The Most Expensive Private School in Oklahoma: And It Has No Teachers

The most revealing document about Alpha School is not its marketing or its investor list. It is a job posting. Reporting by Oklahoma Watch on 21 July 2026 established, from Alpha's own recruitment materials, that the adults employed at the school do not touch any curriculum and are not in the room when students are learning.

Read that again. It is the entire proposition compressed into a single clause. The school has staff, paid between 100,000 and 150,000 dollars a year, two to three times what a first-year public school teacher earns in Oklahoma. And it has designed the role so that the person holding it is somewhere else while the teaching happens.

On Wednesday 12 August 2026, campuses opened in Edmond and in Tulsa. As of 20 July, twenty-six students were enrolled at the Edmond site, branded as the Oklahoma City campus and running from kindergarten through tenth grade, and thirty-two at the Tulsa site, which covers kindergarten through eighth. Tuition is 40,000 dollars a year, discounted by 10,000 dollars for five years for the inaugural cohort. Oklahoma City's most expensive traditional independent school, Casady, charges 27,950 dollars for high school. Tulsa's Holland Hall charges 21,625 dollars. Edmond's Oklahoma Christian School charges 14,640 dollars.

So the claim that Alpha will cost nearly double the state's most expensive private school is not quite right against Casady, where the multiple is closer to one and a half. It is comfortably more than double against the nearest school in Edmond itself. Either way, Oklahoma now has its most expensive private school, and it is the only one in the state with no certified teachers, by design rather than by accident.

That inversion is the story. For a century, what wealthy parents bought when they bought private education was human attention: smaller classes, more contact time, teachers who knew the child's name and noticed when something was wrong. Alpha has taken the price point and removed the product. What remains is a proprietary adaptive platform called 2 Hour Learning, two hours of laptop work each morning with half an hour for each of mathematics, reading, science and social studies, an afternoon of life-skills workshops, and a monitoring system that photographs children when it decides they are not progressing.

What Forty Thousand Dollars Buys When It Does Not Buy a Teacher

Alpha School was founded in Austin in 2014 by MacKenzie Price, who holds a psychology degree from Stanford and is not a trained educator, and Brian Holtz. Its principal backer is Joe Liemandt, founder of Trilogy Software and ESW Capital, who has committed a reported billion dollars to the model. The hedge fund manager Bill Ackman has endorsed it publicly. Tuition across the network runs from around 10,000 dollars to 75,000 dollars a year, with most sites clustered near 40,000.

The corporate architecture matters more than the celebrity endorsements. The schools are nominally non-profit entities contracting for services with for-profit vendors, including 2 Hour Learning for the software, Trilogy Enterprises for financial services and Crossover for recruitment, all connected to the same principals. This is the structure the education writer Peter Greene has flagged repeatedly: non-profit schools buying services from for-profit companies owned by the same people. Tuition does not simply fund a school. It flows through a school into a software business.

As of late July the school had not secured a location in Tulsa, where thirty-two families had already committed. In mid-May it sent prospective parents a report, prepared with AI, setting out two candidate sites; one recipient noted that the report cited incorrect data. In Edmond, Alpha leased more than six thousand square feet of industrial space for the Oklahoma City campus, and when Oklahoma Watch visited, the building was barren and carried no Alpha signage. The failure mode that runs through the curriculum appears first in the sales material.

Expansion is aggressive and roughly triples the network. Oklahoma Watch reported twenty-eight new campuses opening this autumn. Alpha's own press release of 5 August 2026 says twenty-seven, listing sites including Denver, Palo Alto, Chicago, Nashville, Charlotte, Boston, Atlanta and Miami Beach. The discrepancy of one campus is trivial in itself and worth naming anyway, because precision is the currency in short supply throughout this story. As of April 2026 the network had thirteen locations. Either way, a chain of thirteen schools becomes a chain of about forty in a single admissions cycle.

There is also the question of who is allowed in. Alpha's Oklahoma applications warn that the school cannot serve students requiring intensive behavioural, therapeutic or one-to-one academic support, and require families to disclose individualised education programmes, 504 plans or behaviour plans, on pain of withdrawal. That exclusion is not a footnote. It is the load-bearing wall under every academic claim the network makes, and we will come back to it.

