Abstract:

The adoption of Generative Artificial Intelligence (GenAI) in higher education has prompted heterogeneous responses at both international and national scopes. Most universities have taken a predominantly reactive stance, prioritizing basic training for academic staff (teaching and research personnel) over students, while focusing institutional concerns on plagiarism risk, academic integrity, and a perceived reduction in student effort. By contrast, a small number of pioneering institutions are pursuing deeper adoption, recognizing that any strategy based on restricting, preventing, or sanctioning the use of these tools is ultimately unsustainable.

This work argues that an institutional fixation on detecting AI use via anti-plagiarism tools is a dead end. Automated detectors exhibit high false-positive rates, and ongoing technological advances are making generated text increasingly indistinguishable from human writing. Instead, universities need a large-scale institutional effort to define clear, course-specific rules for GenAI use and to transform assessment practices—moving away from memory-based testing toward interdisciplinary authentic assessment approaches that foster metacognitive capabilities such as critical thinking and learner autonomy, as well as student-centered learning. This shift must be accompanied by comprehensive AI literacy that positions students not only as users, but also as critical co-creators, for instance, through systematic, critical verification of AI-produced outputs.

However, university-wide AI adoption is not solely a pedagogical challenge, otherwise it is also organizational, technical, and budgetary. At the organizational and technical level, deploying AI creates substantial challenges in governance, data protection, and infrastructure. From a governance perspective, many solutions require tenant administration, security and compliance management, differentiated licensing strategies, and mechanisms to monitor the cost of token-based or metered consumption models. Regarding data protection, the university community needs literacy in the terms of use of major AI tools and the ability to discern when to rely on commercial solutions from large providers versus local solutions built on smaller language models. Finally, as the literature highlights, providing AI solutions to the entire university community demands significant economic, technical, and human resources. Scaling Pro/Plus licenses (e.g., Microsoft 365 Copilot, ChatGPT Plus, Gemini Advanced, Claude Pro) to all academic staff and/or students faces prohibitive financial barriers for public universities, increases technological dependence on large corporations, and adds administrative and economic challenges associated with token-based pay-per-use API models.

Within this context, the Universidad Politécnica de Madrid (UPM) proposes a strategic and sustainable adoption model. Building on prior experience in high-performance computing (HPC), UPM is advancing a hybrid architecture that combines corporate large language models (LLMs) with in-house developments based on small language models (SLMs). This approach aims to optimize cost control and data privacy while leading a pedagogical integration in which AI functions as an enabler of student autonomy and educational innovation.

Keywords:

Generative Artificial Intelligence, Higher Education, Digital Governance, AI Literacy, Educational Innovation.