By Jonathan Harris, AI author and host of Turing’s Torch AI Weekly.
What this guide covers
AI applications in education divide roughly into three categories: tools that support learners directly, tools that support educators, and tools that support administration. Each has a distinct evidence base and distinct risks.
Learner-facing tools include adaptive learning platforms that adjust content and pacing to individual performance, AI tutors that respond to student questions, writing support tools that give feedback on drafts, and language learning platforms that simulate conversation. The most established of these have decades of research behind them; the newest are still accumulating evidence.
Educator-facing tools include automated marking and feedback generation, early-warning systems that flag at-risk students before they disengage, lesson planning assistants, and tools that help teachers differentiate instruction across ability levels. Administrative tools cover scheduling, resource allocation, and compliance documentation.
The academic integrity question is now live for every institution. Generative AI has made it possible for students to produce plausible assignments without engaging with the underlying material. Institutional responses range from prohibition to redesign of assessments to explicit integration of AI as a learning tool.
Where it works well
Personalisation at scale is the strongest honest case for AI in education. A good human tutor adjusts in real time to what a student knows and does not know. AI systems can approximate this across a class of thirty or a platform with a million users simultaneously. The evidence on intelligent tutoring systems — some now decades old — shows meaningful learning gains in the right conditions.
Feedback speed matters for learning. Students who receive feedback within minutes of an attempt learn more from it than students who receive the same feedback a week later. AI can provide immediate feedback on writing drafts, mathematical working, and code, at no marginal cost per student.
Early-warning systems have shown genuine results in higher education settings. Institutions that acted on predictive signals about at-risk students — offering targeted support before a student disengaged entirely — improved retention rates measurably. The intervention had to be human; the signal was algorithmic.
Where it gets complicated
Engagement metrics and learning outcomes are not the same thing. Optimising for time on platform or completion rates is not the same as optimising for understanding. AI systems that are optimised for the former can produce learners who feel productive while not learning deeply. Distinguishing genuine progress from engagement theatre requires assessment that AI cannot easily game.
Equity effects are real and poorly understood. Wealthy students already have advantages in tutoring and test preparation. AI tools that are high quality but expensive, or that work best for students with reliable devices and internet connections, can entrench existing disparities rather than reduce them. This is not an argument against the tools but an argument for thinking carefully about who gets access on what terms.
The what-is-education-for question is not comfortable but is inescapable. If AI can write, code, calculate, and research — and it can — then skills-based education that focuses on performing those tasks faces a fundamental challenge. Institutions that are honest about this are redesigning around what AI cannot replace: judgement, original inquiry, collaborative meaning-making, ethical reasoning.
FAQ
Do AI tutors actually work?
The research on intelligent tutoring systems, which have been studied for decades, shows genuine learning gains in controlled settings, particularly in mathematics and science. Newer LLM-based tools have a shorter evidence base but early results are mixed — they are capable but also prone to confidently generating wrong explanations.
How should schools handle AI and academic integrity?
There is no single right answer. Some institutions prohibit AI use. Others require disclosure. Others redesign assessments around oral defence, process documentation, or in-class work. The institutions with clearest policies have usually started from a curriculum design question rather than a technology question.
Will AI replace teachers?
No. The evidence from decades of educational technology is that technology that works in education almost always works in conjunction with skilled teachers, not in place of them. What changes is what teachers spend time on.
What is adaptive learning?
An approach in which the learning system adjusts content difficulty, pacing, and feedback based on each student's performance. The system builds a model of what the student knows and chooses the next learning step to be maximally useful for that student specifically.