People are more willing to accept decisions made by an artificial intelligence if a human is in the loop, according to a survey carried out in Germany.
Automated decision-making is already used to shortlist candidates for jobs, determine appropriate sentencing in court cases and even in – occasionally with problematic or controversial results, because artificial intelligences can display the same biases that exist in the data sets that are used to train them.
To assess how people viewed such processes, at the University of Mannheim in Germany and his colleagues surveyed almost 4000 individuals through a polling service that aims to provide a representative cross-section of the country’s population. The team asked what people thought about the fairness and acceptability of automated decision-making in a range of contexts, such as banking, criminal justice and employment.
Advertisement
The respondents said a hybrid approach where an AI informs a human decision-maker, rather than making the decision alone, is fairer than entirely automated decision-making, but only roughly as fair as solely human decisions.
For example, about 30 per cent of people said an automated decision on banking wasn’t at all fair, whereas around 20 per cent said that about human or human-assisted decisions. And 55 per cent of people thought decisions made by humans and AI together in banking contexts were fair or very fair, and less than 20 per cent said they were not at all fair.
Context seemed to be key: people were more wary of AI decisions when the stakes were higher, such as in applying for jobs or court sentencing, than they were in decisions in banking.
When it came to acceptability, the respondents also perceived a combination of human and algorithmic decision-making as being as good as decisions made by a human, and more acceptable than entirely automated decisions.
Kern says that despite the desire of respondents to maintain a human in the loop, he isn’t pessimistic about the future of AI decision-making, because the field is working on fixing its issues around bias. He also points out that humans have their own biases that can lead to unfair decisions.
“Humans could also make things worse,” he says. “I mean, it can go in both directions. They can bring in their own biases and kind of ‘correct’, or in their view correct, a fair system and kind of move it back to where we started off. So a combination [of humans and AI] might be a good choice.”
Patterns
