“The risk categorisation in the AI Act is opaque and difficult to understand for most people. It would make sense to involve members of the public and make adjustments,” recommends the researcher, citing two examples: “Automated fatigue detection in lorries directly intervenes in driving behaviour and processes potentially sensitive biometric data. EU law therefore prescribes strict protective measures.” For the people surveyed – as well as for many lorry drivers – however, the benefits for road safety would be the main priority. “The legal requirement is consequently seen as excessive and a hindrance,” adds Thalmann.
Conversely, many customers feel uncomfortable when they are expected to interact with chatbots on various internet portals. “Although these do not request sensitive data or intrude on personal spheres – and are therefore classified as low-risk under the AI Act – consumers have data protection concerns and fear manipulation or false information,” the researcher found. “Policymakers should therefore place greater emphasis on providing clear information and raising awareness, for example through easily understandable labelling of AI risks,” recommends Thalmann. Furthermore, digital literacy needs to be strengthened: “Those who are familiar with technology are also better able to assess the risks posed by artificial intelligence,” the expert cites as a further finding of his study. People who perceive the technology as particularly risky are hardly persuaded even by additional protective measures.
The study was carried out by an interdisciplinary team from the Smart Regulation research area, in collaboration with Graz University of Technology. The findings have been published in the journal Computer Law & Security Review:
Hanna Kröll, Jürgen Fleiß, Magdalena Eder, Elisabeth Lex, Stefan Thalmann: *AI risk perceptions and the effect of AI Act Safeguards. Results from a representative population sample*