling textbooks. After being “fed” nearly every page of the book, the bot model aced exams, raising the possibility of a new generation of AI “study companions”—specialized tutors expert in a single subject.
“We said, ‘We’re thinking of launching these study companions for just the 10 most common high school or college subjects,’” Xu recalled. “Then they cut off communication with us.”
Amid the flood of new “generative AI” products, SuperFocus’s effort represents yet another striking potential application of this technology—but it also places the startup at the heart of several key battlegrounds in the AI field: What information can be used to train these models? Should AI developers pay royalties to publishers for access to crucial data?
Take Pearson, one of the world’s top textbook publishers, as an example. When asked how the company views third parties using Pearson’s materials to create “AI teachers,” CEO Andy Bird was unequivocal: “I suspect that would cause us problems.”
Bird said Pearson plans to “embrace” the AI education revolution but added, “That doesn’t mean everything is free. You have to start with… respect for intellectual property and copyright.”
All “generative AI” tools work the same way: they are trained on massive amounts of data. Then, when prompted, their core algorithms generate the most plausible “answer” based on all the training data. The mechanism behind these answers is essentially highly refined guesswork—predicting the most likely next word in a sentence based on billions of examples.
If you ask SuperFocus’s “study companion” who the 23rd U.S. president was and what his signature policies were, the answer won’t simply repeat content from the textbook it was trained on. Instead, it will produce a new response grounded in the data it absorbed—in other words, the kind of service a human tutor might provide.
Clearly, publishers don’t share AI advocates’ optimism. The central issue revolves around a legal doctrine called “fair use,” which allows individuals or companies to use copyrighted material without permission under certain circumstances. The boundaries of fair use are blurry. Judges consider factors such as whether the use significantly devalues the original work and how much of the original material was used to create something new.
Given the pace of technological advancement, this is a critical question. In the future, everyone may have a smart, always-available “pocket genius” on their phone capable of answering almost any question, writing essays, or creating art.
Education is among the first sectors to feel the disruption. But there’s a limiting factor: large language models like ChatGPT (Chat Generative Pre-trained Transformer) often “hallucinate”—they don’t actually possess knowledge but instead guess answers to questions. As a result, they frequently fabricate facts and state them with unwavering confidence.
SuperFocus believes it can overcome this problem by giving its AI tool a “memory” limited only to specific, desired materials—such as a company’s customer service manual or a U.S. history textbook—and restricting its answers to that knowledge base.
Whether this approach will hold up remains an open question, but one thing is clear: education is changing rapidly, and practitioners are scrambling to adapt.
“Sticking your head in the sand won’t work,” Bird said, “because this is reality. It’s happening—and I think faster than anyone expected.”