The Strongest Version of the Alpha Argument Deserves a Hearing

Treating this as pure grift would be lazy. The instincts underneath it are neither fringe nor new.

Start with the thing Alpha is right about. Age-cohort instruction, in which thirty children of roughly the same birth year move through the same material at the same speed regardless of whether any of them have understood it, is a nineteenth-century administrative convenience rather than a finding from learning science. A child who has not mastered fractions is moved on to decimals because the calendar says so. A child who mastered them in October spends until June bored. Everyone who has taught knows this, and the profession's own literature has said so for fifty years.

Mastery learning is the standard rebuttal, and it has an evidence base. Chen-Lin Kulik, James Kulik and Robert Bangert-Drowns published a meta-analysis of 108 controlled evaluations in the Review of Educational Research in 1990, finding positive effects on examination performance across colleges, secondary schools and the upper primary grades, with the strongest effects among weaker students. Alpha's insistence that progression follows demonstrated mastery rather than seat time is squarely within that tradition.

Then there is the tutoring argument. Benjamin Bloom's 1984 paper in Educational Researcher reported that students tutored one to one under mastery conditions performed two standard deviations above conventionally taught peers, the famous two sigma problem. Kurt VanLehn's 2011 review in Educational Psychologist found human tutoring at an effect size of 0.79 and intelligent tutoring systems at 0.76, close enough that the gap between a good machine tutor and a good human one looked, for a while, like a rounding error. If software approximates the tutor, and tutoring is the gold standard, the case for a personalised two-hour block writes itself.

The caveat belongs here rather than later, because it is the honest one. Bloom's two sigma has never been replicated at that magnitude. Andre Nickow, Philip Oreopoulos and Vincent Quan reviewed ninety-six randomised tutoring studies in 2020 and found an average effect of 0.37 standard deviations, with none reaching two sigma. The ceiling Alpha is selling was always a research artefact, not a benchmark.

The afternoon programme is also more thoughtful than the mockery suggests. Workshops have included sailing, programming toy cars and Navy SEAL style challenges. One San Francisco session sent middle schoolers to a Starbucks with preloaded cards, and a guide deliberately zeroed one student's balance to teach the management of disappointment. That is a slightly brutal piece of experiential design, but it is design, aimed at something real that conventional schools handle badly.

Finally, the labour argument. Guides earn 100,000 to 150,000 dollars. Oklahoma's state minimum for a certified teacher with a bachelor's degree and no experience was 39,601 dollars in 2025-26. The National Education Association put the national average starting salary at 48,112 dollars for 2024-25. A model paying adults who work with children three times the going rate is doing something the profession has demanded for decades, even if it removed the teaching from the job to afford it.

Where the Evidence Actually Stops

Now the dismantling, which is mostly a matter of asking who checked.

Alpha claims its students grow 2.6 times faster than peers on nationally normed NWEA MAP assessments, that the best perform at 6.5 times, and that students sit in the ninety-ninth percentile. Oklahoma Watch reported an average SAT score of 1545 out of 1600 and average Advanced Placement scores of 4.2 to 4.4. These are extraordinary numbers, and they rest entirely on internal analysis of the network's own MAP data. Independent researchers have not been able to reconstruct how the 2.6 figure was calculated. The underlying data, the methods, the student selection and the wider conditions have not been subjected to external review, and there is no peer-reviewed evaluation of the model.

Absence of verification is not proof of falsity. But consider selection. A school that charges 40,000 dollars a year, declines students with individualised education programmes and 504 plans, excludes those needing intensive behavioural or therapeutic support, and admits children whose parents will pay for an experiment, is not comparing like with like when it benchmarks against national norms. Every one of those filters pushes measured growth upward before a single line of code runs. The 2.6 multiple would need to survive adjustment for all of them, and nobody outside the company has been permitted to try.

Regulators who have looked have not been persuaded. In January 2025 the Pennsylvania Department of Education rejected a cyber charter application from Unbound Academy, the online sibling of the Alpha model, on the grounds that the instructional approach was untested and failed to demonstrate how its tools, methods and providers would align to Pennsylvania academic standards. The department also cited unrealistic enrolment projections and doubts about special education capacity. Arizona's State Board for Charter Schools approved the same operator in December 2024 by four votes to three, which tells you the confidence interval on expert opinion here is wide.

