CtrlK
AI in education refers to the use of artificial intelligence technologies to support teaching, learning, assessment, and administrative processes across educational settings. Applications range from intelligent tutoring systems and automated grading to personalized learning platforms and AI-powered administrative tools. The field traces its roots to the 1960s and 1970s, with early computer-assisted instruction systems like PLATO, and has expanded rapidly since the release of large language models such as ChatGPT in late 2022. As of 2026, AI tools are embedded in classrooms from primary schools through universities, reshaping how students learn and how educators teach, while also raising serious questions about academic integrity, equity, and data privacy.
The earliest attempts to use computers for education date to the late 1950s and 1960s. The most significant of these was PLATO (Programmed Logic for Automated Teaching Operations), created in 1960 by Donald L. Bitzer at the University of Illinois at Urbana-Champaign. PLATO was the first generalized computer-assisted instruction system, allowing educators to create interactive lessons that students could work through at their own pace on networked terminals [1].
PLATO was remarkably ahead of its time. The system introduced several innovations that would not become mainstream for decades, including a plasma display, touch screen input, graphical user interfaces, message boards, chat rooms, instant messaging, and even real-time multiplayer games. The courseware was written in TUTOR, a programming language conceived in 1967 by biology graduate student Paul Tenczar that allowed non-programmers to design lesson modules [1].
By the mid-1970s, PLATO IV supported over 900 terminals at 146 sites across the United States and Canada. Control Data Corporation (CDC) licensed the system from UIUC and began commercializing it, and by the mid-1980s over 100 PLATO systems were operating worldwide, mostly at educational institutions [1]. Although PLATO eventually faded from use, its influence on educational technology was substantial: it demonstrated that computers could deliver interactive, individualized instruction at scale.
While PLATO and similar systems delivered pre-programmed instruction, researchers in artificial intelligence pursued a different approach: building systems that could adapt their teaching strategy based on a model of the student's knowledge. These came to be known as Intelligent Tutoring Systems (ITS).
Jaime Carbonell's SCHOLAR system (1970) was among the first to use AI techniques for instruction, employing a semantic network to engage students in Socratic dialogue about South American geography. Other early ITS projects included WHY (Collins and Stevens, 1977), which explored reasoning strategies, and SOPHIE (Brown, Burton, and de Kleer, 1982), which taught electronic troubleshooting [2].
The most commercially successful ITS of this era was Carnegie Learning's Cognitive Tutor, developed in the 1990s based on John Anderson's ACT-R cognitive architecture at Carnegie Mellon University. The Cognitive Tutor for algebra tracked individual student mastery of specific skills and adjusted problem selection accordingly. A large-scale randomized controlled trial published in the Journal of Research on Mathematics Education found that students using Cognitive Tutor outperformed control groups on standardized tests [2].
Despite promising results, ITS adoption remained limited through the 1990s and 2000s due to high development costs, the need for extensive domain modeling, and the difficulty of scaling these systems beyond narrow subject areas.
The 2010s saw the emergence of commercial adaptive learning platforms powered by machine learning rather than hand-crafted expert rules. Companies like Knewton (founded 2008), DreamBox Learning (founded 2006), and Carnegie Learning (which continued to evolve its Cognitive Tutor platform) used data-driven algorithms to personalize content sequencing, pacing, and difficulty.
Knewton, which raised over $180 million in venture funding, partnered with major publishers including Pearson to embed its adaptive engine into digital textbooks. The company claimed its technology could identify precisely which concepts a student had mastered and which needed reinforcement, adjusting the learning path in real time. However, Knewton struggled commercially and was acquired by Wiley in 2019 for a fraction of its peak valuation, illustrating the gap between the promise and the business reality of adaptive learning.
The Massive Open Online Course (MOOC) movement, led by platforms like Coursera and edX (both launched in 2012), also brought AI into education at scale, using machine learning for course recommendations, automated grading of programming assignments, and peer review matching.
The release of ChatGPT in November 2022 changed the landscape dramatically. For the first time, students and teachers had access to a general-purpose AI that could write essays, solve math problems, explain complex concepts, generate lesson plans, and engage in open-ended dialogue. Within two months, ChatGPT reached 100 million users, and education was one of the sectors most immediately affected [3].
