Education

From PRIMO.ai
Jump to navigation Jump to search

Youtube search... ...Google search


AI in education covers every way that artificial intelligence is used to help people learn, teach, or run schools: AI tutors that walk a student through a problem, assistants that help teachers plan lessons and give feedback, systems that score essays or flag students who are falling behind, and the curriculum that teaches young people how AI itself works. Since late 2022, generative AI and conversational AI tools such as ChatGPT, Gemini, and Claude have turned what was a niche research field into an everyday reality. By late 2025, more than half of U.S. teens said they used AI chatbots for schoolwork, and about six in ten U.S. public school teachers were using AI in their own work.

Where it came from. The field of AI in education (often called AIED) is much older than chatbots. In 1984, educational psychologist Benjamin Bloom reported that students tutored one-on-one with mastery learning outperformed conventionally taught students by about two standard deviations, the famous "2 sigma problem": one-to-one tutoring works, but no society can afford a tutor for every child. Researchers at Carnegie Mellon University, including John Anderson and Ken Koedinger, spent decades building "intelligent tutoring systems" that modeled students' knowledge step by step; these became commercial products such as Carnegie Learning's Cognitive Tutor. Today the field draws on learning scientists (for example Rose Luckin at UCL, Dora Demszky and Susanna Loeb at Stanford), economists who run large field experiments (Philip Oreopoulos, Hamsa Bastani), nonprofits such as Khan Academy, AI developers such as OpenAI, Google, Anthropic, and Microsoft, and international bodies such as the OECD, UNESCO, the World Bank, and the Brookings Institution.

The problem it addresses. Education has three stubborn bottlenecks. Individual tutoring is the most effective form of instruction but is expensive. Teachers spend large parts of their week on planning, grading, and paperwork instead of working with students. And classrooms move at one pace while students arrive with very different gaps in what they know. AI promises cheap, always-available tutoring, relief from routine teacher work, and instruction matched to each student's level. Generative AI also created a new problem: it can produce a finished essay or a solved worksheet instantly, which challenges how schools assign and assess work.

The central mechanism. Effective AI tutoring is a feedback loop: diagnose what the student knows, give a task at the right level of difficulty, respond to each attempt with hints or questions rather than answers, and repeat until the skill is mastered and retained. Older tutoring systems ran this loop with hand-built rules in narrow subjects. Large language models can run it in natural conversation across almost any subject, but by default they are built to be helpful, which usually means simply giving the answer.

The paradigm shift. The defining finding of 2025 and 2026 research is that the same technology can help or harm learning depending on how it is designed and used. When AI does the thinking for students, their performance on the task improves while their learning can decline. When AI is designed to make students think, several randomized trials have found large gains. That shifts the central question from "should schools use AI?" to "which AI, used how, for what?" It also shifts the goals of schooling toward AI literacy, judgment, and the human skills that AI does not replace.

How it differs from earlier approaches. Earlier education technology (drill software, video lectures, adaptive practice platforms) was mostly scripted and lived inside school-approved tools. Generative AI is general-purpose, conversational, inexpensive, and largely used outside institutional control, on students' own phones. It can explain, question, and adapt in ways older tools could not, but it can also hallucinate, flatter, and do the work for the student.


AI can act as a tireless tutor, teaching assistant, and learning coach, but whether it helps students learn depends far more on design and use than on raw model power. Used to hand out answers, it raises performance while eroding learning; used to make students think, it has produced some of the largest gains education research has seen. The technology is now in nearly every student's pocket, so the burden falls on how schools, families, and tool builders shape its use.


How AI Is Changing Teaching and Learning

The Mechanism

AI touches education through several distinct channels. They work differently and carry different risks, so it helps to separate them.

  1. Student-facing tutors and practice. Adaptive practice platforms and AI tutors work on two loops. The inner loop responds to each step a student takes, with a hint, a probing question, or a worked example. The outer loop chooses what the student should practice next, based on a running estimate of what they have mastered. Generative AI mainly improves the inner loop, making feedback conversational and flexible. Education-specific modes now try to prevent answer-giving: ChatGPT's Study Mode (July 2025), Gemini's Guided Learning (August 2025), and Claude's Learning Mode (April 2025) all use Socratic questioning instead of direct answers, and Khanmigo was designed from the start to coach rather than solve.
  2. Teacher-facing assistants. AI drafts lesson plans, worksheets, differentiated reading passages, quizzes, rubrics, parent emails, and feedback on student work. This is where adoption is fastest and evidence of benefit (time saved) is clearest. Related tools include automated scoring of essays and short answers and automated feedback on writing and code.
  3. Human-AI hybrids. Some of the strongest results come from AI that coaches the human tutor or teacher instead of replacing them. Stanford's Tutor CoPilot, for example, suggests expert-style responses to human tutors during live chat sessions; the tutor decides what to send.
  4. Assessment and integrity. Schools use AI to generate and grade assessments, and some try to detect AI-written work. Detection has proved unreliable (see below), which is pushing assessment toward in-class work, oral defense, and grading the process rather than only the product.
  5. System-level analytics. Early-warning systems, scheduling, and enrollment tools use machine learning to predict which students need help. These raise fairness and privacy questions because errors fall on individual children.
  6. AI as a subject. Schools increasingly teach how AI works, how to use it responsibly, and how to evaluate its output (see the K12 section).