Then there is what the curriculum looks like from the inside. On 17 February 2026, Emanuel Maiberg published an investigation in 404 Media, based on leaked internal documents and interviews with former employees, reporting that Alpha's AI-generated lesson plans were poorly constructed and presented students with illogical multiple-choice questions, that internal documentation described lessons that sometimes did more harm than good, and that the company had scraped material from other online course providers to train its own system. Former employees said students frequently needed considerably more than two hours a day, and that many required remedial work before high school coursework. One described students as being treated like guinea pigs. The reporting also found large volumes of student data, including video of children, sitting in a Google Drive folder accessible to anyone holding the link.

Alpha's chief communications officer, Anna Davlantes, told Oklahoma Watch that the 404 Media and Wired stories relied on cherry-picked examples and sources who had been dismissed. The company disputes the scraping allegations and says it respects intellectual property. Both investigations sit behind paywalls this article could not read in full, and the specific hallucinated items and unsolvable questions are described rather than reproduced in the available secondary coverage. That caveat stands on the record. What is not in dispute is that the curriculum is machine-generated at scale and that no certified subject teacher reviews it before a child encounters it.

Who Decides When the Educator Is Deleted From the Allocation

The most useful analytical tool for this problem was published in April 2026 by Conrad Borchers, Olga Viberg and René Kizilcec, and accepted at the thirteenth ACM Conference on Learning at Scale. It is called “Who Decides in AI-Mediated Learning? The Agency Allocation Framework”.

Its central move is to stop treating learner agency as a mood or a proxy for engagement and treat it instead as an allocation problem. In the authors' words, the paper reframes learner agency as the allocation of decision authority across learners, educators, institutions, and AI systems. The framework asks four questions of any system: how decisions are distributed, how choices are architected, what evidence supports them, and over what time horizons their consequences unfold. Among the recurring difficulties the authors identify are reliance on behavioural proxies, the trade-off between efficiency and learner control, and the redistribution of agency through AI-mediated systems. They are careful to say the framework does not argue for more or less automation. It supports analysis of when AI scaffolds learners' capacity to act and when it substitutes for it.

Run Alpha through that grid and the picture is stark, because one of the four parties has been struck out.

Decisions about what a child studies, in what sequence, at what grain size and against what standard of mastery are made by the platform, a product of the institution, a set of interlocking companies owned by the same people. The learner's authority is real but narrow: they control pace within a path they did not choose, and the pace is policed. The educator's authority is nil, not partially delegated but structurally absent, because the guide does not touch curriculum and is not present during instruction. Where a conventional school distributes authority across four parties with different incentives and different information, Alpha collapses three into one vendor and deletes the fourth.

The evidence question is worse. What supports the platform's decisions about a particular child on a particular morning? Behavioural telemetry: keystrokes, mouse movement, time on task, screen patterns, eye tracking. These are precisely the proxies the paper warns about, and they are used not merely to study agency but to exercise it. The system infers understanding from clicking. A teacher infers it from a face.

The time horizon question is the one nobody at a start-up asks. The consequences of a mis-sequenced curriculum, or of an unsolvable question that teaches a nine-year-old that mathematics is arbitrary and that the failure is hers, do not surface within a quarter. They surface years later, in a subject abandoned, in a self-concept quietly revised downward. The framework's insistence on temporal consequence is the part a company measuring growth in MAP percentiles has no instrument to detect.

Nobody Is Left to Perform the Trust Calculus

The second paper used as a spine here needs a correction first. It has circulated as “Teachers' Perceived Benefits and Risks of AI Across Fifty Schools”. That is not its title. The paper at arXiv 2605.08486, submitted on 8 May 2026 by Yan Tao, Olga Viberg, Deepak Varuvel Dennison, Zhikun Wu and René Kizilcec, is “Teachers' Perceived Benefits and Risks of AI Across Fifty-Five Countries: An Audit of LLM Alignment and Steerability”. It studies fifty-five national school systems, not fifty schools. Getting that wrong would be a small error in a piece about a school that generates its own facts, which is reason enough to fix it in public.

