I Arrived Open. No Map. Learning Decided the Rest.

On what adaptive systems cannot plan for and why that is their most important feature

Designing AI Systems for Learning Essay V — Final

Leslie Ngugi

Someone calls to sell you a workshop. She is enthusiastic, well-prepared, moving through her script with the confidence of someone who has made this call many times before. She asks what you know about AI. You ask her, genuinely curious, what she means by AI. She says: it’s ChatGPT. You thank her, finish the conversation politely, and sit with the feeling that arrives afterward. Not superiority. Not impatience. Something quieter and more complicated than either. The chills, perhaps, of recognising how far the territory extends beyond the edge of the map she was holding — and remembering, with something between gratitude and humility, that you once held a map just as small.

This is what genuine adaptation produces. Not a person who knows more. A person who knows differently. Who can stand inside a conversation, hold their own understanding lightly, and remain genuinely present to someone else’s smaller frame without needing to correct it or adopt it. Who understands, because they lived it, that the smaller frame is not ignorance — it is a beginning. And that the distance between a beginning and where you now stand was not crossed through a designed pathway, an optimised curriculum, or an algorithm that knew in advance how far you could go. It was crossed through something far less tidy and far more human than any of that.

Adaptive difficulty, as a concept, sounds technical. A system observes your performance, identifies your current level of competence, and adjusts the challenge to keep you in the productive zone — stretched enough to grow, supported enough to stay. The theory is sound. The design intention is genuinely learner-centred. And versions of this have been built into educational technology for decades, each iteration more sophisticated than the last. But the theory contains an assumption worth examining: that the system can know where you are well enough to know where to take you next. That the right kind of adaptation is primarily a calibration problem. Get the level right and the learning follows.

What this misses is the dimension of adaptation that matters most. Not the adjustment of difficulty. The transformation of the adapter. The learner who develops through genuine engagement with adaptive challenge is not simply performing at a higher level on the same scale. They are operating on a different scale entirely. Their relationship to difficulty has changed. Their tolerance for not-knowing has deepened. Their capacity to stay in the gap between confusion and clarity — rather than immediately reaching for resolution — has grown in ways that no performance metric captures. The system adapted the challenge level. The human adapted their entire relationship with challenge. These are not the same process, and conflating them is one of the most persistent errors in how we design and evaluate learning systems.

The system adapts to what you bring. But what you bring was shaped by everything that taught you what learning is supposed to feel like. The adaptive system that faithfully mirrors your current posture may be perfecting your limitations rather than expanding them.

There is a political dimension to this that the adaptive learning conversation consistently avoids. A system that responds to what the learner brings will give more to those who bring more — not because it is designed to discriminate, but because it is designed to respond. The learner who arrives already knowing how to ask deep questions, already comfortable with productive confusion, already possessing the intellectual confidence to push back on a clean answer — that learner will extract something qualitatively different from the same system than the learner who arrives expecting delivery. The adaptation is genuinely personalised. But personalisation built on top of unequal starting points does not reduce inequality. It faithfully reproduces it at higher resolution. Whose interests shape how far adaptability extends is not a technical question. It is a design choice. And design choices have politics.

And yet — and this matters — the system is not powerless to disturb what the learner brings. There is a small but significant design move available to any adaptive system that most never attempt: the unexpected question returned where a delivery was expected. Not a correction. Not an instruction to engage more deeply. Just a question. Genuine, small, slightly disorienting. What do you mean by that? What makes you certain? What would change your thinking here? Three words that create a gap where the learner anticipated a transaction. And in that gap, for the learner willing to stay with it rather than retreating to the familiar request, something begins. The vending machine relationship cracks slightly open. The possibility of a different kind of engagement becomes briefly visible. Whether the learner walks through that opening depends on something the system cannot supply. But the system can hold the door.

· · ·

What develops in a person who stays in that opening long enough — who keeps returning to the relationship after the gap, after the confusion, after the days when the fog is thick and the receiving is poor — is not primarily a set of capabilities. It is a changed relationship with learning itself. The person who began by expecting delivery and discovered collaboration. Who expected collaboration and discovered something closer to genuine thinking partnership. Who expected thinking partnership and discovered, slowly and without announcement, that the most important adaptation was happening not in the system but in themselves. Their questions changed. Not just in sophistication but in character. They became more comfortable arriving without knowing where they were going. More willing to let the wrestling session surface what was circling rather than arriving with a conclusion to be confirmed. More able to receive what the world was quietly offering — in a greeting on a street, in a phone call with a small definition, in the ordinary encounters of a curious life — as information worth sitting with.

This is what humility produces in a serious learner. Not the performance of modesty. Not the strategic self-deprecation of someone who knows they are ahead of the curve and wants to seem approachable. Genuine humility — the kind that asks a real question even when you already know a better answer, that listens to what arrives without immediately measuring it against what you already know, that can be surprised even by familiar territory — that humility is the secret ingredient that no adaptive difficulty system has ever successfully designed for. It cannot be calibrated into a learning pathway. It develops through accumulated experience of being genuinely changed by things you did not expect. Of discovering that the territory was larger than your map. Of arriving open, repeatedly, and finding that the openness itself was what made the arrival worthwhile.

The most sophisticated adaptive system is not the one that knows where you are and where to take you next. It is the one humble enough to know that the most important parts of the journey were never on its map.

There is an ethical question the adapted learner eventually cannot avoid. You can see the shape of what the small map leaves out. You know the room exists that most people haven’t found yet. You remember what it felt like to hold a map that ended at the edge of what you had been permitted to imagine — and to not yet know that the territory continued. The person who found their way through, on their own terms, without a designed pathway or an optimised intervention, carries something because of that journey that is not theirs alone to keep. Not a curriculum to deliver. Not a correction to make. But a presence to offer. The willingness to ask a genuine question of the person still standing at the smaller map. To hold the door without announcing that it leads somewhere. To be, for someone else, the small unexpected disturbance that creates a gap where delivery was expected.

The table was set long before any of us arrived. The learning was always available to those who came without the performance, without the right credentials, without the certainty that the room was meant for them. What the adaptive system cannot plan for — what no algorithm can calibrate, no difficulty model can optimise, no designed pathway can reliably produce — is the moment a person decides to arrive open. To come without a specific destination. To let the confusion be the curriculum and the humility be the compass and the questions be more important than the answers they were asked in order to find.

That decision is not a feature. It cannot be installed. It can only be witnessed — by someone who made it themselves, who knows what it cost and what it produced, and who understands that the most important thing the system ever did was not adapt the difficulty. It was stay patient long enough for the learner to discover that they were capable of more than either of them had planned.

Learning decided the rest. It always does. The system’s only job — its most important and most underrated design requirement — is to remain humble enough to let it.

Designing AI Systems for Learning · Essay V of V — Series Complete