A Dubai academy is making AI the student

As artificial intelligence becomes more common in classrooms, its role has largely been to explain concepts, answer questions, and guide students through assignments. Dubai-based Crucifer Academy is experimenting with the opposite arrangement, requiring students to teach an AI avatar that asks questions and tests the clarity of their explanations

Known as “AI Kid,” the avatar questions students about concepts they have studied before taking an examination based on what it has learned. The initiative attracted attention in November 2026, when the academy outlined its proposed assessment model ahead of its first student cohort.

The experiment comes as educational institutions worldwide expand their use of artificial intelligence. In September 2025, Estonia introduced its AI Leap programme, initially targeting 20,000 secondary school students, while the United Arab Emirates incorporated AI into public school curricula from kindergarten through Grade 12 during the 2025–2026 academic year.

Yet the rapid adoption of AI has outpaced evidence of its educational benefits. Experiments such as Crucifer's highlight the need to determine which approaches improve academic performance, under what conditions, and how those gains can be reliably measured.

Turning students into teachers

Crucifer Academy, part of Dubai-based Crucifer AI Technologies, is developing four six-month training programs covering AI data engineering, machine learning, robotics and enterprise AI deployment, according to the company.

Central to the proposed curriculum is a seven-stage learning process combining instructor-led lessons, practical exercises, assessments, and AI-assisted teaching. After completing a module, students must explain its concepts to AI Kid, which the company says is designed to express confusion, request alternative explanations, and challenge incomplete reasoning. The avatar then takes an examination intended to test what the student has taught it.

Crucifer describes this process as part of its “Knowledge Guarantee,” under which course completion depends on demonstrated competence rather than attendance alone. According to the academy, human instructors will remain responsible for reviewing students’ conceptual understanding, practical work, and explanations before determining whether they can advance.

The program also incorporates an Applied Intelligence Lab, where students are expected to independently define technical problems and select an approach before using AI to assist with implementation. They must subsequently test their solutions, identify weaknesses, and justify their decisions. The company plans to limit live classes to 30 students and allocate an additional 30% of learning time to questions, reinforcement, and remedial instruction.

The science of learning by teaching

Although Crucifer's use of conversational AI is relatively new, the educational principle behind it has a considerably longer history.

In the early 2000s, researchers at Vanderbilt University and Stanford University developed Betty's Brain, a computer-based learning environment in which students teach a digital character about scientific subjects, including river ecosystems and climate processes. The program was publicly documented by Vanderbilt in March 2008.

Students construct diagrams illustrating relationships between different scientific concepts, which Betty subsequently uses to answer questions and complete quizzes. Another digital character, Mr. Davis, evaluates the answers and provides feedback, allowing students to identify mistakes in their explanations and correct their understanding.

Crucifer's system introduces a new consideration. Unlike Betty's Brain, which relies on information explicitly provided by students, modern AI models already possess extensive knowledge acquired during training. Assessing students through an AI avatar therefore requires distinguishing between the quality of their explanations and the knowledge the technology already possesses.

Can AI deliver better results?

The growing use of AI in education has also prompted researchers to compare its effectiveness with established teaching methods, although evidence from conventional AI tutoring cannot directly establish the benefits of teaching an avatar.

A randomized controlled experiment published in Scientific Reports in June 2025 examined 194 undergraduate physics students at Harvard University. Researchers compared a specially designed AI tutor with classroom active learning across two physics lessons, measuring students’ understanding before and after instruction.

Students using the AI tutor achieved more than twice the median learning gains of those receiving classroom instruction, while completing the material in less time. They also reported greater engagement and motivation. The system incorporated guided explanations, targeted feedback and opportunities to work through problems independently, rather than simply providing unrestricted chatbot access.

These considerations are particularly relevant as countries pursue AI education at a national scale. Estonia's AI Leap programme, for example, reported participation from 154 of the country's 156 upper-secondary schools during the 2025–2026 academic year, involving approximately 20,000 students and 5,000 teachers. Its learning application, ITI, uses questions and explanations to guide students toward answers rather than immediately providing them.

Crucifer's first cohort could provide an opportunity to examine a different approach, provided its outcomes are measured against comparable teaching methods. As artificial intelligence becomes more capable of completing academic tasks, its educational value may depend on how effectively institutions use it to strengthen the reasoning students must still perform themselves.