AI With a Calculator: Higher School of Economics Tests a New Way to Assess Math Skills
The Higher School of Economics (HSE) in Moscow has combined an adaptive algorithm, psychometric models and generative AI in a new system: the algorithm selects questions, psychometrics helps identify individual skills, and AI generates exercises and feedback. The project could significantly change how schools assess students' abilities.
Researchers from HSE's Center for Psychometrics and Educational Measurement at the Institute of Education, working with specialists from the Faculty of Computer Science's Institute of Artificial Intelligence and Digital Sciences, have tested an adaptive system for assessing elementary school students' mathematical literacy. The study involved 284 third- and fourth-grade school students from Moscow and Novosibirsk.
Different Questions for Different Students
The project's goal is to identify which skills a child has already mastered and which still need development. Three technologies work together in the system: an adaptive algorithm, psychometric models and generative artificial intelligence. The pilot showed that this approach can be used in everyday school practice.
The system is built around an adaptive algorithm. It analyzes a student's previous answers and selects the next question to gather as much information as possible about the student's skills. As a result, children may receive different numbers of questions, with difficulty levels tailored to their performance. In any case, the test ends only when the system has collected enough information to make a reliable assessment.
AI Generates the Questions, Psychometrics Measures the Skills
Psychometric models assess not only a student's overall level of mathematical literacy but also its individual components, known as cognitive attributes. For example, the system can determine whether a child can recognize patterns. AI has a different role. “AI helps build a question bank and generate clear, actionable feedback for teachers. Bringing these technologies together reduces manual work and gives educators a detailed picture of which student skills need attention,” says Danil Shvetsov, project lead in the Applied Technology Solutions Department at HSE's Institute of Artificial Intelligence and Digital Sciences.
Teachers noted that receiving different numbers of questions initially surprises students. Over time, however, they come to understand that this is how adaptive testing works.
But HSE's project stands out for more than its adaptive approach. “We brought together our colleagues' psychometric methods, an adaptive algorithm and generative AI in a single system,” Shvetsov emphasizes. His colleague Elena Kardanova, the project's lead and academic director of the Center for Psychometrics and Educational Measurement at HSE's Institute of Education, adds: “This is a good example of AI being put to beneficial use in education. It makes it possible to do things that would otherwise be impossible.”
AI for Good
Every component of HSE's new project – the adaptive algorithm, psychometrics and generative AI – has already been tested in research projects and school services. In 2022, specialists from the Moscow Aviation Institute and Plekhanov Russian University of Economics proposed using neural networks to select both the topic and difficulty of a question based on a student's previous answers and the time taken to respond.
In 2023, researchers at ITMO University introduced a system based on fuzzy logic that took into account both the proportion of correct answers and the number of attempts. That same year, a study involving 4,012 students from 19 Russian regions found that adaptive math learning could help students prepare for the State Final Assessment (GIA). A year later, in 2024, Moscow's Tsifrovoy uchitel (Digital Teacher) service became available to students in grades 5 through 9. By October, more than 173,000 students had used it, and the service had reached approximately 850,000 users over the course of the year.
China offers another example of how such services are being deployed internationally. In 2026, the country officially demonstrated how AI can be used to grade homework. Mao Ning, a spokesperson for China's Ministry of Foreign Affairs, presented a video showing an AI system scanning handwritten assignments, recognizing even difficult handwriting, identifying errors and assigning scores.
Students and Teachers Have the Final Say
The Russian government is actively supporting the development of educational technology while seeking to preserve the teacher's central role. In September 2026, Minprosveshcheniya (the Ministry of Education) issued recommendations on using AI in schools, again emphasizing that teachers have the final say. In October, the Russian government announced the launch of more than 40 pilot projects in education and science. These include digital assistants for teachers and tools for analyzing learning outcomes.
The market is growing alongside the expansion of digital education. During the first nine months of 2026, revenue at Russia's 100 largest EdTech companies reached 119 billion rubles (about $1.45 billion), up 12.1% from the same period a year earlier. This growth is likely to continue over the next few years, particularly as digital solutions become more diverse and play a larger role in the learning process.
The team behind the math-focused AI project has already identified several ways to expand it. These include broadening the scope of testing, adding new subjects, adapting the system for other age groups and moving beyond simple knowledge assessment toward personalized learning. The last step is arguably the most important. The platform could then do more than identify learning gaps: it could also select exercises designed to help students close them.
