Executive summary
AI integration in education has moved from experimental pilot programs to large-scale implementations that fundamentally change how students learn and teachers teach. Schools worldwide are now deploying adaptive learning systems, AI tutors, and personalised education platforms that deliver measurable improvements in student outcomes, and giving teachers powerful new educational tools.
- Squirrel AI in China taught 112,718 students simultaneously with personalised math lessons, creating over 108,000 unique learning paths in real-time
- Alpha School’s model compresses traditional academics into just two hours using AI tutors, freeing afternoons for hands-on projects and life skills
- Iowa invested US$3m to provide AI reading tutors to all elementary schools statewide, with voice recognition technology providing instant feedback
- Medical schools like Dartmouth use AI-simulated patients for clinical training, allowing unlimited practice in multiple languages
- Teacher adoption has surged, with 50% of educators reporting increased AI usage throughout the 2023-24 school year
- “No two students are alike, and thus no two learning paths should be alike…Every child deserves access to an education as unique as they are,” says Derek Li, co-founder of Squirrel AI.
What distinguishes AI approaches from traditional ones is its ability to support personalised learning at scale. Intelligent tutoring systems – software that mimics one-on-one tutoring – have demonstrated measurable gains in learning compared with standard classroom approaches. The transformation to personalised learning is already underway, with AI enabling this shift at unprecedented scale, addressing long-standing educational challenges through data-driven personalisation.
Education has always grappled with a fundamental challenge: how to meet the diverse needs of individual learners within the practical constraints of classroom instruction. Teachers themselves have long recognised that students arrive with different backgrounds, learn at different paces, and respond to different teaching approaches. Yet most educational systems still organise learning around fixed curricula delivered to groups of 20 or 30 students at once.
Recent years have seen various attempts to address this challenge through technology. Early computer-assisted learning programmes offered branching pathways through material. Online platforms enabled self-paced study. Now, AI is emerging as another tool in this ongoing effort to personalise education. AI systems can analyse patterns in how students interact with material, identify areas where they struggle, and adjust the presentation of content accordingly. In some classrooms, these systems are being used to generate practice problems tailored to each student’s current level of understanding, provide immediate feedback on assignments, or flag concepts that might need additional review.
The conditions for implementing such systems have only recently aligned. Sufficient computing power, reliable internet infrastructure, and accumulated educational data now make it possible to deploy AI tools in everyday classroom settings. Schools worldwide are beginning to experiment with these technologies, producing a range of outcomes that merit careful examination. In the following sections, we will examine how schools are implementing AI, what early results suggest about its possible impact, and what questions remain unanswered as this experiment in AI-assisted education unfolds.
“No two students are alike, and thus no two learning paths should be alike…Every child deserves access to an education as unique as they are.”
Derek Li, co-founder of Squirrel AI
1. Personalised learning at massive scale: Squirrel AI’s record-breaking adaptive classroom
How do you personalise learning for 100,000 students at once? You use AI.
When we think about personalised education, we typically imagine small class sizes with dedicated teachers providing individual attention. Squirrel AI Learning challenged that assumption in September 2024 when it simultaneously taught 112,718 students across China in a single online math lesson, setting a Guinness World Record in the process. The platform’s AI system generated over 108,000 unique learning paths and customised exercises tailored to individual student levels in real time, with each student progressing at their own pace while the system continuously adjusted difficulty and content based on their responses and engagement patterns.
This isn’t simply a case of impressive numbers: the platform demonstrates how AI could potentially deliver deeply personalised education at scales that would overwhelm entire faculties. Each student received individualised attention comparable to having a private tutor, something that would have been impossible to achieve with conventional teaching methods. The system tracked not just whether students answered correctly but also how long it took them to respond, how they approached the problem, and which areas they had difficulties understanding. It then used this data to optimise each student’s learning pathway.
The technology behind mass personalisation
Squirrel AI’s platform uses machine learning algorithms to analyse student performance across multiple dimensions simultaneously. The system quickly identifies knowledge gaps, predicts which concepts may present difficulties for a student, and preemptively provides additional support or practice in those areas. During the aforementioned mega-session, the AI managed to maintain consistent performance while handling what amounts to a small city’s worth of students, suggesting that educational AI may finally be ready for a wider rollout.
The company’s success in China offers takeaways for education systems worldwide. If AI can effectively teach 100,000 students mathematics simultaneously while maintaining the quality of instruction, similar systems could help address teacher shortages, provide quality education in underserved areas, and offer advanced courses in schools too small to justify specialised instructors. “No two students are alike, and thus no two learning paths should be alike,” says Derek Li, Squirrel AI’s co-founder. “Every child deserves access to an education as unique as they are.”
2. Reimagining the school day: Alpha School’s AI-powered academic model
Instead of delivering lectures and grading papers, teachers will transition to the role of a mentor who guides students through their learning journey.