What makes a tutor work. The Harvard physics study described below gives a clear recipe. Its tutor was not a stock chatbot: the instructors supplied their own worked solutions, told the model to guide rather than tell, broke problems into steps to manage cognitive load, and asked students to explain their reasoning. Research consistently points to the same design features: the AI knows the correct solution path, withholds the final answer, asks students to generate and explain, gives specific feedback on errors, and connects to what the student has already practiced. Khan Academy reported in 2026 that giving Khanmigo structured access to each student's recent practice history measurably improved its tutoring.

Dependencies and conditions. Results depend on factors that have little to do with model quality:

  • Engagement. Students must actually use the tutor, and use it to think. The largest 2026 trial found that most students simply did not.
  • Teacher integration. Programs embedded in class time with teacher guidance, such as the World Bank's Nigeria pilot, outperform tools left for students to discover on their own.
  • Guardrails. Without them, students use AI as a crutch.
  • Infrastructure and equity. Devices, connectivity, and trained teachers are unevenly distributed.
  • Data governance. Tutoring logs, keystrokes, and sometimes video are sensitive data about children.

Where uncertainty remains. Most rigorous studies last weeks or a few months. We do not yet know the long-term effects of growing up with AI assistance on writing ability, persistence, critical thinking, or social development, and effects differ widely across students, subjects, and settings.

Why This Matters

Scientific importance. AI tutoring is the first scalable way to test long-standing learning-science ideas (mastery, spacing, Socratic questioning) with millions of students and fine-grained data. It also gives researchers a new natural experiment in cognitive offloading, a question that matters far beyond school.

Practical consequences for students. Well-designed AI tutoring can deliver large, cheap gains: a World Bank pilot in Nigeria produced about 0.3 SD in six weeks, ranking among the most cost-effective education interventions studied in low- and middle-income countries. Tutor CoPilot raised mastery at a cost of about $20 per tutor per year. Poorly designed or unsupervised use can quietly hollow out skills while grades look fine.

Practical consequences for teachers. Teachers who use AI weekly report saving about 5.9 hours a week, roughly six weeks over a school year, time they say goes into individualized instruction, feedback, and parent communication. But guidance lags far behind use: in 2025 only 19 percent of U.S. teachers said their school had an AI policy, and a 2026 follow-up found only 18 percent had received formal guidance from administrators. Teachers' unions are now negotiating how AI enters classrooms; the American Federation of Teachers launched a $23 million National Academy for AI Instruction in July 2025 with funding from Microsoft, OpenAI, and Anthropic.

Assessment and credentials. If AI can produce a passable essay in seconds, take-home essays and problem sets no longer reliably measure what a student can do alone. Schools are shifting toward in-class writing, oral examinations, project work with documented process, and assessments that ask students to critique or improve AI output.

Equity. AI could narrow gaps by giving every student access to tutoring that wealthy families already buy, and the Nigeria pilot found especially large gains for girls. It could also widen gaps: students with strong foundational skills and supportive adults tend to use AI to extend their thinking, while others use it to avoid thinking. Access to devices, connectivity, and teacher training is uneven.

Child development and wellbeing. Twelve percent of U.S. teens told Pew they had used chatbots for emotional support or advice. The Brookings Global Task Force concluded in January 2026 that, at this point, the risks of generative AI for children's foundational cognitive and social development overshadow its benefits, while stressing that the trajectory can still change.

What would change if the optimistic case proves right. If guarded AI tutoring reliably delivers effects of 0.3 SD or more at scale, it would be among the most cost-effective tools education has ever had, and the human role in school would tilt toward motivation, relationships, discussion, and judgment. Models such as Alpha School (below) are betting on that future.

What remains uncertain. Long-term effects, effects for younger children, how to sustain engagement without coercion, how to assess learning when AI is everywhere, and how to protect children's data and attention.

Recent Developments

The past two years produced the first rigorous field evidence on generative AI and learning. The overall picture is consistent: design and use determine the outcome.