The initial reaction from educational institutions was largely one of alarm. Several school districts, including New York City Public Schools, banned ChatGPT on school networks in January 2023, citing concerns about academic integrity. However, most of these bans were reversed within months as educators recognized that prohibition was neither practical nor desirable. By mid-2023, the conversation had shifted from "how do we block this" to "how do we integrate this thoughtfully" [3].
AI is now applied across a wide range of educational contexts. The following table summarizes the major application areas as of 2026:
| Application area | Description | Key examples | Status (2026) |
|---|---|---|---|
| AI tutoring systems | One-on-one tutoring that adapts to individual student needs, providing explanations, hints, and practice problems | Khan Academy Khanmigo, Duolingo Max, Socratic by Google | Widely deployed; Khanmigo used by 700,000+ K-12 students |
| Personalized learning | Adaptive platforms that adjust content sequencing, difficulty, and pacing based on learner performance | DreamBox, Carnegie Learning MATHia, ALEKS | Used by 43% of teachers according to industry surveys |
| Automated grading and essay scoring | AI systems that grade objective assessments and provide feedback on written work | Gradescope (Turnitin), ETS e-rater, Grammarly for Education | 41% of teachers report using automated scoring tools |
| Content generation | AI tools that help educators create lesson plans, quizzes, worksheets, and course materials | ChatGPT, Claude, MagicSchool AI, Eduaide.AI | Growing adoption; MagicSchool AI has 5M+ educator users |
| Language learning | AI-powered conversation practice, pronunciation feedback, and adaptive vocabulary training | Duolingo Max (Roleplay, Video Call), Speak, ELSA Speak | Duolingo Max available in 40+ language courses |
| Accessibility | AI tools for text-to-speech, speech-to-text, real-time captioning, and translation to support students with disabilities or language barriers | Microsoft Immersive Reader, Google Live Transcribe, Otter.ai | Integrated into major LMS platforms |
| Administrative automation | AI for scheduling, enrollment management, early warning systems for at-risk students, and institutional analytics | Civitas Learning, EAB Navigate, Anthology Reach | Adopted by hundreds of universities |
| Assessment and proctoring | AI-powered remote exam proctoring and plagiarism detection | Proctorio, ExamSoft, Turnitin AI detection | Controversial; accuracy of AI detection debated |
Khan Academy was among the first educational organizations to partner with OpenAI to build an AI tutoring product. Khanmigo, launched in 2023 and powered by GPT-4, functions as a personal tutor and teaching assistant. Unlike general-purpose chatbots, Khanmigo is designed to guide students toward answers rather than simply providing them. When a student asks for help with a math problem, the system asks probing questions and offers hints, mimicking the Socratic method [4].
Khanmigo expanded rapidly during the 2024-25 school year, growing from 40,000 to 700,000 K-12 students, with projections to surpass one million in the 2025-26 academic year. The tool is now used across hundreds of school districts in the United States and has begun reaching classrooms in India, Brazil, and the Philippines. For individual users, Khanmigo costs $4 per month or $44 annually, while teacher-support tools are offered for free [4].
Khanmigo covers math, science, reading, writing, computer science, and test preparation. For teachers, it can generate lesson plans, create assessments aligned to specific standards, and provide insights into student progress.
Duolingo, the language learning platform with over 100 million monthly active users, launched Duolingo Max in March 2023 as a premium tier powered by GPT-4. The subscription introduced two AI-driven features: Roleplay, which places learners in simulated real-world conversations (ordering coffee in a Parisian cafe, negotiating at a market), and Explain My Answer, which provides detailed, contextual feedback on why an answer was correct or incorrect [5].
In 2024, Duolingo added Video Call with Lily, enabling users to have voice conversations with an AI character, providing speaking practice that was previously difficult to access without a human conversation partner. Internal data from Duolingo shows 34% better grammar retention with the Explain My Answer feature, and 78% of Roleplay users report increased speaking confidence within four weeks [5].