Today’s school schedules can often feel antiquated – students sitting in rows, moving through subjects at predetermined intervals regardless of their individual learning speeds or interests. Alpha School in the US has completely reimagined this model, using AI to compress a full day’s worth of academic instruction into just two focused hours each morning. Students spend the first part of their day interacting with personalised AI tutors that guide them through math, reading, science, and social studies at their individual pace. The system assesses their knowledge in real time, identifying gaps and ensuring they achieve mastery before progressing. Rather than moving through material based on arbitrary timelines, students advance only after demonstrating genuine understanding.
The time saved through AI-powered learning frees up the entire afternoon for other activities that Alpha School considers equally important for student development: hands-on projects, life skills workshops, entrepreneurship training, and collaborative problem-solving exercises. Teachers, who are here referred to as ‘guides’, no longer spend time lecturing or grading traditional assignments in core subjects. Instead, they serve as mentors and motivators, helping students stay engaged with their coursework in the morning and leading interactive workshops in the afternoon on topics ranging from public speaking to financial literacy.
Balancing efficiency with human connection
Alpha School’s approach addresses two persistent criticisms directed at educational technology: that purely digital learning isolates students from human contact, and that one-size-fits-all instruction wastes time. By having students learn core academics individually but in a shared physical space, followed by collaborative afternoon activities, the school maintains the social benefits of traditional schooling while mitigating its weaknesses. Students still interact with peers and adults throughout the day, but the interactions focus on application, creativity, and skill-building rather than passive listening to lectures.
Early results suggest students are advancing faster academically while also developing stronger practical skills. The model has attracted attention from education reformers worldwide who see it as a template for making schools both more efficient and more relevant to modern life. While it’s currently only a private school available to a limited number of students in Texas and Florida, Alpha School plans to expand its programme by establishing both virtual and physical charter schools across the US, potentially reshaping how we think about the structure of the school day itself.
3. State-wide AI implementation: Iowa’s universal reading tutor initiative
AI ensures that each student receives personalised instruction, resulting in higher student engagement and better learning outcomes.
Reading fluency forms the foundation of educational success, yet many students struggle with pronunciation, pacing, and comprehension without receiving adequate individual attention. Iowa addressed this challenge head-on by becoming the first US state to provide AI reading tutors to every elementary school, investing US$3m in 2025 to deploy the EPS Reading Assistant across public and private schools statewide. The AI system uses speech recognition technology to listen as children read aloud, providing instant corrective feedback and support to build essential reading skills. A digital avatar named Amira serves as each student’s personal reading coach, assessing pronunciation, fluency, and comprehension in real time while adapting instruction to individual needs.
The programme aims to address a critical resource constraint in elementary education: the impossibility of providing each student with sufficient guided reading practice using human staff alone. By scaling personalised reading instruction across an entire state, Iowa demonstrates AI’s capacity to supplement teachers’ abilities, while substantially expanding access to individualised support.
Data-driven literacy at scale
Each reading session begins with students selecting appropriate-level texts while Amira monitors their performance through advanced speech recognition. When students struggle with pronunciation or stumble over words, the AI provides immediate correction and guidance, helping to build confidence and accuracy. The system adapts to each child’s pace, offering encouragement for progress while gently correcting mistakes. Teachers receive detailed analytics on each student’s reading performance, including strengths, areas needing improvement, and progress tracking over time.
This data enables targeted intervention and allows educators to focus their limited one-on-one time on students requiring the most support. Early results are promising, indicating improved reading fluency and increased student confidence during reading activities. The program particularly benefits students who might feel embarrassed reading aloud in front of classmates, providing a safe environment for practice and improvement. Iowa’s success could serve as a model for other states seeking to address reading achievement gaps through technology-enhanced instruction.
4. Dartmouth’s AI training: medical students practice on virtual humans
What if medical students could practice their skills without putting patients at risk?
Medical students face a challenging paradox: they need extensive practice with patient interactions before treating real people, but opportunities for safe, controlled practice are scarce and expensive. Dartmouth’s Geisel School of Medicine solved this problem by developing an AI-powered patient simulator that allows medical students to practice diagnosis and bedside manner on virtual patients.
Professor Thomas Thesen and his team built an app that uses GPT-4 to role-play a patient with symptoms, allowing students to conduct a virtual interview and exam. The AI Patient Actor app creates realistic patient personas complete with medical histories, symptoms, and even emotional responses, providing unlimited practice opportunities without the logistical challenges associated with recruiting actors to play patients. Students can repeat challenging cases multiple times, experimenting with different approaches to improve their skills.
Transforming medical education through AI simulation
The system goes beyond simple question-and-answer interactions. AI patients exhibit realistic patient behaviours like anxiety or confusion, and maintain consistent medical narratives throughout the consultation. Each one presents a medical case (for example, memory loss consistent with early Alzheimer’s) and can respond to the student’s questions and even emotional cues – showing concern when discussing serious symptoms or expressing relief when reassured about benign conditions. This helps students develop empathy and communication skills alongside clinical knowledge.
