AI in Education

AI in education refers to the use of generative AI across K–12 and higher education, including direct student use, teacher use, institutional deployment in admissions and grading, AI literacy curricula, and AI-powered edtech products. Generative AI entered classrooms and university curricula faster than prior educational technology: ChatGPT reached 100 million users faster than any earlier consumer product, and within weeks of its November 2022 launch was in regular use for homework, essay drafting, and exam preparation. The policy response three years on is still forming, with regulatory authority split among the US Department of Education, state education departments, accreditors, individual institutions, and, in the US K–12 context, often the classroom teacher. A central tension runs through the debate: AI functions both as a tutoring and accessibility tool and as a shortcut that may deskill students in the competencies education is meant to build.

Scope

The term covers several distinct uses:

• Student use — AI tools used directly by students for research, drafting, tutoring, coding, and problem-solving.
• Teacher use — AI tools used by teachers for lesson planning, differentiation, and grading assistance.
• Institutional use — AI in admissions, financial aid, advising, plagiarism detection, and administrative operations.
• Curriculum content — formal teaching about AI (AI literacy).
• Edtech products — AI-powered learning platforms such as Khanmigo and Duolingo Max.

Key debates

Augmentation versus substitution

The augmentation frame treats AI as a tutor or scaffold that supports learning. The substitution frame treats AI as a task-completer, a tool students use instead of learning. Early empirical work suggests both are occurring, with the mix depending on assignment structure, assessment design, and teacher framing.

Equity

AI tools that function as a leveler for working adults (see Brynjolfsson et al. 2023) may function as an accessibility and equalizing tool in education, enabling non-native English speakers, students with learning disabilities, and students without home tutoring resources to access scaffolded support previously available only through paid tutors. Unequal access to high-quality AI tools, and to the skills required to use them well, produces a second-order inequality.

Deskilling risk

The education-specific version of the AI Deskilling concern holds that students who outsource drafting, analysis, and problem-decomposition to AI during formative years may fail to develop the underlying competencies. The phenomenon is difficult to measure because traditional assessments measure combined output; only unaided testing reveals the baseline drift.

Institutional response modes

Institutions have converged on three broad postures:

1. Prohibit and detect — ban AI use, deploy detection tools, and discipline violators. This was the dominant posture in 2023 and was widely abandoned as detection failed. A 2026 revival appears in UC Berkeley School of Law's Summer 2026 policy, which returns to a default-prohibition posture and replaces AI-detection with per-se evidentiary presumptions keyed to specific LLM failure modes, so that a hallucinated citation raises a presumption of prohibited AI use. The approach anchors enforcement on observable artifacts of AI use rather than statistical detection, sidestepping the unreliable-detection-tool problem, and may serve as a template for other top-tier law schools.
2. Permit and disclose — allow AI use provided students disclose it. This is the most common current policy.
3. Integrate and assess for AI-aware skills — redesign assignments around the assumption that AI is available, assess in-person for foundational skills, and explicitly teach AI-collaboration competencies.

Representative deployments and use cases

ChatGPT in classrooms

Within weeks of ChatGPT's November 2022 launch, K–12 and university students were using it at scale. Survey data across 2023–2025 consistently shows majority use among secondary and tertiary students for at least some schoolwork. Early institutional responses, including NYC public schools and a number of universities, banned ChatGPT on institutional networks; most reversed within a year.

Khan Academy and Khanmigo

Khanmigo, Khan Academy's AI tutor, launched in limited beta in 2023 and rolled out through 2024–2025, is the most prominent institutionally deployed AI tutor. Built on GPT-4 and later successors, it uses a Socratic-method scaffold designed to avoid direct answer-giving in favor of prompted student reasoning. Khan Academy has partnered with school districts in several US states for teacher-assisted deployment. Khanmigo serves as a test of the augmentation-with-guardrails approach, as it is engineered specifically not to substitute for student work.

Duolingo Max, Chegg, and paid edtech

Commercial edtech has integrated AI broadly. Duolingo Max (2023) added AI conversation practice; Chegg lost market capitalization rapidly as ChatGPT eroded its homework-help business; many incumbents have repositioned around AI integration. The effect on student learning is underdetermined by current evidence.

AI in admissions

Several universities use AI tools in application screening; the practice is opaque and contested. Common Application processes and merit-scholarship reviews have been accused of using AI scoring, and disclosure practices are uneven. This is a specific case of the broader Algorithmic Accountability and Bias Audits concern, in that the domain most regulated in employment is largely unregulated in admissions.

Automated essay grading

Automated essay scoring predates generative AI, for example ETS's e-rater, used in GRE analytical writing scoring. Generative AI has expanded the use case to AI-assisted rubric scoring, feedback generation, and comment drafting for teachers. Empirical evidence on accuracy and bias is mixed; concerns about unreliability and bias-compounding have produced cautious guidance from state education departments.

Plagiarism detection failures

GPTZero, OriginalityAI, Turnitin's AI-detection feature, and OpenAI's own AI Text Classifier (retired in 2023) have all faced accuracy problems. False positives disproportionately flag non-native English writers and certain writing styles. False negatives are common with minor edits. By 2024, Turnitin had walked back confidence claims and OpenAI had withdrawn its classifier. The practical consequence is that AI-detection-based discipline regimes are unreliable, which has contributed to a move away from prohibit-and-detect postures.

Accessibility benefits

AI tools provide documented accessibility gains: text-to-speech for dyslexic students, real-time translation for multilingual classrooms, scaffolded explanations for students with learning disabilities, and voice-input for students with motor impairments. These gains are well-documented but less prominent in public debate than deskilling and cheating concerns.

Policy and regulation

A growing set of jurisdictions are mandating AI literacy. The EU AI Act Article 4 requires providers and deployers of AI systems to ensure staff and users have a sufficient level of "AI literacy," in force February 2025. California and Oregon have adopted state-level K–12 AI literacy frameworks. The US Department of Education issued non-binding guidance in 2023, "Artificial Intelligence and the Future of Teaching and Learning," that most state agencies have adopted as a baseline.

Other jurisdictions have moved toward restriction rather than literacy mandates. On June 19, 2026, Norway imposed a near-ban on generative-AI use by elementary-school children and restricted its use for older students, citing concerns about learning (Source: reuters.com). The action illustrates the "prohibit and detect" posture applied at the national level for younger pupils, in contrast to the literacy-mandate approach elsewhere.

The broader policy landscape is summarized below:

Relationships

• depends-on: AI Deskilling — the underlying mechanism by which AI substitution could degrade learning
• related: AI and Productivity — task-level productivity gains map imperfectly to educational outcomes
• related: Algorithmic Accountability and Bias Audits — admissions and grading tools are in scope of high-risk accountability regimes
• related: EU AI Act (Regulation 2024/1689) — Article 4 AI literacy requirement
• related: AI Compliance Industry / Regulatory Fragmentation — uneven institutional policy landscape
• instance-of: sectoral AI adoption under partial regulation