When a computer algorithm treats us unfairly, we get angry and take revenge almost as we would on a living offender. This is the conclusion of a group of psychologists based on a series of experiments published in the Journal of Experimental Psychology: General. However, the feeling of restored justice after punishing a machine does not arise in all cases.
The researchers Alexander Eng of the Asia School of Business in Kuala Lumpur and Kai Chi Yam of the National University of Singapore studied how people react to unfair decisions by automated systems. The basis is the concept of mind perception: in everyday life we assess whether another being has the capacity to think and act intentionally (agency) and the capacity to experience pain, fear, joy (experience). In humans these qualities are linked: an adult can both cause harm and feel punishment. In machines, according to people, the first is present, while the second is almost absent.
In a pilot experiment, 380 adults from the US played the "ultimatum" game, an economic game in which a partner proposes dividing a small sum and the participant can accept or refuse. Refusal means that no one gets anything, that is, it is a price paid for punishment. When the partner was an algorithm and offered an unfair share (5 pence out of 50), participants' anger averaged 3.12 points out of 5 versus 1.31 for a fair offer. They were as angry at the algorithm's unfairness as at a human's (3.12 versus 2.91) and rejected its offer in 57.8% of cases, roughly the same as a human's offer (62.2%).
In the first main experiment (123 participants), all received an unfair offer from a human or an algorithm. Those who refused (78 people) reported a higher sense of justice, 3.57 points out of 5, than those who accepted (2.63). The gap was larger when punishing a human than an algorithm. In the second experiment (739 people), participants could not change partners, and algorithms and humans competed for the right to stay in the game. People attributed less capacity to feel to algorithms (1.86 versus 3.20), but rejected unfair offers equally often (71.0% and 65.4%). Those who punished felt more justice (3.53 versus 2.71), and this link was the same for machines and humans.
The third experiment (735 people) with a "trust game" produced an unexpected result: the more strongly participants recommended removing the dishonest partner, the lower their sense of justice. Possibly because the recommendation did not lead to actual punishment. In the fourth experiment (3,684 people), after refusing an unfair offer, the partner could send a message: expressing regret, promising to be fairer, or both. Messages from the algorithm increased the sense of justice (3.57 and 3.62 points) compared with silence (3.42). For humans the difference was insignificant.
An additional experiment (1,093 people) tested whether people vent their anger on an innocent party. It turned out that those who punished their partner shared slightly more generously with a stranger than those who did not, regardless of who the offender was.
The authors emphasize limitations: all games were online, the stakes were symbolic (50 pence), and participants themselves decided whether to punish. In real situations, for example when a loan is denied or someone is fired, the effects may differ. In addition, the results on the sense of justice were mixed: in one experiment there was a difference between machine and human, in another there was not. The scientists also note that their scale for measuring the sense of justice showed low internal consistency, and call for more precise instruments to be developed. According to Eng, people are willing to pay to punish an algorithm that can neither receive money nor experience loss.