Abstract:

The integration of AI into the field of education is problematic. UNESCO urges policymakers to be aware that AI integrations might increase inequalities and unfairness, risking an increase in the global digital divide. The city of Helsinki banned the use of AI in elementary school citing increased concerns of data protection.

This paper argues against bans, promoting a proactive approach to building competence among practitioners as well as the use of local LLMs for strengthening digital sovereignty and ensuring integrity while enabling AI integration in educational institutions.

The paper presents concerns about dataset transparency and data stream integrity, both posing unacceptable risks to data privacy and control in commercial cloud AI services. These risks conflict with the GDPR as well as EUs artificial intelligence regulations. Furthermore, we argue that there is an inherent ethical and pedagogical risk in using models that are trained on opaque sets of data containing bias in the educational context.

The paper includes a critical review of the performance of local LLM models versus commercial cloud based LLMs, including an evaluation of implementing AI on edge. The results indicate that the performance of modern local LLM models is comparable to commercial models. In addition, the results show that the use of local models presents additional benefits to digital sovereignty and data control.

Pedagogically, local open models can be promoted as objects of improvement and further study instead of simply being external passive tools. This proactive approach for both students and employees in educational institutions promotes AI literacy and serves as a pragmatic way forward.

Keywords:

Local LLM, AI literacy, Digital sovereignty, Education.