Randomized trials: when AI tutoring works

Harvard physics AI tutor (Scientific Reports, June 2025). Gregory Kestin, Kelly Miller, and colleagues ran a randomized crossover trial with 194 students in Harvard's large introductory physics course in fall 2023. Each student learned one topic in an in-class active-learning session (already a research-backed best practice) and another at home with "PS2 Pal," a GPT-4 tutor the instructors built with their own worked solutions and explicit pedagogical rules. Measured against a combined pre-test, median learning gains with the AI tutor were more than twice as large as with the classroom session, students reported higher engagement and motivation, and the median time on task with the tutor was about 49 minutes. What it establishes: a carefully designed AI tutor can beat a strong classroom method on short-term learning of well-defined content. What it does not establish: results for younger students, weaker or less motivated learners, open-ended subjects, or long-term retention. It was one course, two topics, at a highly selective university.

World Bank pilot in Nigeria (2025). Martín De Simone and colleagues evaluated a six-week after-school program in Edo State in June and July 2024, in which first-year senior secondary students used Microsoft Copilot (powered by GPT-4) twice a week for English, with teachers guiding the sessions. In a randomized evaluation, participants scored 0.31 SD higher on an assessment of English, AI knowledge, and digital skills, and 0.23 SD higher in English alone, equivalent by the authors' estimate to 1.5 to 2 years of typical schooling. Each additional day attended added roughly 0.03 SD, girls gained the most, and participants also did better on regular end-of-year exams. Limitations: a working paper rather than a peer-reviewed article, short duration, an outcome measure that partly tested AI and digital skills taught in the program, and attendance disrupted by flooding and strikes.

Tutor CoPilot (Stanford, 2024 to 2025). Rose Wang, Ana Ribeiro, Carly Robinson, Susanna Loeb, and Dorottya Demszky ran the first preregistered randomized trial of a human-AI system in live tutoring, with about 900 tutors and 1,800 K-12 students in Title I schools. The AI suggested expert-style responses that tutors could use, edit, or ignore. Students whose tutors had access were 4 percentage points more likely to master math topics (66 versus 62 percent), and students of the lowest-rated tutors gained 9 points. Tutors using the tool asked more probing questions and gave less generic praise. Cost was about $20 per tutor per year. Tutors noted that some suggestions were not grade-appropriate.

Randomized trials: when AI harms or adds little

Generative AI without guardrails (PNAS, June 2025). Hamsa Bastani, Osbert Bastani, Alp Sungu, and colleagues gave nearly 1,000 high school math students in Turkey access to either a standard GPT-4 chat interface ("GPT Base"), a version with prompts designed to protect learning ("GPT Tutor"), or no AI. During practice, GPT Base students did 48 percent better and GPT Tutor students 127 percent better than controls. On the subsequent exam without AI, GPT Base students did 17 percent worse than students who never had access, while GPT Tutor students did about the same as controls. Chat logs showed students using the unguarded tool as a crutch and copying its occasional errors. Significance: this is the clearest experimental demonstration that AI can raise performance while lowering learning, and that simple guardrails largely prevent the harm. It does not show that guardrailed AI produces large gains.

Khanmigo two-year trial (NBER, August 2026). Philip Oreopoulos and Nina Low ran a cluster randomized trial in 18 Tennessee middle schools from 2024 to 2026, assigning students who were at least a grade behind to use Khan Academy with Khanmigo, configured to coach rather than answer, during daily remedial math time. Assignment raised achievement by about 1.3 national percentile ranks per term, or 0.06 to 0.08 SD per year, with an implied 0.14 SD for a full year of active use. Those gains resemble what Khan Academy practice produces without AI. The reason appears to be engagement: 96 percent of students tried Khanmigo, but the median student messaged it on only a third of practice days and in only 17 percent of sessions where they made a mistake, and most messages were bare answers or clicks on suggested prompts. Sal Khan has acknowledged that for many students the tutor "was a non-event," and Khan Academy has since built Khanmigo directly into practice sessions and plans to reward its use after mistakes. Significance: access is not use. A well-designed tutor added little when help-seeking was optional.

Laboratory evidence and debate: cognitive debt

In June 2025 Nataliya Kosmyna and colleagues at the MIT Media Lab posted "Your Brain on ChatGPT," in which 54 university students wrote essays using ChatGPT, a search engine, or nothing, while wearing EEG caps; 18 returned for a fourth session with conditions swapped. The ChatGPT group showed the weakest neural connectivity, remembered least of what they had written, and felt the least ownership; students who switched from ChatGPT to writing unaided remained under-engaged. The authors call this "cognitive debt." The paper was released before peer review, which the lead author said was deliberate given how quickly schools were adopting AI. A December 2025 published comment urged more conservative interpretation of several results. Status: suggestive and widely discussed, but small, laboratory-based, limited to one task, and not yet independently replicated. It is consistent with, but much weaker evidence than, the field trials above.

