Does AI personalise learning effectively?

Effective personalisation requires more than progress tracking. It demands accurate modelling of what a student currently understands, what misconceptions they hold, and what learning path maximises both efficiency and retention. The best systems, built on Bayesian knowledge tracing or reinforcement learning, estimate knowledge state in real time and adjust accordingly. However, many commercial platforms labelled as AI-powered offer only superficial adaptation. The difference between genuine adaptive learning and basic progress tracking matters enormously for learning outcomes.

What are the risks of AI in education?

Several are well-documented. Algorithmic bias can direct students from disadvantaged backgrounds toward less challenging content. Data privacy is a particular concern for minors, as educational platforms collect granular interaction data that can reveal learning patterns and emotional states. Over-reliance on AI tutoring risks atrophying the social and collaborative dimensions of learning. And the commercial incentives of edtech platforms do not always align with pedagogical best practice, creating tension between engagement metrics and genuine learning.

How are teachers responding to AI tools?

Responses range from enthusiastic adoption to deep scepticism, often within the same school. Teachers who find AI tools helpful report that automated marking frees time for individual attention and that adaptive platforms help differentiate instruction across mixed-ability classrooms. Sceptics point to inconsistent quality, the loss of professional autonomy, and a growing expectation that teachers absorb technical support responsibilities without additional training or compensation. Teacher unions in several countries have called for formal consultation before AI platforms are mandated in state schools.

Can AI help close the global education equity gap?

It can help, but only if infrastructure and content prerequisites are in place. AI tutoring does not work without reliable internet, devices, and electricity, resources that remain scarce in the communities where the gap is widest. Where conditions allow, AI tools can multiply the reach of limited teaching staff and provide instruction in local languages. UNESCO's position is that AI should complement, not substitute for, public investment in trained teachers and school infrastructure.

Home›Research›AI in Education

AI in Education Research

Adaptive learning, intelligent tutoring, and the digital divide in AI-enhanced education

Overview

Education sits at an inflection point. The one-size-fits-all classroom model has never been particularly effective, and the pandemic laid its limitations bare. AI offers a different proposition: instruction that adapts to each learner in real time, feedback that arrives immediately rather than weeks later, and assessment that measures understanding rather than test-taking ability. The question is whether these promises hold up at scale, across diverse populations, and under the messy conditions of real schools.

The evidence, though still accumulating, is cautiously encouraging in specific areas. Adaptive learning platforms powered by Bayesian knowledge tracing and reinforcement learning have demonstrated measurable improvements in controlled settings. Intelligent tutoring systems are now deployed in school districts serving large numbers of students. Automated essay scoring and formative feedback generation have reached reliability levels that, for structured tasks, approach human performance. But the gap between what works in a controlled trial and what works in an under-resourced classroom remains wide.

This page presents verified research developments in AI education, sourced to peer-reviewed publications, institutional reports, and documented deployments. We note where evidence is strong and where it remains preliminary.

Key Developments

Khan Academy's Khanmigo

Khanmigo, Khan Academy's AI tutor built on GPT-4, launched in March 2023. A pilot programme reached approximately 65,000 students across 53 US school districts. In August 2024, Khan Academy launched Khanmigo for Teachers, providing free AI tools for educators in partnership with Microsoft. The tool was featured on CBS 60 Minutes in December 2024. However, a February 2024 Wall Street Journal evaluation found that Khanmigo made basic math calculation errors, highlighting the limitations of large language models in educational settings. Khanmigo is available to individual users at a subscription cost of four dollars per month.

UNESCO Reports on AI in Education

UNESCO's Global Education Monitoring Reports have examined AI deployments in education across multiple countries, highlighting a significant gap between policy adoption and actual evaluation of learning outcomes. The reports call for greater attention to equity, data privacy, and teacher training in AI education deployments. UNESCO has emphasised that hundreds of millions of children worldwide lack minimum proficiency in reading and mathematics, and that AI tools must be evaluated for their impact on these fundamental gaps rather than adopted on technological promise alone.

Adaptive Learning Research at Carnegie Mellon

Carnegie Mellon University has conducted extensive research on adaptive learning systems using cognitive tutors, a line of work spanning several decades. Published studies have shown that adaptive, AI-driven learning systems can reduce the time needed to learn material compared to traditional instruction. Meta-analyses in the Review of Educational Research and other journals have examined the effectiveness of AI-generated feedback, finding positive results for structured tasks but noting limitations for higher-order skills. The evidence base is growing but still limited by relatively small sample sizes and short study durations in many published trials.

AI Translation and Language Access

Meta's No Language Left Behind initiative has developed translation models covering over 200 languages, including many low-resource languages that previously had limited machine translation support. Google Translate and other commercial services have similarly expanded AI-powered translation capabilities. These tools have the potential to improve educational access for speakers of underrepresented languages, though translation quality varies significantly across language pairs. Educational deployments of AI translation remain largely experimental, with limited peer-reviewed evaluation of learning outcomes.

Challenges: Digital Divide, Privacy, and Bias

The digital divide means that students in low-income countries have far less access to AI-enhanced learning tools, risking a widening of existing educational inequalities. Concerns about data privacy and student surveillance are particularly acute when commercial technology companies provide educational platforms that collect granular interaction data. Teacher training and readiness remain significant barriers to effective AI integration in classrooms. Researchers have also documented the risk of AI perpetuating biases present in training data, which could direct students from disadvantaged backgrounds toward less challenging content or misassess their capabilities.