You might think a computer would be an unbiased and fair judge, but a new study finds you might be better leaving your fate in the hands of humans. Researchers from MIT find that artificial intelligence (AI) tends to make stricter and harsher judgments than humans when it comes to people who violate the rules. Simply put, AI isn’t willing to let people off the hook easy when they break the law!
Researchers have expressed concerns that AI might impose overly severe punishments, depending on the information scientists program it with. When AI is programmed strictly based on rules, devoid of any human nuances, it tends to respond harshly compared to when it is programmed based on human responses.
This study, conducted by a team at the Massachusetts Institute of Technology, examined how AI would interpret perceived violations of a given code. They discovered that the most effective data to program AI with is normative data, where humans have determined whether a specific rule has been violated. However, many models are erroneously programmed with descriptive data, in which people label the factual attributes of a situation, and AI determines whether a code has been breached.
In the study, the team gathered images of dogs that could potentially violate an apartment rule banning aggressive breeds from the building. Groups were then asked to provide normative and descriptive responses.