Its opening sentence is the one that matters: teachers' trust in artificial intelligence in education depends on how they balance its perceived benefits and risks. The authors use OECD TALIS data across fifty-five countries and territories to measure cross-national variation in how teachers weigh those two things, then benchmark eight state-of-the-art large language models from four providers against that distribution. The models fail. They compress country differences, overestimate both benefits and risks, and show limited gains from identity prompting or enhanced reasoning. The authors' conclusion is a warning against treating model outputs as substitutes for direct engagement with teachers when informing global initiatives.

The finding is about policy consultation. Its relevance to Edmond is structural. The paper describes trust as a calculus, an ongoing act of professional judgment in which a person with domain expertise weighs what a tool gives against what it costs and decides how much authority to concede. That judgment is distributed across millions of classrooms, is sensitive to local context in ways the models cannot reproduce, and constitutes a real, if invisible, safety layer in every school system that has teachers.

At Alpha there is nobody to perform it. The guide is not qualified to evaluate a mathematics sequence and, by design, is not permitted to alter it. The parent has bought the product and cannot inspect it. Oklahoma has no jurisdiction over a private school's instructional method. The only party balancing risk against benefit is the vendor whose revenue depends on the answer, and the paper's finding that models systematically misestimate both sides of that ledger is not reassuring when the model is also the thing being assessed.

The Things a Teacher Does That Do Not Show Up in the Telemetry

Strip the argument down and the question is whether the functions of a teacher are separable, and whether the residue left after separation is worth anything. Alpha's implicit answer is that instruction is content delivery plus assessment, both of which software does faster, and that everything else is motivation, which is what guides are for. The research does not support the first half of that.

Paul Black and Dylan Wiliam's 1998 review, published in Phi Delta Kappan as “Inside the Black Box”, identified more than forty studies showing effect sizes between 0.4 and 0.7 from strengthening formative assessment, with the largest gains among low-achieving students. Formative assessment is not testing. It is the continuous, mostly unrecorded practice of finding out what a learner currently believes and adjusting on the spot. Barak Rosenshine's 2012 synthesis in American Educator made the same point operationally: introduce material in small steps, check for understanding by questioning, provide models and scaffolds, monitor independent practice. Every one of those principles describes an adult who is present and paying attention.

The relational evidence is separate and at least as strong. Debora Roorda, Helma Koomen, Jantine Spilt and Frans Oort published a meta-analysis in the Review of Educational Research in 2011 covering ninety-nine studies, finding medium to large associations between the affective quality of teacher-student relationships and school engagement, and small to medium associations with achievement. John Hattie's synthesis of more than a thousand meta-analyses places those relationships among the higher-impact influences, with formative evaluation and teacher clarity higher still. Hattie's effect sizes have been criticised on methodological grounds and are not precision instruments, but the ranking is consistent with the underlying literature.

W. Ian O'Byrne, an associate professor of literacy education at the College of Charleston who has studied digital literacy for two decades, put the pedagogical objection cleanly in The Conversation on 15 June 2026: effective learning is not always efficient. He drew on Robert Bjork's work showing that the struggle to retrieve an answer before receiving it is what makes the knowledge stick, on Piaget's account of learning through the collision between expectation and outcome, and on Vygotsky's insistence that development is fundamentally social. Efficiency, in learning, is frequently the enemy. A system optimised to remove friction removes the thing that does the work.

This is where the anti-pattern detector reveals itself as a caricature of noticing. It fires when progress stops or screen behaviour looks unusual. It cannot distinguish a child who has given up from one thinking hard about something difficult, one who has spotted an error in the question, or one upset about something that happened at home. A teacher makes that distinction in about two seconds, from a face, and it determines everything they do next. The educational neuroscientist Jared Cooney Horvath, quoted in a National Education Policy Center piece by Emily Cherkin and Denise Champney in February 2026, made the reductio explicit: if learning could truly be accelerated the way the marketing claims, students would finish the whole of school by the second grade.

Surveillance Is Doing the Work That Attention Used to Do

Since there is nobody in the room, the room has to be instrumented, and the instrumentation is the part with the clearest legal history.

