Review: Teaching Machines

Teaching Machines: The History of Personalized Learning

Author: Audrey Watters
Publisher: MIT Press
Published: August 2021
Review date: (updated) 20th December 2021

Review by: David Longman, TPEA (1600 words)

An important theme to emerge from reading ‘Teaching Machines’ by Audrey Watters (MIT Press, 2021) is that the ‘industrial age’ of mechanised educational technology has not come to an end, as some might believe. Instead, it is thriving.

In her introduction Watters summarises, among others, the argument of Sal Khan (the creator of Khan Academy) that for the first time [my italics] online learning enables a truly personalised approach to learning in or out of school. At last, online learning enables us to break free from the stifling effects of an outmoded education based on regimented, bureaucratic organizations that fail to enable effective learning. It is, however, a too familiar critique of education that has served its time as a rationale for disruptive innovation in education, one often taken up by those proponents of learning technology who argue that schooling is somehow broken.

As Watters argues and aims to demonstrate in this book, the ‘end of history’ story on which Khan bases his claims is wrong. She goes further. Not only is he wrong about the unchanging face of schooling but he denies history. It reflects a general ‘Silicon Valley’ inspired ideology that the past is irrelevant and that only the future matters. In this way, old ideas can be recast as unprecedented, innovative and disruptive to a moribund educational system:

“What today’s technology-oriented education reformers claim is a new idea — ‘personalised learning’ — that was unattainable if not unimaginable until recent advances in computing and data analysis has actually been the goal of technology-oriented education reformers for almost a century.” (p9)

Teaching Machines is a well-researched and largely well written illustration of this history, providing critical commentary on the notion that machines can be so designed as to afford effective learning with minimal intervention from teachers. It also illustrates that implementing machine-based learning at scale has long been a challenging ambition (again an issue not always acknowledged by contemporary disruptors). Although her focus is American education the story is relevant to the design and implementation of technology in similar school systems.

The scope of the book is confined to key figures and projects in the quest to develop mechanical tools for school learning and teaching covering the years from about 1920 to 1970. The technology under discussion is literally boxes containing gears, rolls of paper or similar media, occasionally electric lights (and in an early version even a dispenser offering chocolate bars), all controlled by simple levers and keys for user input. Although now superseded, many of the main features of these pre-digital devices remain familiar today in much educational software:

• to present a ‘unit’ of content, usually offering a question or task;
• to provide a means for a learner to respond;
• to provide feedback on the response (ideally immediate);
• to move to a new ‘unit’, or repeat the current one, based on that feedback.

A central figure running through this story is B.F. Skinner, not so much the originator of Behaviourist psychology (its origins trace back to the early 1900s), as its post-war bête-noire doggedly promoting his mechanical teaching machine as a device to change the manner and quality of school-based learning. The story of Skinner’s machines covers a surprisingly short period from about 1949 to about 1969, with antecedents in the 1930s. However, by the end of this period there was limited tangible effect on education beyond a few successful if not entirely rigorous trials.

Although today mechanical teaching machines tend to be known as ‘Skinner machines’, he was not the inventor of the concept. First World War recruitment revealed low levels of health and education in the population. Then, as now, education was perceived to be poorly managed with overworked and ineffective teachers. At war’s end a new emphasis was placed on testing for attributes such as intelligence or retention of learning. A key figure here was Sidney Pressey and his ‘automatic teaching machine’ that first appeared in 1923. Pressey foresaw an “industrial revolution” in an education system he regarded as stuck at the “crude handicraft stage”. However, like Skinner after the Second World War, he experienced frustration with efforts to develop his device as a commercial enterprise (though he was unfortunate to be doing this in the midst of the Great Depression).

