Education Technology - Sponge cake version

As of the present (2024), I have spent little more than half a decade in ed-tech, and all of my life before then in education in one way or another. So luckily, my career aligns with my ideals, my stray curiosities overlap with my professional industry. Since I was a child I’ve always loved to help my friends learn, which then led me to being a tutor as the first thing I happily did for ‘work’ (the other being writing educational articles in a tech magazine), and then working at Gradeslam writing software for an education platform. I’ve been thinking about these things for many years now.. on land, in the sky, on the water, and in my dreams. So what have I been doing with my life? .. I mean … What even is education really?

What is Education?

I’m not totally satisfied with any definition I’ve found yet. Some are too specific, some too contextual. I could get into examples but this is no academic paper so I’ll just posit my own definition - Education is a purposeful proliferation of information that is relevant to a being surviving and thriving in the world. The world is both physical (for which relevant information includes chemistry, ecology etc), and symbolic (for which relevant information includes the social sciences, history, etc) so an education should help one live in a cultural urban jungle as much as a literal forest. Coincidentally, Math is therefore special, since it is both physical and symbolic, as it deals symbolically with the physical world, and also has to do with the ‘physical’ aspects of symbols (how they make sense). That’s another infinite rabbit hole, not for the moment.

Education is a “proliferation” of information because it consists of both Learning and Teaching. In the base case, a Learner could obtain an education from the world and their experience in it, without a dedicated Teacher, one avenue for which is something we colloquially call “the school of hard knocks”. A Teacher expedites Learners’ journeys, increasing success rates statistically as well as individually, somewhat like a mountain guide who knows which passes to take and which cliffs to avoid falling off of. Our educational institutions are systems that facilitate Learning and Teaching.

Acquiring random information is technically learning, as it might change the quality of your experience, but in my view, it isn’t educational if you have no meta-information/inkling about the potential utility and relevance of the information to your life. A successful education should be useful to the ones being educated, as well as to society and civilization at large, as we all tend towards “a good life” in our own, hopefully mutually beneficial, ways.

Enter Technology

Technology is the application of knowledge, to help us do something we want, which is usually a better version of something we already do. Then we want to be better at the improvement and go to higher orders (mucho meta) and ultimately end up with things that can now be called unnatural (which is not always a useful amount of abstraction - like bits of information stacked into bitcoin).

Education, in its most simplistic sense, starts as something natural, since various species of animals exhibit behaviours dedicated to the proliferation of information relevant to survival. Education technology should help make education better and more effective. “Better and more effective” are mostly subjective qualifiers until we really decide on, what are requirements that could define a ‘successful’ education? Without this, technology will be aimless at best, and sometimes harmful, whether it take the form of an app on a screen or a sci-fi device that can sense everything around you or even scan your brainwaves.

Apologies but I have to bring a value into this, but I think enough of us might subjectively share it that we can say it tends to an objective value - Freedom. If we truly fundamentally believe in freedom (to the limits of not encroaching on that of others), we can apply an axiom of agency to our definition of education as information transfer. A good education is something that increases the agency (as a capability of consciousness) of the beings being educated, on top of the improved survival. Bad education is information acquisition that can end up causing pathogenicity in the information-related capabilities of the agent/being/Learner. This is a necessary logical pre-condition, just like a tolerant society needs to be kinda intolerant of intolerance to survive. (I say “kinda” because the reality of tolerant or not is on a spectrum of control exerted and freedom encroached.)

Consider the phenomenon of domestication - It is a technology, to alter ecology of entire species and change their way of being, for food and pets. In some sense, domestication is also education (of the animals), although if the animals have anything close to our complexity of consciousness, we can sense that we’d call it a “bad” education. There’s a higher degree of selfish motive from the system designers/administrators, and the animals have no say in the change, other than the behaviours resulting from their DNA. Hence the tropes of dropping out of uni to change the world, and skilled labour being called sheep, pervasive in rebellious youth and counterculture. It’s only natural we start out mentally colorblind, then life keeps getting better as we get educated. Or is that a tautology?

So let’s make it simple, what is ideal education technology?