The system gives instant feedback on the student’s performance – whether they asked the right questions, their bedside manner, and if their diagnosis is correct. This instant feedback loop appears to accelerate learning compared to traditional methods, where students might wait days or weeks for assessment results. Students can even toggle language settings to practice multilingual patient care, preparing them for diverse clinical environments. Dartmouth’s broader AI curriculum initiative reflects growing recognition that future physicians must understand and work alongside AI systems. “The goal is to educate future physicians in the responsible use of AI and digital health in medical practice to improve patient outcomes and to build a cohort of leaders in the field,” says Thesen.
5. Classroom AI companions: Germany’s humanoid teacher experiment
Can a robot teacher do a better job than its human counterpart?
In December 2024, Willms Gymnasium in Delmenhorst, Germany, became the first European school to have a humanoid robot teach an entire lesson. Developed by Hong Kong company Hidoba Research and named Captcha, the robot led students through an interactive lecture titled ‘The difference between how AI thinks and how humans think’. It engaged students in discussions, asked probing questions to check their understanding, and moderated debates about AI’s societal impact.
The robot’s visit, which was initiated by students who met Captcha at a UN technology summit, generated national media attention and sparked important conversations about AI’s role in education. Students reported that while the robot’s arguments were logical and well-structured, it lacked the personal warmth and intuitive understanding of human teachers. Captcha can speak five languages, track faces and maintain eye contact, and adjust its personality settings, with the default being a ‘cheeky teenager boy’ designed to engage young learners.
Testing the boundaries of AI in teaching
The experiment provided valuable insights into what aspects of teaching can be automated and what remains uniquely human. Students found the experience engaging and memorable, with many reporting increased interest in AI and technology careers. Teachers observed that while the robot could deliver content effectively and manage structured debates, it couldn’t read the room’s emotional temperature or adjust its approach based on subtle social cues.
The demonstration highlighted both the advantages and limitations of robotic educators. While a humanoid robot might help address teacher shortages or provide specialised instruction in remote areas, the experience of German students and teachers reinforced the notion that effective teaching requires emotional intelligence, creativity, and adaptability that current AI cannot fully replicate – and maybe never will. Schools considering similar technology will need to position AI as a complement to, rather than a replacement for, human educators.
“Some kids go home to parents who can help revise their papers. Others go home to an empty house. We needed a way to give every student the same type of feedback and support.”
Kyle Beimfohr, Digital Learning Coach at Zionsville Community Schools
6. Zionsville’s MagicSchool implementation: ensuring educational equity through AI
AI promises to democratise access to education, ensuring that each student receives the support they need to fulfil their potential.
Zionsville Community Schools in Indiana took a comprehensive approach to AI integration, implementing the MagicSchool platform to simultaneously reduce teacher workload and build student AI literacy within a controlled environment. Rather than ban AI tools or allow unrestricted access, the district created what they call a ‘walled garden’ where students can learn responsible AI use while developing skills essential for future careers. The phased rollout prioritised teacher adoption first, with educators initially using AI to streamline lesson planning, rubric creation, and feedback generation during the first semester. Only after teachers became comfortable with the technology did the district expand access to students, who then used AI tools for instant writing feedback, historical character interviews, and differentiated learning experiences.
Levelling the playing field
The implementation produced unexpected educational moments that actually deepened the learning experience. During a sixth-grade World War II research project, students used MagicSchool’s character chatbot to interview historical figures. One student chose a German fighter pilot, and when the AI responded in German, it sparked a deeper classroom discussion about historical context, communication, and perspective. Similarly, fourth-graders studying the Revolutionary War engaged with AI representations of George Washington and King George III, gaining a nuanced understanding of opposing viewpoints.
Digital Learning Coach Kyle Beimfohr emphasises the equity dimension of AI access: “Some kids go home to parents who can help revise their papers. Others go home to an empty house,” Beimfohr explained. “We needed a way to give every student the same type of feedback and support.” The platform also includes built-in digital literacy prompts that teach students about AI limitations, bias awareness, and protecting personal information.
Learnings
The six implementations of AI in education we’ve examined in this article reveal that the sector is in the midst of an important transition period where traditional teaching methods increasingly intersect with emerging technologies. The diversity of approaches – from AI tutors to adaptive assessment systems – suggests that we’re still discovering which applications prove most valuable in different educational contexts. The results so far present a complex picture: encouraging gains in some areas, unexpected challenges in others.
As it stands, we don’t have a clear verdict on AI’s role in education just yet, but rather a series of questions that educators, policymakers, and communities must consider. What capabilities might we lose if we delegate too much instruction to algorithms? What new skills might students develop when freed from repetitive practice to engage in more creative and collaborative work? How do we ensure that AI serves to reduce rather than amplify existing educational inequalities? The answers to these questions will vary across communities and cultures, but the implementations described here provide early data points in what will likely be a lengthy process of figuring out how AI can best serve our educational goals.
