Abstract:

This study examines the integration of a structured observation-based tool within a human-centered system designed to identify early learning strengths and difficulties in primary school children. While standardized assessments often fail to capture subtle, context-dependent indicators of learning vulnerability, early signs typically emerge through observable behaviors during everyday learning activities. This work builds on observation-based approaches grounded in ecological validity and developmental theory, addressing the gap between theoretical constructs and their practical observability in educational settings.

The proposed framework translates complex developmental processes into standardized, observable indicators that can be reliably documented in classroom contexts. These indicators are embedded within structured learning activities, particularly robotics-based workshops, which provide rich, iterative, and ecologically valid environments for eliciting cognitive, motor, language, and socio-emotional behaviors. Such contexts enable repeated observations of meaningful learning processes while maintaining pedagogical relevance.

The observational tool constitutes the first layer of a multi-layered system combining structured data collection, expert interpretation, and artificial intelligence (AI)-assisted reporting. Observations are recorded using standardized Likert-type scales, ensuring consistency across practitioners and contexts. A neuropsychologist contributes a second layer by contextualizing behavioral patterns, identifying potential developmental vulnerabilities, and ensuring that interpretations remain grounded in developmental science. This human expertise plays a critical role in transforming raw observations into meaningful and clinically informed insights.

The resulting structured dataset is then integrated into an AI-assisted system that supports the generation of individualized profiles and pedagogical recommendations. Importantly, the system operates under strict human supervision and does not automate decision-making. Instead, it supports professional judgment by synthesizing structured, interpretable data into coherent outputs for teachers and parents. This human-in-the-loop configuration aligns with current recommendations for responsible AI use in sensitive domains.

The findings highlight the central importance of data quality, structure, and theoretical grounding in enabling meaningful AI-supported applications. The observational tool acts as an interface between real-world behavior and computational analysis, making it possible to generate reliable and interpretable outputs. In this sense, the effectiveness of the system depends less on algorithmic complexity than on the rigor of the input data and the integration of expert knowledge.

Overall, this study demonstrates how structured observation, combined with domain expertise and AI, can produce comprehensive and interpretable representations of child development. It contributes to the development of human-centered educational systems by proposing an integrated framework in which observation, expertise, and AI are iteratively refined. Future work should further evaluate the robustness of this approach across contexts and explore how AI can continue to support transparent, ethical, and scientifically grounded educational decision-making.

Keywords:

Structured observation, artificial intelligence in education, early detection, learning difficulties, robotics in education, pedagogical recommendations, developmental psychology.