To slay this monster of a question with verbiage containing adequate mass of meaning, delivered accurately enough, I want to take the simplest case and look back at my lived experience as a tutor, helping a student learn, using primitive pencil and paper. What happens in such a situation, and in what order? What are the relevant events and behaviours that we want more of? I will try to recount an abstracted experience in phases:

1. Map the world and align on a destination to get the journey going

To begin, I need to get in alignment with the student. I try to understand what the Learner is trying to learn, and the problem they are facing, to establish common purpose. Then to know how best to help, I also need additional information about the Learner, their competence (and interest) with that subject matter, and any relevant proclivities. I get to know their familiarity with adjacent, constituent, or derivative things, and I establish common vocabulary and meaning so I can communicate with them efficiently. Their knowledge and competence is the minority of the picture though, I mainly make sure they are interested and invested in getting through their learning journey.

1. Identify concept in question, relevant knowledge components (KCs)
2. Infer/Assess interest and familiarity with proximal KCs

2. Keep the journey going and improve the map. Look out for hazards and deal with them. Stop and admire the scenery along the way.

Then we get to prodding the problem. I ask questions around it to get clarity, and usually discover one of two things - a misconception, or a roadblock. The misconception looks like a normal road but it never takes you to your destination, and the roadblock sometimes looks like a chasm when it’s actually just a pothole. This phase involves some backtracking and introspective investigation, like auditing your holiday packing before a trip. If I have identified the missing/incorrect piece of understanding, I try nudging the Learner towards it without directly pointing (to encourage the behaviour of looking around), else I go for a carpet bombing approach to then help the Learner build completely new structures for themselves. My favourite tools here are analogies and models. This is the part that we’d most likely think of when we hear ‘teaching’. Some of what we’d call ‘teaching style’ also should tend to objective aims for teaching - using content, form and tone that is respectful and encouraging, destroys learned helplessness, and increases grit and metacognition. In the happy path, by this point, the student is also identifying their own gaps and asking for information as needed. The key (for me) is to avoid force-feeding, and help the Learner fish for themselves instead of giving them fish. Maybe a bite to keep them alive if their information is really not incrementing for whatever reason.

3. Identify missing/incorrect concepts. Shine (gradually bright) light on it if unnoticed. Disentangle the faulty connections with existing knowledge.
4. Provide just enough scaffolding as needed, titrating information until progress is made.

3. At the destination, make sure you’re really there. What’s fun around here?

Now it looks like the problem is solved, on the surface, but we don’t really know. Either way, the Learner has put in effort, which is worthy of praise, and any accomplishments in learning are likely to last longer if celebrated. We want to re-inforce the new mental structures (..as much as possible in the moment, since proper long term memory will require practicing recall and re-inforcement over longer time scales.) Here my pedagogical tool of choice is self-explanation and its variations, asking the student to make an analogy, construct a concept map, or assessment questions for their future self or other students. (If possible, these can be turned into anki flash cards.) If the student still has fuel in the tank and is revving to go, we can even go further than the problem du jour, or move on to the next mountain.

5. Re-inforcement / competence building tasks (hypothesised Learning Events)
6. Assess competence and goto 3 or 4 as needed (infer LE efficacy, update ZPD and learner KC tree)

Applying knowledge of knowledge to its acquisition for learning

As other thought leaders in ed-tech have posited, a future where you’re holed in to some device to learn seems dystopian. Also, to be in alignment with the definition above (regarding utility of learning in the world) we implicitly require measurements and demonstrations of competence to be ‘in-situ’ as much as possible, using toy models if not the real thing. It may be useful to consider examples of subject matter here - mathematics on one hand which is perhaps the subject with the most objectivity and abstraction, amenable to interactive visualizations we can code and customize, such that the education can live almost entirely within technology. On the other hand, we may have something highly hands-on like carpentry, where most of the education happens ‘out there’ in the real world (until we can simulate it well enough, like the virtual labs for medicine etc that the pandemic made us make).

The causal chain that we can use to improve is: Changes in instruction -> changes in learning -> changes in knowledge -> changes in robust learning measures. For the academic robustness, we can look to learning science stuff, like the KLI framework, which is built on that same causal chain.

The biological technology, and learning to dance with it

Here is the thing I keep bumping into. Almost none of what makes learning work is in the software. It is in us, in a piece of biological technology we were issued at birth and cannot send back for repairs: a memory and an attention system with very particular habits. You can build the loveliest app in the world, but if it fights those habits it loses, the way a beautiful boat loses to a tide it decided to ignore. So the interesting design question stops being “what can the screen do” and becomes “what is the hardware already doing, and how do we move with it rather than against it”. A few of its habits matter more than the rest.

The first is that memory forgets, and it forgets on a schedule you can actually read. Cepeda and his colleagues sat more than a thousand people down with a pile of obscure trivia, brought them back to review it after gaps ranging from minutes to months, then tested them anywhere up to a year later. Out of that fell a ridgeline: for any given “how long do you need to hang onto this”, there is a best gap before review, and that best gap is a shrinking slice of the horizon. Somewhere around a fifth to a third of the way out if you need it for a week, nearer a twentieth to a tenth if you need it for a year. Cram it all into one sitting and it evaporates; leave the right amount of forgetting between visits and it sets like concrete. The kind part is that you do not have to be exact, a whole band near the peak works nearly as well, you just have to stay off the two cliffs of cramming and waiting far too long. Every “let’s come back to this next week” a good tutor mutters is a rough shot at that ridgeline.

The second habit is that we think through a very small window. Working memory holds almost nothing and holds it briefly; long-term memory is effectively bottomless; and the small comedy of the whole thing is that everything has to squeeze through the small window on its way to the big room. Cognitive load theory, which is John Sweller’s baby, splits the squeeze in two. There is the difficulty that belongs to the material, the honest number of moving parts you have to hold at once, and there is the difficulty you bolt on by presenting it badly, splitting a diagram from its label so the eye has to keep ferrying back and forth to reassemble the meaning. The second sort is pure tax, and it turns out to be the more expensive one. When Lepine and his colleagues watched financial professionals work alongside GPT-4o and estimated both kinds of load from the transcripts, the self-inflicted, presentation kind dragged on the quality of the work about three times as hard as the material’s own difficulty. There is a sting in the tail, too: the scaffolding that rescues a beginner turns into clutter for an expert, one more thing they now have to notice and step around. Which is why “just enough help” is not a platitude but a moving target, and why a system that dumps the same hints on everybody is quietly making its strongest learner slower.

The third habit is the sneaky one, because it makes progress feel like its opposite. Drill one kind of problem over and over in a tidy block and you will get visibly, satisfyingly better at it inside the hour. Then you will forget it. Shuffle the problem types together so you never quite know which is coming next, and you will fumble more today and keep far more of it tomorrow, and, the real prize, carry it across to problems you have never met. Schorn and Knowlton found this even in people learning finger sequences they could not consciously describe, so it is not some trick of the studying-conscious mind; it reaches all the way down. It works because the block lets you coast on autopilot, running the same move without deciding anything, while the shuffle forces you to keep asking “wait, which kind is this one”, and that small act of deciding is exactly what the test, and the world, will ask of you later. The uncomfortable consequence for anyone building a learning product is that a smooth, confident, everything-is-green practice session is often a warning light, and a dip in today’s score can be the machine working as designed. Try charting that on a progress bar.

The fourth is that “knowledge” is not a single substance, and pretending it is one is why the research keeps looking like it is arguing with itself. One camp swears by testing and spacing, another by worked examples and patient explanation, and they are both right, about different things. Koedinger and his colleagues’ KLI framework brokers the peace by noticing that these methods aim at different kinds of knowledge component: a bare fact you just need fluent wants retrieval and spacing, a rule you have to induce from scattered cases wants varied examples and feedback, an idea you have to genuinely understand wants you to explain it back in your own words. So the honest first move is not to pick your favourite method and preach it, but to take the thing being learned apart into its pieces, ask what each piece actually is, and only then reach for the matching tool. Which, if you glance back up the page, is quietly what the six tutoring steps were doing the whole time.

None of these four are features I can code. They are facts about the learner, older than any curriculum and indifferent to my roadmap, and the best a piece of education technology can do is the thing the pencil-and-paper tutor does without ever naming it: watch which of them is happening right now, and move with it. That is the dance. The software does not get to lead.

So how do we build education technology for Learning in the second millennium CE? (Part 2)