AI judges a real possibility, but discrimination a risk, Leiden professor says

Computers and artificial intelligence could make excellent judges in the near future, as long as there are plenty of safeguards in place against discrimination, professor Jaap van den Herik of Leiden University said to newspaper AD. 

A major advantage of an AI judge is that computers are very good at searching through old lawsuits. Human judges also look at previous convictions when making a ruling, to take the suspect's criminal record and rulings on similar cases into account. But where human lawyers can find around 80 percent of relevant cases computers can find 98 percent, according to the professor.

The big problem with an AI judge is that they can take over old mistakes made by human judges. There are old cases in which the ruling was "literally colored", Van den Herik said. On average, people of color were more severely punished than white people convicted of similar crimes. A computer will discover this pattern and may even build on it.

A ruling always mentions the place of birth of the suspect, for example. And with punishments for people born in Curacao usually being higher than people born in the Netherlands in the past, the AI may assume it is appropriate to give people born in Curacao higher punishments. 

But with enough safeguards in place, discrimination like this can be avoided, Van den Herik believes. