The history of learning machines, from Sidney Presser and B.F. Skinner to McGraw-Hill

Inventors Have Tried to Reinvent the Textbook for Nearly a Century. They’ve Mostly Failed.

Inventors Have Tried to Reinvent the Textbook for Nearly a Century. They’ve Mostly Failed.

Comments
Slate Plus
Your all-access pass
Oct. 26 2015 9:01 AM
Comments

The Fascinating, Mostly Failed History of “Teaching Machines”

And how far we’ve come since the 1920s.

151023-Pressey-Testing-Machine
Pressey’s “teaching machine.”

public domain/wikipedia

If textbook publishers and ed-tech startups have their way, American classrooms may soon become very different learning environments. A new wave of software that draws on 21st-century technologies like cloud computing, online behavioral tracking, and machine-learning algorithms is beginning to replace print textbooks, as I reported in a lengthy Slate feature this week. These technological leaps won’t just affect the materials students learn, however. As I explain in my piece, they stand to reshape the processes of teaching itself.

Will Oremus Will Oremus

Will Oremus is Slate’s senior technology writer. Email him at will.oremus@slate.com or follow him on Twitter.

But while much of the technology is new, the concepts underlying these changes aren’t nearly as novel as you might think. “This whole ‘learn at your own pace’ idea has been the goal of education technology since the 1920s,” notes Audrey Watters, an education writer who counts herself as a critic of the “adaptive learning” movement. “ ‘Programmed instruction’ was one of the early names for it.”

Indeed, the history of 20th-century ed tech, as recounted by Bill Ferster in his 2014 book Teaching Machines, can be read as a litany of failed efforts to build the very kind of adaptive textbooks that are finally catching on today.

As early as 1926, a behaviorist psychologist named Sidney Pressey assembled from old typewriter parts a device he called the Automatic Teacher. It presented the student with a series of multiple-choice questions, one at a time, arranged from least to most difficult. Only by answering one question correctly could you move on to the next one. Pressey’s stated goal was to relieve teachers of the onerous task of grading students’ exercises so they could focus on more rewarding forms of interaction. By 1929 he had inked a deal with a scientific-instruments manufacturer to produce and sell the device to schools. But production costs ran over budget, the onset of the Depression dampened demand, and Pressey was out of business within a year.

151023-skinner-machine
Skinner’s machine.

public domain/wikipedia

His device wasn’t forgotten, though. In the late 1950s, the legendary behaviorist B.F. Skinner revived some of Pressey’s ideas in a device of his own, called the Teaching Machine. Inspired by the Socratic teaching method exemplified in Plato’s Meno, Skinner divided complex lessons into a series of small questions, each requiring a response from the student and each building on the correct response to the one that came before. Skinner’s machine required students to master one concept before moving onto the next. This same “mastery learning” approach motivates ALEKS and other adaptive courseware today.

Skinner’s contraption made headlines—“Will Robots Teach Your Children?” Popular Science panted—and by the 1960s teaching machines had proliferated. But, as Ferster explains, they struggled to find a mass market, due to a combination of high costs, skepticism from teachers and school administrators, and structural inertia. Mastery learning, in which students move through the material at their own pace, proved to be a tough sell in an education system predicated on moving every student through 12 grade levels at the same pace. In many ways, it still is.

Today’s technology, including high-speed Internet and powerful mobile devices, makes the latest generation of adaptive software far more user-friendly than its predecessors, not to mention much faster and cheaper. For the first time, adaptive learning appears to be commercially viable. The question now is: How intelligent can it get?