Todd Feathers reported for Wired on 27 October 2025, in “Parents Fell in Love With Alpha School's Promise. Then They Wanted Out”, that the platform tracks screen activity, keystrokes, mouse movement and eye position, and uses those signals to flag anti-patterns. Monitoring extended into students' homes by default, and parents who objected had to opt out manually. In the incident that has come to define the story, a student sitting on her bed in pyjamas one evening doing schoolwork was flagged for an anti-pattern and sent a video of herself, captured by her own laptop webcam, as documentation of the violation. Oklahoma Watch summarised the same finding in July 2026, noting that Alpha's programming takes photographs of students when the software detects a lack of progress or unusual screen patterns.

There is settled American precedent for exactly this, and it is not favourable. In February 2010 the family of Blake Robbins, then a fifteen-year-old at Harriton High School, sued the Lower Merion School District in Pennsylvania after discovering that district-issued laptops could photograph students remotely. Discovery established that Robbins had been photographed roughly four hundred times in a fortnight, including while he slept in his bedroom, and that the district had captured thousands of webcam images and screenshots across the student body, defending the practice as a means of locating missing computers. In October 2010 it settled the Robbins and Hasan claims for 610,000 dollars.

The distinction Alpha would draw is consent. Parents agree to the terms; Lower Merion's families did not. That distinction is real and it is thin. The person photographed in her bedroom did not consent, was not the contracting party, and had no means of refusing. It is the same structural gap that runs through every children's technology market: the party who signs is not the party whose bedroom it is.

The legal position is worse than most parents will assume. The Family Educational Rights and Privacy Act applies to institutions receiving funds administered by the US Department of Education. The overwhelming majority of private K-12 schools do not, and sit outside FERPA entirely. The Children's Online Privacy Protection Act reaches operators of services directed to children under thirteen, and the Federal Trade Commission's school-authorisation route lets a school consent on parents' behalf, but only for school-authorised educational purposes and no unrelated commercial use. The Commission declined in its 2025 rulemaking to finalise the proposed ed-tech amendments, deferring to potential FERPA changes, leaving the school-consent question unsettled at precisely the moment it matters. State law has holes in the shape of this product: only Illinois, Texas and Washington have dedicated biometric statutes, only Illinois provides a private right of action, and eye tracking and keystroke dynamics sit awkwardly against definitions written for fingerprints and face scans.

Oklahoma has just legislated on this subject in a way that does not reach Alpha. Governor Kevin Stitt signed Senate Bill 1734, the Oklahoma Responsible Technology in Schools Act, on 12 May 2026, after it passed the Senate forty-two to nil and the House eighty-nine to nil. It prohibits artificial intelligence being used for instructional purposes except through educator-directed use, obliges teachers to review anything AI produces before it is used in the classroom, bars AI as the primary basis for grading, discipline or promotion, requires parental notification and opt-out, and mandates district AI policies. Unanimous in both chambers. Inapplicable to private schools. The state has written down what it believes about machines teaching children, and the school it most obviously describes is exempt.

The Profession Is Being Redefined by the Price Signal

The second half of the question is what this does to teaching as an occupation, and here the answer has less to do with any individual school than with the signal a price sends.

Alpha has performed an unbundling. The teacher's job, historically a single role, has been separated into curriculum design, which belongs to a software vendor; instruction, which belongs to a model; assessment, which belongs to a platform; and motivation and pastoral care, which belong to a guide who is paid extremely well, is deliberately recruited from unconventional backgrounds, requires only a bachelor's degree in any subject, passes a cognitive test and a background check, and trains for several months in Austin. Alpha's own materials describe teachers becoming guides, shifting from traditional roles to mentoring and motivating learners, focusing on emotional growth and life skills.

That is a coherent theory of labour. It is also, precisely, deprofessionalisation: the removal of discretion from a role while retaining responsibility for outcomes. The guide is accountable for a child's engagement and cannot alter anything that produces it. Wired's reporting described guides not permitted to help a stuck student, who instead had to schedule a meeting with an academic specialist, and found that twenty-seven of the thirty-one academic coaches listed for 2023-24 were remote workers in the Philippines and Colombia employed through Liemandt's other ventures rather than credentialed educators.