By 2025, Duolingo had used AI to launch 148 new language courses, dramatically accelerating content production through AI-assisted creation with human oversight. Duolingo Max is priced at $29.99 per month or $167.99 annually [5].
MagicSchool AI has emerged as one of the most widely adopted AI platforms designed specifically for educators. By 2025, it reported over 5 million educator users. The platform provides over 60 AI-powered tools tailored to teacher workflows, including:
| Tool category | Examples |
|---|---|
| Lesson planning | Generate standards-aligned lesson plans, unit plans, and scope-and-sequence documents |
| Assessment creation | Create quizzes, rubrics, and differentiated assessments with answer keys |
| Differentiation | Generate materials at multiple reading levels for the same content |
| Communication | Draft parent emails, recommendation letters, and IEP (Individualized Education Program) goals |
| Student feedback | Provide first-pass feedback on student writing with specific, actionable suggestions |
Other teacher-focused AI platforms include Eduaide.AI, SchoolAI, and Brisk Teaching. These tools aim to save teachers time on routine tasks, with surveys finding that teachers who use AI tools at least weekly save an average of 5.9 hours per week, adding up to roughly six extra weeks of reclaimed time across a standard school year [10].
Student adoption of generative AI tools has been swift and continues to accelerate. A Higher Education Policy Institute (HEPI) survey conducted in February 2025 found that 88% of students reported using generative AI tools such as ChatGPT for assessments. Pew Research Center data from January 2025 showed that 26% of U.S. teens already use ChatGPT for schoolwork, and a September 2025 RAND survey found that 54% of K-12 students used AI for school, up more than 15 percentage points in two years. By early 2026, an estimated 86% of all students in higher education utilize AI as their primary research and brainstorming partner [6][10].
Students use LLMs for a variety of academic tasks: brainstorming essay topics, getting explanations of complex concepts, debugging code, summarizing readings, practicing for exams, and generating first drafts. Some of these uses are clearly constructive (using an LLM as a study aid is analogous to asking a knowledgeable friend for help), while others cross into academic dishonesty (submitting AI-generated text as one's own work).
The line between legitimate use and misuse is not always clear, and it varies by institution, course, and assignment. This ambiguity has been one of the central challenges for educators since 2023.
LLMs have also become popular tools for educators themselves. Teachers use AI to generate lesson plans, create differentiated materials for students at different levels, write quiz questions, provide feedback on student drafts, and handle administrative tasks like writing recommendation letters or parent communications.
Platforms like MagicSchool AI, which reported over 5 million educator users by 2025, provide teacher-specific AI tools with built-in guardrails. These platforms are designed to generate content aligned to educational standards (such as Common Core or Next Generation Science Standards) and to save teachers time on routine tasks, potentially freeing them to spend more time on direct instruction and student interaction.
The integration of AI into K-12 education has progressed rapidly, though unevenly, across the United States and globally.
As of March 2025, 28 states have published or adopted AI guidance for K-12 education, up from fewer than 10 in early 2024. These guidelines vary significantly in scope and specificity:
| State / Initiative | Key provisions |
|---|---|
| Connecticut | Launched an AI Pilot Program in spring 2025 across seven districts, introducing students in grades 7-12 to state-approved AI tools with hands-on learning experiences |
| California | Issued comprehensive AI guidance for schools covering responsible use, data privacy, and equity considerations |
| Virginia | Published a framework for AI literacy and ethical use in K-12 classrooms |
| North Carolina | Developed AI integration guidelines with emphasis on teacher training and digital literacy |
| U.S. Department of Education | Issued federal guidance in 2025 on AI use in schools, including recommendations for equity, privacy, and age-appropriate deployment |
Approximately 74% of districts planned to provide teacher AI training by fall 2025, reflecting growing recognition that effective AI integration requires teacher preparation [10].
A significant challenge exists in educator preparedness. While 63% of U.S. teens report using AI tools like ChatGPT for schoolwork, only 30% of teachers report feeling confident using these same AI tools. Student use of AI for school-related purposes jumped 26% since the previous school year, while educator use rose 21% over the same period [10].
This readiness gap creates situations where students may be more familiar with AI capabilities and limitations than their teachers, complicating classroom management and assessment design.
