Artificial intelligence (AI) is already an integral part of many everyday applications, from search engines to loans. How decisions are made is often unclear. At a press conference on Thursday, the Austrian Chamber of Labour (AK) presented a study by the Institute of Technology Assessment (ITA) of the Austrian Academy of Sciences. The authors ask how comprehensible artificial intelligence is and demand more transparency to enable the foresight of negative effects or also – to settle damages already done.

EU efforts fall short

"The EU is currently regulating the legal framework for an AI law, but is forgetting about the consumers", criticises AK consumer protector Daniela Zimmer. "AI can be useful, but it can also be socially explosive, for example if a numerical value calculated by AI reduces life chances. Therefore, consumers need protection against the erosion of their fundamental rights and freedoms, against intransparency, discrimination and other risks of harm posed by analytics software," says Zimmer, and calls for an amendment of the EU draft for a new AI law.

Project leader Walter Peissl, who co-wrote the study together with ITA researchers Daniela Fuchs and Titus Udrea, is concerned about the uncertainty surrounding the spread of AI systems: "It is not clearly defined what AI includes. It is not just a matter of a too narrow - purely technical - definition that would leave many applications unregulated. The interests of consumers are neglected in both the development and regulation of AI. However, transparency about the algorithms used is necessary and also possible. If these requirements are not met, the use of AI should be avoided, because the responsibility must remain with the people," says Peissl.

Social effects of AI just as important as economic success

Above all, it is important to understand that every technology also leaves its mark on society: "AI decisions that are not transparent or perceived as unfair can divide society and endanger democracy," stresses Zimmer. One of numerous recommedations given in the study is to generally register AI systems whose decisions affect humans and to certify them according to expected risk. That would strengthen consideration of consumer interest, says Peissl.

"We have to ask ourselves how and with what consequences AI is used in everyday life, because our social interaction with each other is being massively reshaped by it," Peissl emphasises. As examples, he mentions credit scores when granting loans or automated application procedures as well as microtargeting activities in social networks. "We must therefore always consider AI in a social context, and this must also be laid down as a law."