Major syntheses

OECD Digital Education Outlook 2026 (January 2026). The OECD's review concluded that generative AI can support learning when guided by clear teaching principles, but that successfully completing a task with AI does not automatically produce learning, and that outsourcing tasks to general-purpose tools raises performance without real learning gains. It reported that 37 percent of lower secondary teachers used AI for their job in 2024, 57 percent agreed AI helps them write or improve lesson plans, and 72 percent believed AI can harm academic integrity. It urged governments to fund education-specific tools rather than relying on general chatbots.

Brookings Global Task Force (January 2026). After consulting more than 500 students, teachers, parents, and experts in 50 countries, reviewing over 400 studies, and running a Delphi panel, Rebecca Winthrop and colleagues released "A New Direction for Students in an AI World: Prosper, Prepare, Protect." Its headline conclusion: at this point in its trajectory, the risks of generative AI in children's education overshadow its benefits, because the risks threaten the foundational cognitive and social development that students need to benefit at all. Its 12 recommendations include using tools that "teach, not tell," co-designing tools with educators and families, and moving schoolwork away from transactional tasks that AI can complete.

Adoption and practice

  • Students. Pew Research Center's survey of 1,458 U.S. teens in fall 2025 (published February 2026) found that 64 percent use AI chatbots, about three in ten daily, and 54 percent use them for schoolwork. One in ten said they do all or most of their schoolwork with chatbot help, and 59 percent said students at their school use chatbots to cheat at least somewhat often. Pew's earlier surveys found ChatGPT use for schoolwork rising from 13 percent of teens in 2023 to 26 percent in 2024.
  • Teachers. A Gallup and Walton Family Foundation survey of 2,232 U.S. public school teachers found that 60 percent used AI during the 2024-25 school year; weekly users saved about 5.9 hours a week, and schools with an AI policy saw a 26 percent larger time dividend. A May 2026 follow-up with 2,069 teachers found that six in ten use AI for work but most received no formal guidance on tasks such as grading, feedback, and tutoring.
  • Products. OpenAI launched Study Mode (July 2025) and a free ChatGPT for Teachers program for verified U.S. K-12 teachers (November 2025); Google launched Gemini Guided Learning (August 2025) and offers Gemini free in Workspace for Education; Anthropic launched Claude for Education with a Learning Mode (April 2025), focused on universities. Tutoring companies such as Varsity Tutors moved toward "Live + AI" hybrids. Evidence on these general-purpose study modes is still emerging, and students can switch them off.

Policy and regulation

  • United States. Executive Order 14277, "Advancing Artificial Intelligence Education for American Youth" (April 23, 2025), created a White House Task Force on AI Education, launched a Presidential AI Challenge for students and educators, and directed agencies to prioritize AI in teacher-training grants and to build K-12 AI education resources through public-private partnerships.
  • China. The Ministry of Education issued guidelines on AI education in primary and secondary schools in 2024 and 2025, and Beijing made AI instruction compulsory from fall 2025, requiring at least eight class hours a year across more than 1,400 schools. Chinese guidance also tells teachers not to use generative AI as a substitute for their core teaching role.
  • European Union. The AI Act classifies AI used to determine access to education, evaluate learning outcomes, or monitor students during tests as high-risk, and has banned emotion recognition in educational institutions (except for medical or safety reasons) since February 2025. The Digital Omnibus, which entered into force in July 2026, pushed the high-risk obligations for these systems from August 2, 2026 to December 2, 2027.
  • AI literacy frameworks. In June 2026 the European Commission and the OECD published the final AI Literacy Framework for primary and secondary education, "Empowering Learners for the Age of AI," which organizes AI literacy into four domains (engaging with, creating with, managing, and shaping AI) and will inform the first PISA assessment of Media and AI Literacy in 2029 (see K12).

Academic integrity and detection

AI-text detectors have not lived up to their promise. OpenAI withdrew its own classifier in July 2023 after it identified only 26 percent of AI-written text while flagging 9 percent of human writing. A 2023 Stanford study (Liang and colleagues, published in Patterns) found seven detectors flagged 61 percent of essays by non-native English writers as AI-generated, versus near zero for native speakers, because simpler vocabulary looks "machine-like" to these tools. Several universities, including the University of Waterloo in September 2025, have turned off AI detection, and others now treat a detector score as a reason for conversation, never as evidence on its own. The practical shift is toward assessment design: in-class work, drafts and version history, oral explanation, and assignments that require personal context or critique of AI output.

Future of Learning

Youtube search... ...Google search

  • Code Royale ...world's 1st educational eSport.Learn Machine Learning.

The talks below span 2017 to 2023. The earlier ones describe the promise of intelligent tutoring and data-driven personalization before generative AI arrived; read them alongside the Recent Developments above, which show which predictions held up. Game-based learning and immersive reality remain active areas, though rigorous evidence on VR and AR learning gains is still thinner than for tutoring.