Research on AI-powered personalized learning has produced increasingly strong evidence of effectiveness, though with important caveats.
A 2024 study by researchers at Harvard found that students using an AI tutor in a computer science course learned 1.5 times faster and scored significantly higher on assessments than those in a traditional lecture format. The study controlled for prior knowledge and motivation, suggesting that the AI tutor's ability to provide immediate, individualized feedback was a genuine advantage [8].
A 2025 Harvard University physics study produced even more striking results, finding that students using AI tutors learned more than twice as much in less time compared to those in traditional active-learning classrooms [10].
A meta-analysis published in Computers & Education in 2025 reviewed 83 studies of AI-assisted learning and found a moderate positive effect (Cohen's d = 0.41) on learning outcomes compared to traditional instruction. The effect was strongest for personalized tutoring systems and weakest for simple content recommendation engines [8].
Khan Academy's internal research has shown that students using Khanmigo spend 30% more time on task compared to students using the platform without the AI tutor, suggesting that the conversational interface increases engagement [4].
In language learning, Duolingo's data indicates that the Roleplay feature in Duolingo Max leads to measurable improvements in conversational confidence and grammar retention. The company reports that users who engage with Roleplay sessions at least three times per week show 34% better grammar retention on follow-up assessments [5].
Despite these promising findings, 86% of educational organizations use generative AI, the highest of any industry in the United States, yet most AI-based educational tools have not undergone independent validation, and few have been tested through rigorous methods such as randomized controlled trials. The gap between adoption rates and scientific validation remains a significant concern [10].
AI is transforming both how students are assessed and how assessments are created.
Teachers increasingly use AI to generate assessments. AI tools can create multiple-choice questions, short-answer prompts, and essay questions aligned to specific learning standards. More advanced tools can generate assessments at multiple difficulty levels, create alternative versions of the same test to reduce cheating, and produce detailed rubrics.
Automated essay scoring (AES) systems use NLP to evaluate student writing on dimensions including organization, vocabulary, grammar, and argument quality. The Educational Testing Service (ETS) has used its e-rater system as a component of GRE essay scoring for over two decades. Newer AI-powered systems from companies like Turnitin (through its Gradescope platform) can provide more detailed, formative feedback.
However, AES systems have known limitations:
Schools that have redesigned assessments to be more AI-resilient, moving toward oral exams, in-class writing, process portfolios, and project-based evaluations, report 40% fewer AI-related integrity issues compared to those relying solely on detection [6].
The most immediate and widely discussed concern about AI in education is its impact on academic integrity. AI-related misconduct grew from 1.6 to 7.5 cases per 1,000 students between 2022 and 2026. Over 8 in 10 students reported using generative AI during the school year in 2025, and for many, that included graded assignments [6].
The detection problem is severe. While 68% of teachers report using AI detection tools, only 54% of faculty feel effective at identifying AI-generated content. Research from the University of Reading found that 94% of AI-generated submissions went undetected in a blind evaluation. This creates an uneven playing field: students who use AI without disclosure may receive higher grades than those who do their own work [6].
Three in four chief technology officers at educational institutions say that artificial intelligence has proven to be a moderate or significant risk to academic integrity at their institution [6].
AI detection tools themselves have introduced a new form of inequity. Studies have found that non-native English speakers face a 61.2% false positive rate (meaning their original work is flagged as AI-generated), compared to just 5.1% for native speakers. This disparity has led institutions including Princeton and MIT to advise against relying solely on AI detectors [6].
The bias arises because non-native speakers sometimes produce writing that is more formulaic or uses simpler sentence structures, patterns that detection algorithms associate with AI generation. As a result, the students most likely to be falsely accused of cheating are often those who are already marginalized in educational settings.
Educators worry that students who rely heavily on AI tools may fail to develop critical thinking, writing, and problem-solving skills. If a student can get an AI to write every essay, they may never learn to organize their thoughts, construct an argument, or revise their own prose. This concern is particularly acute in foundational courses where the process of struggling with material is itself the learning objective.
Similar concerns have been raised about math and science education. When students use AI to solve problems step by step, they may get correct answers without understanding the underlying concepts. The risk is that AI becomes a crutch rather than a scaffold.
