The potential and expectations for artificial intelligence (AI) are high in health and care services, which is reflected in the national digitalization strategy and the National Health and Coordination Plan [1][2].The latter points out that AI can contribute to faster and more precise diagnostics, better decision support for healthcare personnel, simplified logistics, automation of administrative tasks, and enable citizens to better monitor their own health. AI will also make a significant contribution to sustainable development of our shared health service.
Development and use of artificial intelligence (AI) and especially large language models has seen significant growth in recent years. Use of large language models has the potential to improve both efficiency and quality in health and care services, for example through automatic text production, structuring of free text in patient records, machine-assisted health professional coding, plain language assistance, translation, speech recognition, knowledge support, health research, sorting and triaging patient inquiries, clinical chatbots (questions and answers), citizen-directed chatbots, health education, and decision support in treatment [3].
On the other hand, there are significant risks associated with adopting generative AI. This report describes important risks that the Norwegian health and care services must consider to be able to use AI systems based on large language models in a responsible manner. The risks relate both to inherent risks associated with this type of AI models and to their use in health and care services.
Currently, language models are best suited for linguistic and administrative tasks. Quality assurance is important regardless of use. AI systems with language models that are to be used to provide healthcare are most likely medical devices, and thus regulated through the act on medical devices.
Adaptation to Norwegian conditions is one way to reduce risks associated with the use of large language models in health and care services. The report describes what such adaptation entails, how adaptations can be made, what is needed, and who can make adaptations.
Finally, the report recommends measures that can reduce risks and facilitate use that is adapted to Norwegian conditions in health and care services:
- Establish a quality framework for large language models for Norwegian health and care services
- Build and share competence on development and use of large language models
- Establish common data foundation
- Investigate the need for infrastructure with computing power adapted to the health sector's needs
- Investigate governance structure for large language models in the health and care sector
Most measures will need to be solved in collaboration between a number of actors, such as the Norwegian Directorate of Health, Norwegian Health Network (NHN), the Research Council of Norway, the National Library of Norway, the higher education sector, regional health authorities, regional ICT companies, and KS. It is proposed that the Norwegian Directorate of Health take particular responsibility for establishing a quality framework.
[3] The Norwegian Directorate of Health has, among other things, written about opportunities with large language models in this report: https://www.helsedirektoratet.no/rapporter/store-sprakmodellar-i-helse-og-omsorgstenesta