Load video
YouTube
Reimagining the future of Learning with the use of technology and AI
This conference, in English, is part of the "CLO Forum - Important Trends". Artificial intelligence (AI) and technologies such as virtual reality (VR), augmented reality (AR), and holograms can help change the world of learning. To make the most of these technologies, we need to reimagine what learning experiences look like. In this session, I’ll describe a future vision of learning that merges progressive learning theory with new technologies. This conference, on February 1st in the London room, was presented by David Guralnick, President and CEO, Kaleidoscope Learning, Columbia University, & the International E-Learning Association.

Load video
YouTube
Revolutionizing Education with ChatGPT: How AI is Transforming the Way We Learn
As Artificial Intelligence (AI) improves - the line between human and machine generated content begins to blur. How will we prevent students from cheating on assignments by having an AI such as ChatGPT complete assignment for them?

The Learning Mindset Podcast: https://thelearningmindset.buzzsprout... Join me on my podcast where I share ideas on how we can take a science-based approach to learning. On the podcast, I share ideas, book reviews, best practices, and more about learning.

Load video
YouTube
Artificial Intelligence and the future of learning - Professor Rose Luckin - LHL - 16/05/2017
UCL Knowledge Lab, UCL Institute of Education Tuesday 16th May 2017 What makes a good teacher and a successful lesson? Artificial Intelligence (AI) in the classroom is the key that is unlocking the answer, according to Rose Luckin, Professor of Learning with Digital Technologies. Intelligent software holds opportunities for individualised teaching that understands when and how lessons veer off topic, at what point pupils lose interest and when they are at their most engaged and receptive, providing smart interventions that help teachers find the best learning strategies for their classes. In this UCL Lunch Hour Lecture, Professor Luckin will discuss what AI is telling us about how we learn and predict how machines and humans will interact in the classrooms of the future.

Load video
YouTube
Neuroscience, AI and the Future of Education | Scott Bolland | TEDxSouthBank
Currently around 63% of students are disengaged at school, meaning that they withdrawal either physically or mentally before they have mastered the skills that are required to flourish in later life. In this talk Scott Bolland explores the science of learning, the mismatch between how we teach and how the brain natural learns, and the important role that artificial intelligence could take in addressing the limitations in our current education system. Dr Scott Bolland is the founder of New Dawn Technologies, a high-tech software company aiming to revolutionise education through the use of artificial intelligence. He has spent the last 20 years actively researching and teaching in the field of cognitive science – the scientific study of how the mind works – which spans disciplines such as psychology, philosophy, neuroscience, artificial intelligence and computer science. He holds a PhD in this field, as well as a university medal for outstanding academic scholarship. This talk was given at a TEDx event using the TED conference format but independently organized by a local community. Learn more at https://ted.com/tedx

Load video
YouTube
Learn Anything FASTER With ChatGPT (13 ChatGPT Prompts For Studying)
ChatGPT is incredibly powerful but most people have only just scratched the surface. ChatGPT has some amazing applications for students studying that will make you more productive when it comes to preparing for exams or getting coursework completed. If you want to get the most out of ChatGPT and create original and impactful content you'll want to check out these prompts for learners. So here are 13 ChatGPT prompts for studying so you can study smarter, not harder.

13 ChatGPT Prompts For Studying - Get The Studying Prompts here

Load video
YouTube
[Highlights] Death of Classroom: AI, Online Learning & the Future of Education in Asia
New technologies from artificial intelligence and virtual reality to massive open online courses are beginning to disrupt existing models of learning. These technologies could help to meet rising demands for education around emerging Asia. From the rise of new global educational providers to the development of "micro" degrees and new VR tools for online learning, these technologies will also have profound implications for the way in which Higher Education policies are developed across Asia - both in developed economies like Singapore seeking to re-equip their workforce for a new era of globalisation and emerging nations attempting rapidly to increase the scale and quality of their education systems. Find out key highlights shared by the speakers on their views of the death of traditional classrooms. The panelists are: Mr Johannes Heinlein, Vice President, EdX; Mr Adrian Lim, Director (Digital Participation & Foresight, Digital Readiness Cluster), Info-communications Media Development Authority (IMDA); and Dr Suzaina Kadir, Associate Dean (Admissions, Partnerships and Programmes), Deputy Director (Academic Affairs) and Senior Lecturer, Lee Kuan Yew School of Public Policy. The session is moderated by Mr James Crabtree, Senior Visiting Fellow, the Centre on Asia and Globalisation, Lee Kuan Yew School of Public Policy.