AI educational tools collect substantial amounts of student data, including performance metrics, learning patterns, interaction logs, and sometimes biometric data (in the case of proctoring tools). This data collection raises privacy concerns, particularly for minor students. In the United States, the Family Educational Rights and Privacy Act (FERPA) and the Children's Online Privacy Protection Act (COPPA) govern student data, but the application of these laws to AI tools is not always clear.
The European Union's General Data Protection Regulation (GDPR) imposes stricter requirements, particularly regarding automated decision-making and the processing of children's data. Schools using AI tools must ensure compliance with these regulations, which can be complex when tools are provided by third-party vendors whose data practices may not be fully transparent.
AI tools in education risk widening existing inequalities. Students at well-funded schools with strong technology infrastructure and institutional AI subscriptions have access to tools that students at under-resourced schools do not. Paid tiers of AI tools (like Duolingo Max at $29.99/month or ChatGPT Plus at $20/month) are out of reach for many families.
This digital divide extends globally. While schools in North America and Western Europe are debating how to integrate AI into curricula, schools in many developing countries lack reliable internet access, let alone AI-powered learning platforms. Without deliberate efforts to ensure equitable access, AI in education could become another mechanism through which privilege compounds.
Educational institutions have taken varied approaches to AI use. A 2025 analysis of 174 universities found that only 30% had an explicit AI policy, while the remaining 70% had no AI-specific guidelines. Among those with policies, approaches range from broad permission with disclosure requirements to strict restrictions on summative assessments. Two-thirds of higher education institutions have or are developing guidance on AI use [7][10].
Common policy elements include:
| Policy approach | Description | Examples |
|---|---|---|
| Full prohibition | AI tools banned for all academic work | Initial NYC Public Schools ban (January 2023, later reversed) |
| Restricted use | AI allowed for some tasks but prohibited for graded submissions unless explicitly permitted | Columbia University, many UK universities |
| Permitted with disclosure | AI use allowed but students must document what tools they used, what prompts they provided, and how they integrated the output | Harvard Graduate School of Education, Duke University |
| Course-level discretion | Individual instructors set their own AI policies for their courses | University of Texas at Austin, many Australian universities |
| Encouraged integration | AI tools actively incorporated into coursework and assignments | Some computer science and business programs |
The trend as of 2026 is moving toward course-level discretion with disclosure requirements. Institutions increasingly recognize that a one-size-fits-all policy is impractical given the diversity of disciplines and assignment types [7].
Several tools have been developed to detect AI-generated text, including Turnitin's AI detection feature (integrated into its plagiarism detection platform), GPTZero, Originality.ai, and Copyleaks. However, their reliability remains contested.
Duke University does not recommend AI detection software as part of an AI policy because the products are unreliable. OpenAI itself withdrew its own detection tool (AI Text Classifier) in July 2023 due to low accuracy. Australia's higher education regulator TEQSA warned in 2025 that AI-assisted cheating is "all but impossible" to detect consistently, and urged universities to redesign assessments rather than depend on AI detectors [7].
Schools that have redesigned assessments, moving toward oral exams, in-class writing, process portfolios, and project-based evaluations, report 40% fewer AI-related integrity issues compared to those relying solely on detection [6].
A growing number of schools and districts are developing AI literacy curricula as a core competency. These programs teach students:
Several organizations have developed AI literacy frameworks, including ISTE (International Society for Technology in Education), AI4K12 (a national initiative for K-12 AI education), and UNESCO's AI competency framework for students [10].
Despite the concerns, research has documented several positive effects of AI in education:
A 2024 study by researchers at Harvard found that students using an AI tutor in a computer science course learned 1.5 times faster and scored significantly higher on assessments than those in a traditional lecture format. The study controlled for prior knowledge and motivation, suggesting that the AI tutor's ability to provide immediate, individualized feedback was a genuine advantage [8].
Khan Academy's internal research has shown that students using Khanmigo spend 30% more time on task compared to students using the platform without the AI tutor, suggesting that the conversational interface increases engagement [4].