Skinner’s own early work at Harvard in the 1930s focused on developing the idea of behavioural conditioning using devices he made known famously as “Skinner boxes”. With these he trained pigeons to get food when the correct lever was pressed. His insight that these techniques might be applied to school learning came in 1953 when he visited his daughter’s fourth grade class (the top-end of UK Key Stage 3). As Skinner told the story he was shocked to see that the classroom teaching of basic arithmetic failed to meet what he regarded as the minimum conditions for learning, namely progression matched to ability (i.e. ‘stimuli’ in behaviourist terms) and timely feedback (for the pigeons an edible reward). Good students were held back, he observed, while those who needed help could not keep up. “The teacher”, he declared, “is out of date” and cannot provide adequate and timely feedback (i.e. reinforcement) to many children at once. Echoing Pressey (who he met and corresponded with during these post-war years), Skinner also declared that an “industrial revolution in education” is needed.

Interestingly, even as Skinner promoted his approach to Behaviourism as the true science of learning a paradigm shift was already beginning to take place that would challenge this viewpoint. Cognitive Science began its emergence as the new science of psychology and human learning (an early conference was organised by Jerome Bruner in 1959) and the term ‘Artificial Intelligence’ as a framework for understanding human thinking was adopted by a community of scientists and engineers at the now notorious Dartmouth College conference in 1965.

There is much in this book to interest students of the history and origins of contemporary education technology as well as commentators on the current scene. In 1958 Sputnik stunned American national pride in its cultural and scientific prowess leading to many reforms. In education a new focus was brought to the teaching and learning of critical subjects as mathematics and science. While this gave a boost to the idea of teaching machines many problems about their design and usefulness remained to be solved. Content was a major issue and the field of Programmed Instruction (PI) grew in importance as more work was done to enhance the quality of both what was ‘taught’ by machine and how it was taught through sequence and structure.

It is a familiar story to contemporary readers that efforts to verify the value of this new technology through large-scale school-based investigations were largely ineffective or that by contrast projects by publishers to produce cheaper textbooks designed on PI principles (what today we might call branching texts) were relatively successful. Like Pressey before him Skinner had chronic difficulties in persuading his manufacturing partners to produce machines to a quality benchmark that satisfied him. Even so, it was hard to compete because his machines still lacked content. He later entered into partnerships with encyclopedia publishers because their established door-to-door sales model helped to forge a strong link with the idea of home-based learning and ‘free’ teaching machines were offered with every encyclopedia sale.

However, these efforts did little to enlarge the market or make teaching machines more acceptable to educators. In schools, the familiar issues of staff training, machine reliability and the scarcity of curriculum content dogged the enterprise. Moreover, by the turn of the decade in 1960–70 new ideas about pedagogy were emerging. This included a backlash against Behaviourism and increased dissatisfaction with post-war efforts to reform the teaching of mathematics and science. But Watters argues that teaching machines did not simply die out; they were absorbed:

“… many of the key figures in the teaching machine movement did not suddenly stop working in teaching or training when the focus turned to computer-based education. Many of the ideas that propelled programmed instruction persisted and spread into new practices and new technologies.” (249)

Here she makes a case for the idea that behaviourism and its core concept of conditioning did not disappear from the mainstream of educational technology, as its most articulate critics might argue, but continues to inform the design aims of present-day educational technology. Indeed, it is fundamental to the massively successful “industrial” character of our new digital technology culture for it too relies on sophisticated techniques of “behavioural engineering” to actively nudge (or push) our preferences, desires, ideas and opinions towards ends that we often serve unwittingly. This is an important point of view and one that needs careful consideration.

It’s a case of: “Behaviourism is dead! Long Live Behaviourism!”

If there are limitations to this valuable critique of teaching machines it is, perhaps, that the idea of personalised learning, the subtitle of this book, does not get quite enough detailed critical attention that the reader might expect. We are also left unsure about the details of programmed instruction methods using the ‘old’ media of paper rolls, punched card discs, projected slides and so forth so that we might ponder how much of that early work has been carried forward. There is also a lack of illustrations leaving the reader to somehow imagine the various machines described in the text but which are nowadays quite unfamiliar. Perhaps MIT Press could have included more. Audrey Watters has curated a good collection of images and technical descriptions on her blog (for example see this page) and more of that material could have been included to enhance this noteworthy book.