Name in original language
Uso de ALIA (Modelo Fundacional de Lenguaje en Español y Lenguas Cooficiales) en agentes inteligentes de asistencia al ciudadano en Entidades Locales
Initiative overview
ALIA is Spain’s public and open AI infrastructure for Spanish and co-official languages. This ensures digital sovereignty and provides language models and related resources that public administrations, researchers and companies can use to build trustworthy AI services. Currently, every public administration that has implemented an AI-based assistant has done so by using a proprietary model (for example, ChatGPT), which not only costs money but also poses the potential risk of sending private data to companies whose values might not be aligned with public service. In order to test the usefulness of ALIA models in real-life scenarios, this initiative involves close collaboration with local entities to replace their current AI assistants with ALIA-based ones. If the project succeeds, public entities and citizens alike will benefit from a better, personalised service maintained by a free, public and sovereign infrastructure fully aligned with public service values. The project’s long-term goal is to keep making ALIA as robust and complete as possible so it can end up being used not only in every public AI project but also by companies and citizens.
As an open-source "building block", ALIA also empowers the domestic economy. It allows SMEs, researchers, and developers to build specialised applications on a secure, sovereign foundation. This strategy prevents vendor lock-in, ensures that the economic and strategic value of AI remains within the national ecosystem, and guarantees that the AI tools used by the State are aligned with national values and administrative reality.
Results, outcomes and impacts
ALIA has established a secure foundation for AI in Spain, delivering publicly available language models and an open resource hub (ALIA Kit) with models, datasets, documentation and integration guidance for Spanish and co-official languages. Tests submitted by local entities where ALIA has been used report that the public model, while not as effective as the model it replaced, almost reaches the same number of correct answers in many cases, demonstrating that there is significant potential for the future of ALIA models. Entities have expressed their interest in continuing to use ALIA as their primary model for citizen assistance from now on.