The context is a profession already under strain. The Learning Policy Institute's 2026 national scan estimated that across the fifty states and the District of Columbia, some 387,843 teachers were not fully certified for their assignments, and that in the twenty-nine states and the District of Columbia publishing vacancy data a further 37,569 positions were unfilled. The combined 425,412 is a floor rather than a total, and already amounts to around one in eight teaching posts. Attrition drives roughly ninety per cent of annual new-teacher demand, and nearly three-quarters of those who leave or move do so voluntarily for reasons other than retirement. Interest in teaching among American undergraduates has fallen to its lowest level in half a century.

Into that vacuum arrives a model paying three times the starting salary for a job requiring no certification. It is not hard to see which way recruitment flows, or what it does to the argument for spending four years and considerable money on a credential the sector's highest payers have declared unnecessary.

This is the point Oklahoma State Representative Cyndi Munson, a Democrat representing part of Oklahoma City, was making. “These types of schools and models that are being implemented are taking away the seriousness of what it takes to be a classroom teacher,” she told Oklahoma Watch. “It is very disturbing and should worry all of us.” Senator Regina Goodwin, a Tulsa Democrat, put concession and objection in the same breath: “I understand we have challenges with our public education. But the answer is not to just eliminate human involvement that is certified.” When Pennsylvania rejected the Unbound Academy application, the state education association's president, Aaron Chapin, said AI can help teachers but can never replace a teacher guiding a student's learning.

The prestige mechanism is what makes this more than a local dispute. American schooling has always taken its cues on quality from the top of the market. When the most expensive schools in the country announce that the certified teacher is an unnecessary cost centre, that judgment does not stay inside the buildings that can afford 40,000 dollars a year. It becomes available to every legislature looking at a shortage of 425,000 positions and a model that appears to solve it. Jami Cole, an Oklahoma educator with the group Edvocates, named the fear directly: that the next move is at public funds. Alpha was briefly listed on the Oklahoma Tax Commission's approved schools list for the Parental Choice Tax Credit and removed after Oklahoma Watch enquired, the company saying it had not applied. In Texas, education savings accounts begin flowing in the 2026-27 year, and accreditation by a body recognised by the Texas Private School Accreditation Commission, which includes Alpha's accreditor Cognia, is one of the gates.

What Cannot Be Run Twice

The defence of all this is that it is an experiment, and that experiments are how anything improves. The difficulty is the object of the experiment.

A software company can ship a broken build, read the telemetry, and ship a fix on Thursday. A child gets one pass at being seven. If the 2.6 multiple is a selection artefact, if the machine-generated curriculum has taught a cohort that questions sometimes have no answers and that the failure is theirs, if the anti-pattern detector has punished the children who think slowly and rewarded the ones who click quickly, the correction arrives after the exposure. There is no control group and no rerun, and the network is scaling from thirteen sites to about forty before the first independent evaluation exists anywhere.

The clearest admission in the whole enterprise is the admissions policy. A school confident that its personalised, mastery-based, infinitely patient tutor represents the future of learning has told applicants in Oklahoma that it cannot serve students with individualised education programmes, 504 plans or behaviour plans, or those needing intensive behavioural, therapeutic or one-to-one support. Those are, disproportionately, the children for whom differentiated instruction was invented and for whom personalisation should mean most. The model excludes them. A claim to have solved education that is conditioned on not admitting the students education finds hardest is not a claim about education at all.

Applying the Agency Allocation Framework honestly gives the sharpest formulation available. Alpha has not redistributed decision authority between learners, educators, institutions and AI systems. It has concentrated it. Curriculum, instruction, assessment, pacing, attention monitoring and the definition of acceptable behaviour now sit inside a single proprietary system owned by the same people who own the school, with the party who existed to hold that system to account removed from the room by contract. The trust calculus that Tao, Viberg, Varuvel Dennison, Wu and Kizilcec describe teachers performing across fifty-five countries is not being performed badly at Alpha. It is not being performed.

