Svensk hälso- och sjukvårds kunskap kring bias i AI- verktyg för beslutsstöd : Framtidens AI- teknik utan fördomar, mot en jämlik vård

Sammanfattning: With an increased availability of healthcare data and a rapid development of analysis techniques, artificial intelligence (AI) is currently contributing to a paradigm shift in healthcare. Areas that were previously considered only possible to be managed by human experts are now increasingly being managed with the help of AI. A large part of the research in this field points to the positive aspects that the introduction of AI in healthcare can bring. What is less often highlighted is the need for relevant data to ensure that the AI solutions generate results that promote equality. In order for this to happen, it is required, among other things, that the algorithms in the AI solutions are trained on neutral and objective data. The aim of this study is to use an exploratory design to investigate existing knowledge about bias in AI tools for decision support among clinical, Swedish healthcare personnel. This has been done via interviews with representatives from Swedish healthcare who work with AI. In the interviews, the status of knowledge about how data is created, implemented and used in AI for decision support was examined. The awareness of how this affects the results obtained in the form of bias and what is done to counteract this was investigated. The analysis of the respondents' answers showed that there is an indirect image of knowledge regarding bias and its consequences, but that this is not discussed in the terms that the research uses. Various different actions are performed that prevent the emergence of bias. Distinctive about the result, however, was that the respondents in many cases did not describe the motive for these actions. Actions seem to be carried out but this without fully knowing to what benefit.