a group of people walking on a street next to a building

Photo by Thuy on Unsplash

Essays 2026.05.08 Identity

The human school — what AI doesn't reach.

English / العربية

There is a specific claim being made loudly in education right now, across editorial pages and conference programs and venture-backed pitches, and it is worth sitting with. The claim is not that AI has failed in education. It is something more precise: that some educational work is irreducibly human, that no degree of adaptive personalization changes this, and that the line between AI-addressable and human-only is not drawn by cost or convenience but by the nature of the work itself.

This is not a nostalgic argument. It is an empirical one. And 2026 is a good year to test it.

The precision claim

Alpha School operates thirteen campuses across the United States. Its "2 Hour Learning" model compresses core academics — science, math, reading — into a two-hour morning block using what Alpha calls AI-powered adaptive software. Students, the school claims, test in the top one percent on standardized assessments. The remaining school day is devoted to life skills: financial literacy, public speaking, negotiation, arts, sport.

The CBS News investigation into Alpha found something worth pausing on. The software deployed in Alpha's morning block is adaptive applications similar to IXL and Khan Academy — rules-based or shallow-model tools that have existed since the 2010s. Not large language models. Not conversational AI. The "AI-powered" framing is marketing language for a well-established software category that has been in classrooms for over a decade.

This distinction matters, and not to embarrass Alpha. The two-hour learning model's honest version is actually compelling: adaptive software can handle pacing, drill, and formative check-in for core academic skills better than synchronous whole-class instruction, and that frees time. What the CBS finding reveals is that the "AI" in the AI-personalization pitch is being asked to carry a claim beyond its current capability. That gap is load-bearing evidence, not a technicality.

Three regions of educational work

The human-school argument, at its core, proposes a map. Not AI is bad, but AI has a territory, and that territory has edges.

The adaptive region is real and valuable. Software that adjusts difficulty based on a student's last ten responses, that catches the gap in fraction computation before a unit test, that ensures a student who needs twelve repetitions gets twelve and a student who needs three gets three — this is genuine educational work. IXL and Khan Academy do it. Alpha's morning block does it. The tools are real; the efficacy on narrow academic tasks is well-documented. Nothing in the human-school argument disputes this.

The relational region is different in kind. A teacher noticing that a student who was talkative last week has gone quiet. A mentor choosing the moment to press harder and the moment to hold back, based on signals that are not in any data feed. A guide who has watched a student fail three times and decides the fourth attempt should be public, because this student needs the experience of recovery in front of others. These acts require a particular kind of attention: sustained, contextual, caring in the full sense of the word. Adaptive software does not notice. It responds to inputs. The teacher notices — and the noticing is not a function to be optimized, it is the educational act itself.

The cultural region is where the argument is sharpest and most durable. Identity formation — the transmission of cultural memory, the practices that locate a person within a community, the understanding of what is worth preserving — happens in what a school treats as worth remembering. In the rituals it performs. In the silences it keeps. An AI system can reflect back what a community has already decided is important. It cannot make that decision. The decision is irreducibly collective and therefore irreducibly human. No personalization engine reaches this.

HCI 2026 — the conference just named it

The Human Capability Initiative Conference 2026, held May 3-4 in Riyadh at the King Abdulaziz International Conference Center, chose "The Human Code" as its theme. Over 23,000 participants, 550 speakers, with the United Kingdom as Country of Honor.

A conference of this scale naming this question is not incidental. HCI is the largest HCDP-aligned convening in the Saudi calendar — it is where human capability development intersects with sovereign policy. When its theme is "The Human Code," it is saying: the question of what is irreducibly human in education, work, and capability development is now a policy question, not a philosophical one.

That the Riyadh conference chose this frame in the same quarter that the editorial press (two EdSurge op-eds, three weeks apart, independent authors, the same underlying move) and the venture market (Sinai.ai's AI-native interactive books, Arabic.AI's voice infrastructure for language learning) all converged on the same question — that convergence is the news. The human-school argument is not one writer's position. It is the field's question of this moment.

Why it matters — operators, parents, policy

For operators building AI-personalization tools: the CBS finding is not a scandal. It is a clarification. Adaptive software that does what adaptive software does well — pacing, sequencing, drill — is genuinely valuable. The tools that will earn sustained adoption are the ones that are honest about their territory and excellent within it, rather than reaching for claims that require them to be something they are not yet.

For parents and families: the human-school frame is not an argument against digital learning tools. It is an argument for being specific about what they are for. Software that helps a child master fraction computation at their own pace is a good tool. It is not a replacement for a teacher who notices that the same child has been sitting alone at lunch for three weeks.

For policy: the question "what should human educators do when AI handles the pacing and drill?" is not answered by replacing teachers. It is answered by redesigning what teachers spend their time on — and by investing in the relational and cultural work that no software handles. That redesign is the hard, interesting, under-resourced work. It is also the work that matters most.

The closing map

The conservative-sounding move here is not nostalgia. It is precision.

Adaptive AI for pacing, sequencing, and drill. Human presence for noticing, for transmitting culture, for forming identity. Both — not as a compromise, but as a map of different territories doing different work.

The argument is not that AI education tools should slow down. Several of them are genuinely excellent at what they do, and the infrastructure being built right now in Arabic-first AI (Sinai.ai, Arabic.AI) extends that territory into contexts that need it. The argument is that the territory has edges. Knowing where those edges are is the precondition for building well on both sides of them.

Some educational work is irreducibly human. The work that is left over after the adaptive block finishes — the noticing, the culture, the identity — is not the remainder. It is the point.

Sources

Wiki pages drawn from

External sources

  1. "Inside the $40,000 a year school where AI shapes every lesson, without teachers" — CBS News, 2024. https://www.cbsnews.com/news/alpha-school-artificial-intelligence/
  2. Alpha School — official website. https://alpha.school/
  3. "I Tell My Students Writing Is Hard. I Still Ask Them to Do It Anyway." — EdSurge, 2026-04-01. https://www.edsurge.com/news/2026-04-01-i-tell-my-students-writing-is-hard-i-still-ask-them-to-do-it-anyway
  4. "Returning to What it Means to Make School Human Again" — EdSurge, 2026-04-22. https://www.edsurge.com/news/2026-04-22-returning-to-what-it-means-to-make-school-human-again
  5. Human Capability Initiative Conference 2026. https://humancapabilityinitiative.org/
Filed2026-05-08
TrackIdentity
Length1087 words · ~5 min
LanguagesEN ⇄ العربية
← Essays Index →