MESIAS: The New Compass for Assessing AI Ethics?

Artificial intelligence is expanding at breakneck speed, but this growth raises concerns about reliability, safety, and ethical values. To tackle these issues, MESIAS was created – a web platform that evaluates AI projects using ethical principles and international guidelines.

MESIAS provides three main tools: a virtual assistant that guides users, adaptive forms that adjust to each project, and a monitoring dashboard to track outcomes. Its performance was tested in three phases: first, operational efficiency was examined; then, technology experts validated the method; finally, 52 professionals gave feedback. The results are striking: total assessment time dropped by 41.8% and required human resources fell by 40%. Expert acceptance reached 85%, while user satisfaction exceeded 90% for usability, content, follow‑up, and overall experience.

These figures show that MESIAS is more than a checklist; it’s a practical approach to make AI more responsible, especially in public settings where decisions affect everyone. With tools like this, developers and policymakers can make more informed, transparent choices, paving the way for a more ethical digital future.

MESIAS is built on an AI governance framework that incorporates UNESCO's seven ethical principles (humanity, transparency, justice, responsibility, sustainability, privacy, and security) and the European Data Protection Board's guidelines (GDPR). Its evaluation engine uses a multi‑criteria scoring algorithm, where each principle is weighted according to the application context (e.g., public surveillance systems vs. personal assistants).

MESIAS's adaptive forms employ fuzzy logic to handle partial or uncertain responses, allowing a more flexible assessment than traditional binary checklists. The virtual assistant is powered by a large language model (LLM) trained on ethical policy datasets, ensuring contextual answers and mitigation suggestions.

The monitoring dashboard aggregates temporal metrics (assessment time, number of iterations) and qualitative ones (compliance index, risk level). Data are displayed via heatmaps and radar charts, making it easy to spot critical areas.

During validation, the 41.8% reduction in assessment time was linked to automated data‑collection, while the 40% drop in human resources stemmed from eliminating redundant manual reviews. The 85% expert acceptance reflects the model's alignment with international best practices, and satisfaction above 90% indicates a user‑centric interface and a short learning curve.

In summary, MESIAS blends ethical principles, advanced scoring methods, and intelligent interfaces to provide a comprehensive, scalable AI reliability assessment, promoting more responsible decisions in both public and private sectors.