In language learning, Duolingo's data indicates that the Roleplay feature in Duolingo Max leads to measurable improvements in conversational confidence and grammar retention. The company reports that users who engage with Roleplay sessions at least three times per week show 34% better grammar retention on follow-up assessments [5].
A meta-analysis published in Computers & Education in 2025 reviewed 83 studies of AI-assisted learning and found a moderate positive effect (Cohen's d = 0.41) on learning outcomes compared to traditional instruction. The effect was strongest for personalized tutoring systems and weakest for simple content recommendation engines [8].
The AI in education market has grown rapidly and projections vary across research firms:
| Metric | Value | Source |
|---|---|---|
| Global market size (2025) | $7.05 billion to $18.9 billion | Precedence Research, Grand View Research |
| Projected market size (2026) | $9.58 billion | Precedence Research |
| Projected market size (2030) | $41 billion to $48.6 billion | Mordor Intelligence, Grand View Research |
| Projected market size (2035) | $136.79 billion | Precedence Research |
| CAGR (2025-2030) | 20.8% to 42.8% | Various |
| North America market share (2025) | 38% | Precedence Research |
| ITS share of AI education deployments (2025) | 30% | Industry reports |
| Educational organizations using generative AI (2025) | 86% | Industry surveys |
| K-12 student AI use for school (2025) | 54% | RAND survey |
| Higher education students using AI (2026) | 86% | Industry estimates |
The wide range of market estimates reflects differences in how firms define the boundaries of "AI in education" (some include only pure-play AI EdTech, while others include AI features within broader EdTech platforms). Regardless of the exact figures, the growth trajectory is clear [9].
North America dominates the market with approximately 38% share in 2025, driven by high EdTech investment and widespread institutional adoption. Asia Pacific is the fastest-growing region, with countries like India, China, and South Korea investing heavily in AI-powered educational infrastructure [9].
As of early 2026, AI in education is in a period of rapid, sometimes chaotic integration. Several trends define this moment:
Normalization of AI use. The initial shock and resistance that followed ChatGPT's launch has given way to a more pragmatic approach. Most institutions now accept that students will use AI tools and are focusing on how to channel that use productively rather than trying to prevent it entirely.
Assessment redesign. The most forward-thinking institutions are redesigning assessments to be more AI-resilient. This includes greater emphasis on oral examinations, in-class writing under supervised conditions, process portfolios that document the evolution of student work, and project-based assessments that require demonstrating competence in real time. Schools taking this approach report significantly fewer integrity issues [6].
AI literacy as a learning objective. A growing number of schools are treating AI literacy as a core competency, teaching students not only how to use AI tools effectively but also how to evaluate their outputs critically, understand their limitations, and consider the ethical implications of their use. Twenty-eight states have published K-12 AI guidance, and organizations like AI4K12 are developing standardized frameworks for AI education.
Expansion of AI tutoring. AI tutoring products like Khanmigo are scaling beyond pilot programs to widespread adoption. Khan Academy's growth from 40,000 to 700,000 student users in a single academic year suggests that the demand for AI-powered tutoring is substantial [4].
Teacher augmentation over replacement. The discourse has shifted away from fears that AI will replace teachers. The emerging consensus is that AI is most effective as a tool that augments teacher capabilities, handling routine tasks like generating practice problems, providing first-pass feedback on drafts, and answering common student questions, while freeing teachers to focus on the relational and motivational aspects of teaching that AI cannot replicate. Teachers using AI tools at least weekly save an average of 5.9 hours per week.
K-12 policy development. State-level AI guidance for K-12 education is expanding rapidly, with 28 states having published guidance by March 2025. Pilot programs like Connecticut's are providing models for structured AI integration in public schools.
Persistent equity gaps. Despite the potential of AI to democratize education, access remains uneven. Well-resourced schools and families benefit disproportionately, and the digital divide continues to be a limiting factor for equitable AI adoption in education worldwide.
The field is evolving rapidly, and the policies, tools, and pedagogical approaches of 2026 will likely look quite different from those of 2024. What seems clear is that AI is not a passing trend in education; it is a permanent feature of the learning landscape that institutions, educators, and students are still learning how to navigate.