Учёные Aalto University создали ИИ-модель чтения для AR и персонализации текстов

Researchers at Aalto University (Finland), together with international partners, have created the most accurate computational model of human reading to date. The new system uses reinforcement learning — a type of AI from the world of robotics — to reproduce and explain the decisions a reader makes while moving through a text.

Why is "human-like" reading so difficult to model?

Eye movements during reading are among the most complex cognitive processes to model: a person makes dozens of unconscious decisions per second. The researchers reproduced three key types of such movements: saccades (rapid shifts of gaze between words), fixations (pauses for information processing), and regressions (backward movements when a passage is not understood).

Previous models were built on statistics and fixed psycholinguistics rules — and struggled with individual differences between readers. The RL agent works differently: it learns through trial and feedback, gradually forming an adaptive strategy sensitive to context — word length and frequency, syntactic complexity, and the reader's goal.

• Developer: Aalto University (Finland) together with international partners
• Method: reinforcement learning
• Subject: saccades, fixations, and regressions of the eyes during reading
• Result: the most accurate computational reading model as of August 2026
• Applications: text personalization, augmented reality, accessibility, and diagnostics

"The new model uses reinforcement learning — a type of AI from the world of robotics — to explain and reproduce the decisions that readers make while moving through a text," according to the

Aalto University research description.

What will the new model be used for?

Content personalization: if the system knows how a specific reader perceives text, it can dynamically adjust sentence length, formatting, and emphasis. For educational platforms, this affects knowledge retention; for media, it affects depth of engagement and reader retention.

Augmented reality: AR glasses and headsets need to know exactly where the user is looking in real time in order to overlay information precisely in focus without delay or offset. Hardware eye trackers are accurate, but expensive and power-intensive. A predictive model at the software level reduces dependence on specialized hardware.

Accessibility and diagnostics: people with dyslexia or attention deficits have reading strategies that systematically differ from the norm. An accurate model can identify these differences and help design text interfaces that are easier to read for such users.

According to the Aalto University research description, applying RL made it possible for the first time to reproduce the actual decision-making mechanism in reading — rather than an averaged pattern of eye movements from a dataset.

What this means

Academic reading models have existed since the classic psycholinguistic studies of the 1980s. The shift to an RL-based approach changes their nature: from descriptive, they become explanatory and potentially individually adaptive. For AR interface developers, EdTech platforms, and publishers, this is a new personalization tool that requires no expensive sensors. An open question remains: how well the model generalizes across different languages, writing systems, and text types.

Frequently Asked Questions

What is reinforcement learning in the context of reading?

Reinforcement learning is an AI method in which an agent learns through trial and error, receiving feedback from the environment. The same technology is used in robotics and training game-playing agents; Aalto University researchers applied it to modeling how a reader makes decisions while moving through a text.

How does the new model differ from previous ones?

Previous models produced averaged predictions based on statistics and fixed rules. The new one reproduces an adaptive decision-making mechanism — context-dependent and closer to the behavior of a specific individual rather than an averaged group of test subjects.