Algorithms Are Accepted – But Only If Humans Have the Last Say

When it comes to AI based decision-making, it is not the use of algorithms as such that is controversial, but the lack of human control. This is the finding of a new study by the data scientists Florian Keusch and Frauke Kreuter.

At the end of 2020, the Austrian Public Employment Agency AMS began to use an algorithm to tailor job offers and continuing education programs to the individual profiles of jobseekers. This led to a public outrage in Austria. Many were criticizing the process, because the algorithm was based on historical data and thus potentially put people that were already discriminated against on the job market at disadvantage. Being a woman, for example, reduced the total score, and being a mother reduced it even more. This could also have reduced the chances of reintegration into the labor market.

However, algorithms are not only common on the labor market, but also in the banking world, in human resources or in medicine – and the discussion about their use is highly controversial. The data scientist Florian Keusch, professor at the University of Mannheim, and Frauke Freuter, professor at Ludwig-Maximilians-Universität München, analyzed the acceptance of algorithm-based decisions. Their study shows that decisions in which humans are involved are considered to be fairer than the decisions of an algorithm alone.

“The results lead us to the conclusion that the use of algorithms without additional human control is seen as particularly problematic”, Keusch concludes. “Thus, it is not the use of algorithms as such that is controversial”, the professor at the University of Mannheim says.

For their study, which is part of the German Internet Panel (GIP), the researchers conducted an online survey among more than 4,000 people.

Ergebnisbericht der Studie


Prof. Dr. Florian Keusch
Lehr­stuhl für Statistik und sozial­wissenschaft­liche Methodenlehre
Universität Mannheim
Tel: +49 621 181–3214

Prof. Dr. Frauke Kreuter
Lehr­stuhl für Statistik und Data Science
Institut für Statistik
Ludwig-Maximilians-Universität München