Abstract:

A companion paper presented at INTED2026 reported on the interdisciplinary design rationale of SELBI (Supporting Early chiLdhood education in Blue skills with generative artificial Intelligence), examining how collaboration across marine science, early childhood education (ECE), and generative AI was enacted to negotiate disciplinary integrity, developmental appropriateness, and responsible system design. Building on that work and on a nationwide diagnostic survey (n = 137) revealing that ECE educators' familiarity with blue skills remains at floor level, this paper advances the project across three subsequent phases: knowledge-base curation, internal platform validation, and a structured educator training programme.

Following implementation of the locally hosted generative AI architecture, domain experts in marine science and early childhood pedagogy curated a repository of expert-verified documents aligned with Maltese curriculum and policy frameworks. The platform was evaluated through two iterative testing cycles using an 11-criterion analytic rubric spanning pedagogical appropriateness, content accuracy, and safety and inclusivity. Evaluation confirmed strong pedagogical suitability and content accuracy, with the second cycle achieving elimination of fabricated content; a finding that demonstrates the value of retrieval-augmented generation as a grounding mechanism for curriculum planning.

The most substantive contribution concerns the educator training programme, conducted across four sessions with four serving ECE educators from Kindergarten 1 to Year 2. The programme progressed from an introductory session in which educators shared their classroom realities and orientations towards AI, through collaborative workshops in which they generated and critically appraised planning suggestions, to a structured final evaluation. Participants ranged from a newly qualified teacher balancing AI-generated suggestions with emergent classroom realities, to an experienced practitioner who described herself as a sceptic with no prior AI engagement. Educators tested the platform against authentic scenarios, including structuring sea-based learning experiences, adapting activities for children with sensory needs, and responding to children's unanticipated questions. They observed that generated content was adaptable across subjects, year levels, and ability levels, whilst identifying limitations such as outputs defaulting to younger developmental levels and constraints in Maltese language generation. Educators' prompting practices were also shaped by assumptions from commercial AI tools, highlighting AI literacy as a necessary component of professional development rather than a skill that can be presumed.

The final evaluation, structured around five validity dimensions, reinforced these findings. Educators reported that the platform supported children's understanding through learning invitations, play-based engagement, and cross-curricular experiences, with one reflecting that it revealed possibilities for integrating blue skills not previously apparent. Professional scepticism proved educationally productive, strengthening alignment between outputs and classroom realities. The paper contributes to understanding what is required for AI-supported professional learning to be educationally defensible in early childhood, with implications for educator training, curriculum development, and the responsible design of AI tools for sustainability education.

Keywords:

Early childhood education, professional learning, education for sustainability, generative artificial intelligence, blue skills, curriculum planning.