Build a Six Week Custom Course

Sandeep Swadia outlines an "ALTER" framework—Advisor, Librarian, Tutor, Editor, and Roommate—to transform artificial intelligence into a personal education system. Learn how to leverage these five specific roles to build custom curricula, automate study workflows with external tools, and apply analytical rigor to master any subject more effectively. You'll learn how to:

  • Make AI your academic advisor and build a curriculum designed entirely around you
  • Use an AI librarian to separate the handful of sources worth your time from the thousands that aren't
  • Turn AI into a midnight tutor that finds the exact gap in your understanding
  • Let an AI editor pressure-test your work instead of just praising it
  • Give AI the role of a roommate who drags you into worlds you'd never explore on your own

Load video
YouTube

Alpha School

  • Alpha ... A school where kids crush academics in 2 hours, build life skills through workshops, and thrive beyond the classroom.

Alpha School is a private K-12 network, founded in Austin, Texas, built around one bold promise: students finish their core academics in about two hours a day, then spend the rest of the day on life skills, projects, sports, and passions. It is one of the most visible attempts to rebuild school around adaptive software and learning science, and it has become a lightning rod in the debate over AI in education.

MacKenzie Price and Brian Holtz founded the school in 2014 as Emergent Academy, originally a spinoff of the Acton Academy microschool model. It was later renamed Alpha. Price says the idea started when her own daughter complained that school was boring. Software billionaire Joe Liemandt, founder of Trilogy Software and ESW Capital, sent his daughters to the school, later became its principal and "product guy," and says he has committed about $1 billion of his own money to building the software behind it. That software, now called Timeback, is meant to scale Alpha's model far beyond its own campuses; Liemandt's stated goal is a billion children.

Alpha makes three commitments to students: they will love school, they will learn twice as much in two hours a day, and they will develop life skills that last a lifetime. Adults on campus are called "guides," not teachers. Alpha says guides do no academic instruction at all; software delivers the lessons, and guides focus on motivation, mentorship, and the afternoon workshops.

What makes Alpha different from earlier "personalized learning" efforts is the combination it bets on: mastery-based software that teaches each child at their actual knowledge level, aggressive motivation systems (including cash rewards), and a school day redesigned so that finishing academics efficiently earns students their afternoons back. Supporters see a working prototype of AI-era schooling. Critics see unverified results, a selective and expensive student body, heavy student monitoring, and a business structure that raises conflict-of-interest questions. Both views are worth understanding.


Alpha School's central bet is that mastery-based, software-delivered academics plus strong motivation can compress core learning into about two hours a day, freeing the rest of childhood for life skills and passions. Its reported results are striking, but they rest largely on the school's own analyses and have not yet been independently verified.


Here's How I'm Fixing School

Load video
YouTube

Kids Will Love School

Load video
YouTube

How the Model Works

Alpha calls its instructional approach 2 Hour Learning. A typical day starts with a morning launch, followed by a focused academic block on laptops, then lunch and afternoon workshops. CNN described the academic block as four 30-minute sessions in math, science, social studies, and language, plus about 20 minutes for skills such as test taking. Alpha's own materials describe 25-minute, Pomodoro-style subject sessions.

The academic loop. The core mechanism is a cycle that repeats for every student, every day:

  1. Diagnose. Students take standardized assessments, mainly NWEA's MAP Growth test, to find their "knowledge grade" in each subject, which may differ sharply from their age grade. Liemandt says students transferring in from other schools, including expensive private schools, averaged about 2.2 grade levels behind on Alpha's diagnostics.
  2. Target. The software assigns lessons at the student's actual level and fills specific gaps, a process Alpha calls "hole filling." Liemandt claims a motivated student can master a full grade level of one subject in roughly 20 to 30 hours.
  3. Master. Students must show about 90 percent accuracy before moving on, which is much stricter than a typical passing grade of 70 percent.
  4. Retain. Spaced repetition brings material back just before the student is likely to forget it.
  5. Earn time back. When the software shows a student is on pace for the day, academics are done and the afternoon opens up.

What "AI" means here. Despite the "AI school" label, the founders say Alpha's AI is mostly adaptive learning software, similar to IXL or Khan Academy, plus its own Timeback platform, rather than chatbots built on large language models. Price calls general chatbots "cheat bots" and says they are kept out of core instruction. Alpha uses generative AI behind the scenes to produce personalized lessons, analyze how students learn, and coach them on study habits. Students have used both third-party apps and Alpha's own tools, and Timeback is meant to tie them together.

Learning-science foundations. Alpha frames its design as an implementation of long-established research: Benjamin Bloom's 1984 "2 sigma problem," which reported that one-on-one tutoring combined with mastery learning produced dramatic gains; cognitive load theory and worked examples; direct instruction; spaced repetition; and fluency in basic facts to free up working memory. These ideas have real research support (see The Physics: The Science of Learning). One caution: later research suggests Bloom's two-sigma figure came from small, short studies, and that typical tutoring effects are substantially smaller, though still meaningful.