What is lost when the person responsible for a child's education is not a human being is not sentiment, and it is not the smell of chalk. It is a specific set of functions with a specific evidence base: the formative judgment that reads a face and revises the next sentence, the relational scaffold that makes a difficult subject survivable, and the professional scepticism that looks at a badly written question and says, ignore that one, it's wrong. That third function is the one Alpha has most conspicuously deleted, and the 404 Media reporting suggests it was the one most needed. A machine that produces unsolvable questions requires an adult who can recognise one.

Since Wednesday morning in Edmond, twenty-six children and in Tulsa thirty-two have been opening laptops. For two hours each day the most consequential decisions about what they learn, how fast, and whether they are trying hard enough are being made by a piece of software, watched by a camera, and reviewed by an adult who is somewhere else. Their parents have paid more for this than for any other school in the state. It is the first time in the history of American private education that the premium product has been defined by the absence of the thing the premium was always for.

References

  1. Maya Henry, “A $40,000-per-Year AI School With No Teachers Is Opening in Oklahoma This August,” Oklahoma Watch, 21 July 2026. https://oklahomawatch.org/2026/07/21/a-40000-per-year-ai-school-with-no-teachers-is-opening-in-oklahoma-this-august/
  2. Emanuel Maiberg, “'Students Are Being Treated Like Guinea Pigs:' Inside an AI-Powered Private School,” 404 Media, 17 February 2026. https://www.404media.co/students-are-being-treated-like-guinea-pigs-inside-an-ai-powered-private-school/
  3. Todd Feathers, “Parents Fell in Love With Alpha School's Promise. Then They Wanted Out,” WIRED, 27 October 2025. https://www.wired.com/story/alpha-school-ai-parents/
  4. W. Ian O'Byrne, “AI schools like Alpha promise efficiency, but can't replicate the messy process that helps kids learn,” The Conversation, 15 June 2026. https://theconversation.com/ai-schools-like-alpha-promise-efficiency-but-cant-replicate-the-messy-process-that-helps-kids-learn-282464
  5. Emily Cherkin and Denise Champney, “The Price Kids Pay: What Alpha School Really Costs,” National Education Policy Center, 23 February 2026. https://nepc.colorado.edu/blog/fprice-kids-pay
  6. Conrad Borchers, Olga Viberg and René F. Kizilcec, “Who Decides in AI-Mediated Learning? The Agency Allocation Framework,” arXiv:2604.13534, 15 April 2026 (v2, 9 May 2026). https://arxiv.org/abs/2604.13534
  7. Yan Tao, Olga Viberg, Deepak Varuvel Dennison, Zhikun Wu and René F. Kizilcec, “Teachers' Perceived Benefits and Risks of AI Across Fifty-Five Countries: An Audit of LLM Alignment and Steerability,” arXiv:2605.08486, 8 May 2026. https://arxiv.org/abs/2605.08486
  8. Benjamin S. Bloom, “The 2 Sigma Problem: The Search for Methods of Group Instruction as Effective as One-to-One Tutoring,” Educational Researcher, June 1984. https://journals.sagepub.com/doi/10.3102/0013189X013006004
  9. Chen-Lin C. Kulik, James A. Kulik and Robert L. Bangert-Drowns, “Effectiveness of Mastery Learning Programs: A Meta-Analysis,” Review of Educational Research, Summer 1990. https://journals.sagepub.com/doi/10.3102/00346543060002265
  10. Kurt VanLehn, “The Relative Effectiveness of Human Tutoring, Intelligent Tutoring Systems, and Other Tutoring Systems,” Educational Psychologist, 2011. https://www.tandfonline.com/doi/abs/10.1080/00461520.2011.611369
  11. Andre Nickow, Philip Oreopoulos and Vincent Quan, “The Impressive Effects of Tutoring on PreK-12 Learning: A Systematic Review and Meta-Analysis of the Experimental Evidence,” NBER Working Paper 27476, 2020. https://www.nber.org/papers/w27476
  12. Paul Black and Dylan Wiliam, “Inside the Black Box: Raising Standards Through Classroom Assessment,” Phi Delta Kappan, October 1998. https://journals.sagepub.com/doi/10.1177/003172171009200119
