Штучний інтелект і трансформація освітнього оцінювання: від перевірки відповіді до реконструкції мислення

Artificial intelligence and the transformation of educational assessment: from checking answers to reconstructing thinking

Modern standardised examinations are based on reducing knowledge to a result. This approach is historically justified from the point of view of mass appeal, scalability and administrative standardisation: the system must quickly check large arrays of answers and ensure formal objectivity. However, this model does not take into account the nature of mathematical and scientific thinking as a sequential logical process consisting of hypotheses, transformations, argumentative transitions, and verification of assumptions.

In the classical educational paradigm, the final result was considered representative of internal cognitive activity. However, in the context of the digital transformation of education, this correspondence is no longer obvious.

In the context of the development of large language models, a fundamental paradox arises:
the more accurately AI generates correct answers, the less informative result-centred testing becomes. If the system evaluates only the final answer, it is unable to distinguish between independent problem solving and algorithmic generation.

This raises a structural epistemological problem: assessment ceases to measure knowledge as a cognitive ability and begins to measure access to answer generation tools.

The author justifies the need to move from the model:

“correct answer = knowledge”

to the model:

“structure of reasoning = level of competence development”.

This is particularly relevant in STEM education, as it is the logical sequence of proof, justification of formulas, and cause-and-effect relationships that form the core of subject competence. Mathematical proof, physical explanation, or algorithmic transformation cannot be reduced to a final numerical result without losing meaning.

In this context, recording the thought process becomes crucial. The most natural form of such recording remains handwriting. Unlike printed or automatically generated text, handwriting reflects:

the pace of thinking,

the sequence of steps,

corrections and doubts,

the spatial arrangement of formulas,

the individual logic of the proof.

Handwriting analysis has not only a graphological but, above all, a cognitive dimension. The spatial organisation of the solution, the method of grouping transformations, the use of auxiliary symbols — all this forms the topology of thinking. It is this topology that can be the object of digital reconstruction.

Within the modern EdTech environment, it is possible to combine handwritten forms with automated data processing tools. The development of OCR systems for STEM (Optical Character Recognition with support for formulas, symbolic notation, and structural analysis) allows for:

automated recognition of handwritten formulas;

segmentation of mathematical blocks;

construction of symbolic dependency graphs;

analysis of solution topology;

detection of logical gaps;

evaluation of argumentation structure.

Unlike traditional testing, this model does not focus on the final number or formula, but analyses the path to it. AI acts not as a generator of answers, but as a tool for reconstructing the cognitive trajectory.

Handwriting analysis as a dynamic trace of thinking plays a special role in this process. Even in a static image of a manuscript, it is possible to detect:

returns to previous steps;

corrections of false hypotheses;

alternative attempts at solutions;

local areas of uncertainty.

These elements are markers of depth of understanding. They form a cognitive profile of problem solving and allow us to evaluate not only the correctness but also the quality of thinking.

The proposed assessment concept involves the integration of:

handwritten task completion;

STEM-OCR analysis;

structural verification algorithms;

human expert control.

This model combines human pedagogical experience with machine analytics, ensuring a balance between scalability and depth of assessment.

It has been shown that digital processing of handwritten text may be the answer to the crisis of confidence in online assessment. Paradoxically, in the age of generative AI, a return to written problem solving followed by algorithmic analysis allows for the restoration of transparency in thinking.

Thus, at the heart of the new paradigm is not the answer as a product, but thinking as a process. Educational technologies (EdTech) and OCR systems are becoming not tools for simplifying verification, but means of increasing the epistemic validity of assessment.

The proposed concept of process-centred assessment has already been partially implemented in the educational application “Mriya+”, which integrates modules for recognising handwritten STEM text, structural analysis of solutions, and reconstruction of reasoning logic. The platform combines handwritten task completion with algorithmic verification of the sequence of steps, allowing for the assessment of not only the final result but also the quality of the argumentation structure. This creates a practical basis for the implementation of an epistemically transparent assessment model in the digital educational environment.

Conclusions

• The current system of standardised assessment, focused on testing the final result, loses its epistemic validity in the context of thedevelopment of generative artificial intelligence.
• The correct answer is no longer a sufficient indicator of competence, as it can be generated without any real cognitive process.
  In STEM education, the key object of assessment should be the structure of reasoning, not just the final result.
• Handwritten solutions are a natural way to record the cognitive trajectory and allow the sequence of logical transitions to be reconstructed.
• The use of STEM-OCR, educational analytics, and algorithmic reconstruction of proof creates the basis for a process-centric assessment model.
  The future of educational assessment lies in combining human expert control and machine analytics, which ensures epistemic transparency and increases trust in assessment systems in the age of AI.