Motivation. Liemandt says motivation is "90 percent" of the problem. The main motivator is time back, meaning afternoons spent on activities students choose. Alpha also uses points, leaderboards, an internal currency called "Alpha Bucks," and cash. Under its "100 for 100" program, students earn $100 for scoring 100 on a grade-level Texas STAAR test, at any grade level, which encourages them to go back and fill gaps. Liemandt has also described paying students $1,000 to reach the top 1 percent nationally. Alpha argues that incentives break psychological blocks and then build intrinsic motivation; critics worry about pressure and about rewards crowding out love of learning. The 2026 Khanmigo trial described above suggests why Alpha leans so hard on motivation: even good tutoring software produces little when students choose not to engage.

Life skills. Afternoon workshops cover leadership, teamwork, public speaking, financial literacy, entrepreneurship, and grit. Alpha's examples include students running an Airbnb or a food truck, kindergartners climbing a 40-foot rock wall, and younger students training for a 5K. High school students pursue multi-year passion projects, sometimes called "masterpieces." Workshops vary by campus: some focus on sports, others on robotics, coding, or writing. Alpha cites psychologist David Yeager's "high standards, high support" framework in training its guides.

Guides and staffing. Liemandt says guides earn at least $100,000 a year and that about 100,000 people applied for guide roles in one year; these are company figures. Reporting has also found remote staff involved in supporting students. WIRED reported that 27 of the 31 academic coaches listed for the 2023-24 school year were remote workers in the Philippines and Colombia employed through other ventures connected to Liemandt.

Monitoring. Alpha's software tracks engagement closely, including screen activity and, in some configurations, webcam video, and it uses that data to coach students and flag inefficient study habits. The Unbound Academy charter application described AI that analyzes response accuracy, engagement duration, and "emotional feedback via webcam." This monitoring is central to the model's claims, and it is also central to privacy criticism (see Evidence and Criticism). In the European Union, AI-based emotion recognition in schools has been prohibited under the AI Act since February 2025.

Why It Matters

Alpha is the most prominent real-world test of whether software can take over core academic instruction while humans focus on motivation and development. If its results hold up under independent evaluation, the implications are large: a much shorter academic day, a new role for teachers as coaches and mentors, faster catch-up for students who are behind, and a template that could spread through charters, vouchers, homeschooling, and Timeback-powered startups. If the results mostly reflect who enrolls, Alpha would still be an expensive, well-run private school with an unusual schedule, but not proof of a new educational paradigm. The model also forces broader questions that apply to all AI in schools: how much monitoring of children is acceptable, who owns and protects their data, and how families can evaluate learning claims made by companies that also sell the software.


Load video
YouTube

How to Accelerate Learning and Improve Education | Joe Liemandt (Huberman Lab, August 31, 2026, about 2.5 hours). Neuroscientist Andrew Huberman interviews Alpha principal Joe Liemandt about mastery, working memory and fluency, worked examples, spaced repetition, the "100 for 100" incentive program, and Alpha's sports academies and expansion plans. It is the most detailed explanation of the learning-science rationale behind 2 Hour Learning, though the performance figures cited are Alpha's own and have not been independently verified. Watch on YouTube

Load video
YouTube

Joe Liemandt: Building Alpha School, and The Future of Education (Invest Like the Best with Patrick O'Shaughnessy, Episode 439, August 2025, about 2.5 hours). Liemandt explains why he believes the classroom model is obsolete, how generative AI produces personalized lessons and coaching, why "time back" is the main motivator, how guides are paid and trained, and his plan to release the Timeback platform to other school builders. It is a useful companion to the Colossus profile published the same month and gives the clearest view of Alpha's long-term business strategy. Watch on YouTube

K12

Youtube search... ...Google search

CalypsoSoftware_595x638_f8f93ea0-3672-43b4-bbdc-3467fc617cc0_660x440.png

K-12 AI education has two jobs: using AI to help children learn, and teaching children about AI. The second job has moved from enthusiast projects to national policy.

What students should know. The AI4K12 Initiative, a joint project of the Association for the Advancement of Artificial Intelligence (AAAI) and the Computer Science Teachers Association (CSTA) led by Carnegie Mellon's David Touretzky, organizes K-12 AI learning around "Five Big Ideas": perception (computers perceive the world through sensors), representation and reasoning (agents maintain models of the world and reason with them), learning (computers can learn from data), natural interaction (making agents interact naturally with humans is hard), and societal impact (AI can affect society in both positive and negative ways). Its guidelines map each idea across four grade bands (K-2, 3-5, 6-8, 9-12). Touretzky has noted that large language models forced an update to the natural-interaction idea, since modern systems handle metaphor and nuance far better than earlier ones.

