Assessing the Viability of On-Device Language Models in Legally Controlled Measuring Instruments |
|---|
| Mahbuba Moni, Daniel Peters, Florian Thiel, Axel Sikora |
- Abstract:
- The European Union's Artificial Intelligence Act represents a significant regulatory framework that governs the deployment of AI systems across various sectors, including critical infrastructure. A crucial challenge identified is the integration of AI in measuring instruments, especially utility meters used in critical infrastructures, as highlighted in the EU AI Act's Annex III. In legal metrology, AI applications, especially language models, are primarily server-based, requiring substantial resources and continuous connectivity. The integration of AI and specifically of language models into measuring instruments promises unique opportunities and challenges along their lifecycle, impacting manufacturers, notified bodies, market surveillance authorities, and consumers. This paper highlights the need for on-device language models that can perform real-time regulatory compliance checks, facilitate natural-language interactions, and generate legally admissible documentation. We also address the challenges of designing smaller, energy-efficient models and propose a system overview to enhance functionality in resource-constrained environments.
- Download:
- IMEKO-TC8-11-24-2025-036.pdf
- DOI:
- 10.21014/tc11-2025.036
- Event details
- IMEKO TC:
- TC11
- Event name:
- IMEKO TC8, TC11 and TC24 Conference
- Title:
Joint conference of the TCs ‘Traceability in Metrology’ (IMEKO TC8), ‘Measurement in Testing, Inspection and Certification’ (IMEKO TC11), and ‘Chemical Measurements’ (IMEKO TC24).
- Place:
- Torino, ITALY
- Time:
- 14 September 2025 - 17 September 2025