The next-generation college textbook will undoubtedly be an "intelligent e-textbook." But different researchers have different visions of what the textbook of the future ought to be and which CS technology path will produce the most useful–and scalable–product.
For instance, at Seattle-based Vulcan Inc.--the company started by Microsoft cofounder Paul Allen--researchers are two years into a project known as Inquire. When completed, the iPad app will enable students to e-read the popular Campbell Biology college textbook while it answers their questions in real time about the book’s content using the knowledge in the book.
Meanwhile, Connexions, one of the first initiatives to offer free textbooks via the Web, has launched an OpenStax Tutor program that uses machine-learning algorithms to analyze student responses to questions to create an intelligent "computer-made" textbook. In contrast, Inquire is an intelligent man-made textbook.
The process of creating Vulcan’s Inquire app began in 2010 when the research team started encoding the knowledge for the entire Campbell Biology into its knowledge base.
The team assigned the encoding task to what it refers to as a "knowledge factory"–a group of four biologists in India that has encoded 40% of the text. Vulcan plans to scale this up to 18 biologists to more quickly finish the book. The target completion date is year-end 2013.
In the meantime, the researchers beta-tested the app using three groups of 25 college students. One group used the hardcover textbook, a second used Inquire with its questioning-answering portion turned off, and the third used the full app. The students scored 71%, 74%, and 81%, respectively, on a homework test, "which is meaningfully significant; that’s a letter grade difference," says David Gunning, Vulcan’s senior research program manager and the manager of the Inquire project.
"Inquire not only helps students understand what’s in the text, but it does it in real-time," adds Mark Greaves, Vulcan’s director of knowledge systems. "There’s no longer a need for the student to, say, e-mail a professor and sit back and wait for an answer."
Indeed, students can highlight difficult text in the app, which then generates a list of question choices in the right-hand margin. Select one and a page of information specific to the text in question appears. If the student is still confused, she or he can type in their own questions and the AI software will attempt to provide answers.
But this sort of curated knowledge, which relies heavily on human labor, can be both expensive and time-consuming, says Richard Baraniuk, a professor of electrical and computer engineering at Rice University. His OpenStax Tutor is using an alternate technology path that utilizes no human labor. It is being funded by both the U.S. National Science Foundation and Google.
"While Inquire’s goal is right on target," Baraniuk says, "the question is just how scalable is their approach. Let’s say they are successful and want to do 50 more books. That will require an enormous investment in time and money."
Instead of using what Baraniuk calls " ‘knowledge engineering’–or hiring a bunch of domain experts, in this case 18 biologists, and locking them in a room for several years and feeding lots of money under the door"--he is using "machine learning, Google-style."
"Google doesn’t have humans deciding what Web pages to recommend. Instead, it uses machine-learning algorithms," he says, "which is what we are doing for personalized learning."
OpenStax Tutor monitors students as they progress through one of the free, open-source e-textbooks in the Connexions library. A suite of sophisticated machine-learning algorithms analyzes student responses to practice problems embedded into the e-text and to homework problems assigned by the instructor.
"It then automatically assembles a concept map similar to that created by Inquire’s team of domain experts," says Baraniuk. Such a process, he claims, could convert a suitable electronic version of Campbell Biology into an intelligent textbook in a few hours of computer cycles.
Rice students have been beta-testing OpenStax Tutor for about a year, and Baraniuk plans to unveil a first book later this fall.
"I’m not for a second claiming a machine-learning approach will immediately outperform what a specialized team of humans can do," he says. "But I am claiming that, just as in any engineering design, there are tradeoffs. Inquire’s human-centric approach works at one end of the scalability spectrum while OpenStax Tutor’s machine-learning approach works at the other."
Vulcan’s Greaves agrees. "Yes, they can do this at low cost and with considerable cleverness once they have the source modules. Meanwhile, Inquire can endow textbooks with AI so they are able to answer human questions about the material. Both methods ought to help students–but in very different ways–and both point to an exciting new kind of textbook."
Paul Hyman is a science and technology writer based in Great Neck, NY.