  13. Barak Rosenshine, “Principles of Instruction: Research-Based Strategies That All Teachers Should Know,” American Educator, Spring 2012. https://www.aft.org/ae/spring2012/rosenshine
  14. Debora L. Roorda, Helma M. Y. Koomen, Jantine L. Spilt and Frans J. Oort, “The Influence of Affective Teacher-Student Relationships on Students' School Engagement and Achievement: A Meta-Analytic Approach,” Review of Educational Research, 2011. https://journals.sagepub.com/doi/10.3102/0034654311421793
  15. “Hattie Ranking: 256 Influences and Effect Sizes Related to Student Achievement,” Visible Learning, accessed 15 August 2026. https://visible-learning.org/hattie-ranking-influences-effect-sizes-learning-achievement/
  16. Carly Sitrin, “Pennsylvania rejects AI cyber charter school,” Chalkbeat Philadelphia, 29 January 2025. https://www.chalkbeat.org/philadelphia/2025/01/29/pennsylvania-rejects-ai-cyber-charter-school-unbound-academy-application/
  17. “New Arizona charter school will use AI in place of human teachers,” KJZZ, 18 December 2024. https://www.kjzz.org/education/2024-12-18/new-arizona-charter-school-will-use-ai-in-place-of-human-teachers
  18. “Lower Merion School District Settles Webcam Spy Lawsuits For $610K,” CBS News Philadelphia, 11 October 2010. https://www.cbsnews.com/philadelphia/news/lower-merion-school-district-settles-webcam-spy-suit/
  19. US Department of Education, “Letter to Parent Regarding Applicability of FERPA to Private Schools,” Student Privacy Policy Office, October 1999. https://studentprivacy.ed.gov/resources/letter-parent-regarding-applicability-ferpa-private-schools-october-1999
  20. Public Interest Privacy Center, “New COPPA Update: A Setback for Schools and Student Privacy?” publicinterestprivacy.org, 2025. https://publicinterestprivacy.org/new-coppa-update/
  21. State of Illinois, “Biometric Information Privacy Act, 740 ILCS 14,” Illinois Compiled Statutes via Justia, accessed 15 August 2026. https://law.justia.com/codes/illinois/chapter-740/act-740-ilcs-14/
  22. State of Texas, “Capture or Use of Biometric Identifier, Business and Commerce Code Section 503.001,” via FindLaw, accessed 15 August 2026. https://codes.findlaw.com/tx/business-and-commerce-code/bus-com-sect-503-001/
  23. Lightspeed Systems, “Oklahoma SB 1734 Is Now Law: What Every District Needs to Do Before 2027-28,” 2026. https://www.lightspeedsystems.com/blog/oklahoma-sb-1734-is-now-law-what-every-district-needs-to-do-before-2027-28/
  24. Oklahoma State Department of Education, “2025-26 State Minimum Salary Schedule,” oklahoma.gov. https://oklahoma.gov/content/dam/ok/en/osde/documents/divisions/school-personnel-records/2025-26%20State%20Minimum%20Salary%20Schedule.pdf
  25. National Education Association, “2024-2025 Teacher Salary Benchmark Report,” nea.org, April 2026. https://www.nea.org/sites/default/files/2026-04/2024-2025-teacher-salary-benchmark-report-final-new.pdf
  26. Learning Policy Institute, “An Overview of Teacher Shortages: 2026,” learningpolicyinstitute.org, 2026. https://learningpolicyinstitute.org/product/state-teacher-shortages-2026-factsheet
  27. Learning Policy Institute, “Teacher Turnover in the United States: Patterns, Drivers, and Policy Strategies,” learningpolicyinstitute.org, accessed 15 August 2026. https://learningpolicyinstitute.org/product/teacher-turnover-united-states-factsheet

Tim Green

Tim Green UK-based Systems Theorist & Independent Technology Writer

Tim explores the intersections of artificial intelligence, decentralised cognition, and posthuman ethics. His work, published at smarterarticles.co.uk, challenges dominant narratives of technological progress while proposing interdisciplinary frameworks for collective intelligence and digital stewardship.

His writing has been featured on Ground News and shared by independent researchers across both academic and technological communities.

ORCID: 0009-0002-0156-9795 Email: tim@smarterarticles.co.uk

Listen to the free weekly SmarterArticles Podcast

Discuss...