We need to place high-quality curated knowledge for the next generation in a modern digital context.
It breaks my heart when the costs of college-level textbooks are labeled as "junk fees." Or, when people complain that today's textbooks are "overstuffed, chopped-up monstrosities" that are boring and do little to advance student learning.
Yet all this angst over textbooks misses these crucial facts:
- Textbooks provide a carefully curated body of knowledge.
- Textbooks need to be nearly 100 percent accurate.
- Textbooks offer the view of an expert educator. Their quality cannot be compared to the free content that we find online.
It is time to reaffirm the indispensability of textbooks' high-quality curated knowledge for educating the next generation -- while moving them into a modern digital context and making them better at the same time.
Digital technology makes it possible to improve the clarity, precision and utility of a textbook. A process known as "knowledge engineering" powers those improvements. It involves a systematic choice of concepts, a rigorous mapping of the relationships between those concepts and logical definitions that allow them to be used by artificial intelligence and understood by humans.
Authoring textbooks so that their concepts can be read as computer code will usher in a transformation of publishing technology potentially greater than the development of Gutenberg's printing press. We should embrace this process now.
Consider an illustrative application of this idea. While writing a biology textbook, we can logically define two biological terms as follows: A Eukaryotic cell is a cell that contains a nucleus surrounded by a membrane. An enzyme is a macromolecule that plays the role of a catalyst and has the function of decreasing the activation energy.
In the definitions above, "Eukaryotic cell," "nucleus," "membrane," and "enzyme" are concepts that have meanings specific to biology; "is a," "contains," "plays," and "has the function" describe relationships. The concepts and relationships taken together are referred to as a "controlled vocabulary," and they are applied across the whole book. Once the controlled vocabularies are created, they can be used to introduce more complex concepts.
The simple act of using a controlled vocabulary is vital for learning any complex topic. In a subject such as biology, students have to master thousands of new words. Other complex fields, such as law, finance and medicine, are no different.
Knowledge engineering is not a new idea -- library scientists and linguists regularly practice it. The creators of the "Longman Dictionary of Contemporary English," for example, engineered it in a way that over 70,000 words can be defined using a core vocabulary of only 2,000 words.
Books are linear while knowledge is not. Mapping of relationships across vastly different topics makes learning exciting and supports many novel capabilities, including precise connections across chapters, visualizations at multiple levels of detail and the development of textbook expert chatbots that can answer students' questions on the fly.
As an initial test of this approach, a team led by SRI International authored several chapters of a biology book. The resulting "intelligent textbook" was tested in a pilot study. SRI reported that the average homework and test scores of students using the intelligent textbook were 10% higher than those of students using the regular textbook.
Similar results were observed in a monthlong classroom trial of the intelligent textbook led by Lena Tibell of Linköping University. During the trial, students' germane mental load -- an indirect measure of learning -- increased significantly. This indicated that the students were learning much more. At the University of Pittsburgh, Prof. Peter Brusilovsky has shown the effectiveness of knowledge engineering for an information retrieval textbook.
Knowledge engineering also enables us to create concept-checkers for textbook authors that are more powerful than spell-checkers. For example, an author can ask the questions: Have any concepts or functions been left out of the textbook? Are any concept definitions inconsistent? This kind of support is unimaginable using current authoring tools.
However, creating such textbooks will not be inexpensive. It is also not always possible to give a logical definition for every concept in a textbook.
Our ability to give logical definitions often ends at the boundary of current knowledge. Reaching those boundaries can be used to prompt wonder and curiosity. For example, one biological concept that is difficult to define is "life" itself. To define such concepts, we must articulate exceptions and multiple points of view.
To seize these opportunities, we need a sustained effort to overhaul the authoring of textbooks using knowledge engineering. For maximum near-term impact, we should begin with textbooks for gateway courses such as biology and psychology.
We must train and nurture a new profession of knowledge curators for curricular resources to enable this transformation.
Most importantly, the textbooks of the future must continue to preserve and communicate the knowledge of the best educators of our generation even more accurately and precisely than before. Such textbooks, using the discipline of knowledge engineering, can support the curation, preservation and learning of all forms of human knowledge.
Vinay K. Chaudhri led AI research at SRI International, including Project Halo and CALO (later SIRI). He taught knowledge graphs and logic programming at Stanford University, where he helped launch the Logic for All initiative.
This commentary first appeared in The Hechinger Report.