It’s the Instructional Design, StupidAnd how the debate over AI and screens is cutting across the old Prog vs. Trad lines.
In 1983, Richard Clark asked the field to consider a grocery truck. Media (i.e., technology), he argued, are “mere vehicles that deliver instruction but do not influence student achievement any more than the truck that delivers our groceries causes changes in our nutrition.” The truck matters for cost and reach. It does not matter for whether you end up eating vegetables. Swap the truck for a train and the groceries are the same groceries. What matters is the cargo: the method, the sequencing, the examples, the practice, the feedback. Clark drew the practical corollary bluntly. If two media can carry the same method to the same criterion, you are obligated to choose the cheaper one, or, as I’d argue, the more scalable one. I think about the truck a lot, especially now that I find myself embroiled in the debate on AI-enabled instruction. People look at the device and forget that the device is just the truck. People look at the human teacher — the choral responses, the brisk pacing, the warm-strict adult working the room — and forget that this, too, is merely a delivery system for something underneath. Again and again, people seem to care more about the carrier they can see than the design they cannot. The great media debateClark’s truck did not go unanswered. Robert Kozma’s 1994 reply made the argument that particular media have particular capabilities — the symbol systems they can employ, and the operations they can perform on those symbols: displaying, storing, transforming, responding — and that these capabilities, together with the methods that employ them, interact with the cognitive and social processes by which learners construct knowledge. Media and methods are “inexorably confounded” in good design: media constrain and enable methods, and methods in turn exploit what a medium can do. You cannot ignore the attributes of the medium, because the attributes determine what the learner can actually do with the instruction — manipulate a model, watch a process unfold in motion, see one example transform into another. Clark, unmoved, answered with his replaceability challenge: show me a media attribute with a unique cognitive effect, one that cannot be achieved any other way. If any attribute can be replaced by another and produce the same learning, then the attribute was never the cause; some underlying method variable was, and we’re back to choosing the cheapest, most scalable vehicle. The debate is usually taught as a dispute without a winner. It’s more useful to read it as a division of labor. Clark tells you where the causal power lives: in the method, the design. Kozma tells you that designs are not free-floating — they have to be executed through a medium’s actual capabilities, and a serious designer studies those capabilities rather than assuming them away. Neither man believed the device teaches. If Clark is right, the medium is beside the point — a computer can deliver instruction just as a teacher can, provided the design is sound. If Kozma is right, the medium matters, but only in this sense: its capabilities must be able to execute the method. Notice that neither position is grounds to reject instruction just because it comes through a screen instead of a human. Both views send you to the same place: the design. The “new” media debateAs someone who participated in the robust “Prog vs. Trad” debates on social media, I am realizing that the lines are being redrawn. Many of the skeptics of AI-enabled instruction are friends of explicit teaching — but friends of it at the level of small-di: the visible delivery techniques and craft of a skilled adult working a room. What’s missing from their awareness is Big DI: the design tradition the field has never cared to embrace — the “picky details” the delivery techniques were built to serve. And I do mean a tradition, not a product line. The published DI programs remain the best-tested curricula ever built, but the knowledge that built them lives in Theory of Instruction and related works. It is there to be used and extended, and almost nobody has.¹ Theory of Instruction (Engelmann & Carnine, 1982) is not a book about teacher behaviors. It is a logical analysis of communication: how to construct a sequence of examples and non-examples that admits exactly one interpretation, so that the learner cannot induce a misrule no matter what irrelevant features happen to co-occur. Faultless communication consists of the principles of sameness and difference that govern juxtaposition; the minimally different pairs that isolate the relevant feature; the progression from overt to covert responding; the specification of an observable response at every step so mastery is demonstrated rather than assumed; the correction procedures that treat an error as information about the communication rather than a fault in the child. None of that machinery is definitionally tied to a human delivery system. In fact, the relationship runs the other way: many of the features of DI that proved most controversial were invented because of the limitations of the human delivery system — one adult, thirty children, and no way to be everywhere at once. So you require unison responses, because a choral answer is how one adult samples every learner’s response on every trial. You script the wording, because learning falls apart when the communication is imprecise, and you require fidelity to the script because an improvising teacher drifts — and drift harms the kids who need precision most. The scripts, the signals, the unison answers — the very things critics point to as evidence that DI is rigid and dehumanizing — were never the point. They were brilliant adaptations to the capabilities and limits of the classroom-as-medium. Enter the app, which doesn’t need the workarounds: it attends to each learner individually rather than rounding up to the group, never tires of the script, never feels micromanaged by it, and delivers it exactly, every time, to every child. A new medium is free to solve the same problems its own way. Siegfried Engelmann — the senior author of Theory of Instruction and the father of Big DI — himself kept changing trucks. In the 1980s he built the Systems Impact videodisc courses, putting the presentation of examples on video: a medium that could show one example transform into a minimally different one, with perfect timing, every time, in a way a teacher at a chalkboard cannot. Later he built Funnix, a computer-delivered beginning reading and math program from the same analysis that underlies DISTAR. And when the University of Illinois built PLATO, the pioneering computer-based education system, engineers supplied the vehicle — and people trained in Direct Instruction, like Marty Siegel, designed cargo for it: reading and basic-skills courseware built on Engelmann’s stimulus analysis, which found some of its most consequential uses among learners the classroom had already failed — adults rebuilding basic skills, incarcerated students caught between basic education and the GED. What stayed constant across every one of those trucks was the design — and the design is the test. A machine that lets a child mindlessly tap through a series of disconnected tasks hasn’t passed the test. But a machine that adheres to a Theory of Instruction analysis will not only teach well — it can place each learner at the right entry point, catch and correct every error the moment it’s made, and enforce a mastery criterion with a consistency that’s structurally impossible in a class of 30. Redrawing the linesWhen people say a computer can’t deliver instruction, they have observed — correctly — that almost all educational software is bad. But Clark’s truck is still a truck. The typical edtech product is a truck stuffed with random freight — activities, videos, badges, no theory holding any of it together — or loaded with the same discovery-oriented cargo that fails in classrooms too. There is also a moral version of the objection to edtech, and it deserves an answer on its own terms. In the popular imagination, the screen is the babysitter, the pacifier, the withdrawal of adult attention. That is the picture that makes AI-enabled instruction feel icky, and it is not wrong as a description of most screens in most children’s lives. But a well-designed program is not the withdrawal of adult attention. It is the concentration of it: hundreds of hours of an adult’s analytical attention spent on sequences, examples, and corrections, delivered intact to every child who runs the program — including the children who were never going to be seated in front of a masterful teacher. If two vehicles can carry the method to the same criterion, and one of them can reach children the other never will, then choosing it isn’t icky. Refusing to consider it is. So yes, the lines are being redrawn — and not only in online debates. Families are increasingly opting out of institutional schooling altogether in search of instruction that treats evidence as something other than optional. The line isn’t “Prog vs. Trad”, but one between instruction that was designed meticulously and instruction that wasn’t — between lessons that were scientifically informed and tested against learner error, and lessons that have never been tested against anything: improvised anew every year, in a million rooms at once, by a system that seems content, even proud, to perpetually reinvent the wheel. The alternative is what Engelmann's Big DI tradition has asked of us all along: analyze the content, control every variable of the communication, test the design against learner error, and revise until the errors stop — then deliver it as efficiently as possible, starting with the kids who can least afford to wait. That is careful, unglamorous work, and it demands a fierce commitment to student accomplishment — not to grocery trucks, not to a certain aesthetic, not to any carrier at all. ¹ One notable exception is Kris Boulton, whose Unstoppable Learning Substack is a rare example of someone taking Engelmann and Carnine’s analysis seriously as a living research program — extending techniques like atomisation and transformation sequences to secondary mathematics, with a book, The Unstoppable Learning of Mathematics, on the way. References Boulton, K. (n.d.). Unstoppable Learning [Substack publication].
Clark, R. E. (1983). Reconsidering research on learning from media. Review of Educational Research, 53(4), 445–459. Clark, R. E. (1994). Media will never influence learning. Educational Technology Research and Development, 42(2), 21–29. Engelmann, S., & Carnine, D. (1982). Theory of instruction: Principles and applications. Irvington. Kozma, R. B. (1994). Will media influence learning? Reframing the debate. Educational Technology Research and Development, 42(2), 7–19. Zach Groshell is free today. But if you enjoyed this post, you can tell Zach Groshell that their writing is valuable by pledging a future subscription. You won't be charged unless they enable payments.
|