From frameworks to assessment. In June 2026 the European Commission and the OECD, with support from CodeAI (the organization formerly known as Code.org) and international experts, published "Empowering Learners for the Age of AI," a final AI Literacy Framework for primary and secondary education. It defines AI literacy across four domains (engaging with AI, creating with AI, managing AI, and shaping AI), treats deciding not to use AI as a legitimate and competent choice, and says AI literacy cannot be left to computing teachers alone. It will inform PISA 2029, the first international assessment of Media and AI Literacy. UNESCO published separate AI competency frameworks for students and for teachers in 2024.

National programs. In China, Beijing made AI education compulsory from fall 2025, with at least eight class hours a year across all grades in more than 1,400 schools, often folded into information technology classes. In the United States, Executive Order 14277 (April 2025) created a Presidential AI Challenge and directed federal agencies to support K-12 AI education and teacher training. Hands-on tools such as Cognimates, ReadyAI's curricula, and robot-programming kits remain useful for younger students because they make abstract ideas such as training data and bias concrete.

Using AI with children. The research summarized above suggests a few practical rules for K-12 settings: use tools designed to teach rather than answer; keep AI use inside structured time with an adult who knows the goal; check AI output together, treating errors as teaching moments; protect foundational skills (reading, writing, arithmetic fluency) with plenty of practice done without AI; and review a tool's data practices before students use it.

AI for Kids

Children now meet AI as toys, voice assistants, tutoring apps, and chatbot companions. Social robots such as the one described below were designed to combine natural language processing, memory, planning, and reasoning; today, general chatbots on phones and tablets reach far more children than robots ever did. That raises new questions. Twelve percent of U.S. teens told Pew in 2025 that they had used chatbots for emotional support or advice, and the Brookings Global Task Force flagged risks to social and emotional development from AI companions. Parents and educators should look for age-appropriate design, clear data practices, and tools that encourage children to think, create, and connect with people rather than replace those experiences. Embodied AI and robotics kits remain a strong way to teach how AI perceives and acts in the physical world.

Load video
YouTube
Building an AI Robot for Kids featuring Dr Wilfred Odoyo
Dr Wilfred Odoyo is the current director of global AI business at AI Brain, a company based in Korea, with the goal of building fully autonomous AI by unifying the three essential aspects of intelligence, namely Problem Solving, Learning and Memory. It was selected as one of Top 20 Artificial Intelligence Companies in the world by Datamation on September 20, 2016, alongside tech giants such as Amazon and Google. He holds a Ph.D. in Computer Engineering, professing his expertise on Computer Vision, Artificial Intelligence, and Pattern Recognition with interest in Biometric Systems and Robotics. He studied at the prestigious Chosun University in South Korea, where he had been a student and researcher between 2002 to 2012, having pursued his high school studies at St. Joseph's Rapogi.He is also the chair of the Kenyan Professionals and Business Associations of Kenyans in South Korea, which has a membership of about 300 Kenyans. In 2015, Dr Odoyo led a team in the development of Tyche (https://tyche.club/). Tyche, named after the Greek goddess of fortune, (pronounced “Tai-kee”) is an AI robot companion for kids that talks, listens, thinks and reasons to plan and solve tasks. Tyche fuses natural language processing, memory, planning, and reasoning to converse with people. During this event , Dr Odoyo will share his experience while leading the team developing Tyche, his career journey and advice to AI practitioners and enthusiasts. The main presentation from this event can be accessed on https://bit.ly/3gU0ST7

Further Viewing and Sources

Viewing and Reading

Additional Viewing

Load video
YouTube

A new direction for AI and students: Findings from the Brookings Global Task Force on AI and Education (Brookings Institution, January 14, 2026, about one hour). Rebecca Winthrop, director of the Center for Universal Education, and University of Virginia student and co-author Rida Karim discuss the task force's report with journalist Kara Swisher. They explain why the task force concluded that generative AI's current risks to children's cognitive and social development overshadow its benefits, and what "teach, not tell" tools and other recommendations look like in practice. Watch on YouTube

Load video
YouTube

AI and the future of teaching and learning (Brookings Institution, June 29, 2026, about one hour). Andreas Schleicher, the OECD's director for education and skills, and Rebecca Winthrop of Brookings, moderated by Michael B. Horn of the Clayton Christensen Institute, compare the findings of the OECD Digital Education Outlook 2026 and the Brookings task force report. The conversation focuses on when narrowly designed AI tools help students and teachers, why widespread use of general-purpose chatbots can diminish learning, and what policymakers should do next. Watch on YouTube

Key Sources

Experimental evidence on AI and learning

Syntheses, surveys, and adoption

Policy, literacy, and integrity

Retrieved from "https://primo.ai/w/index.php?title=Education&oldid=53439"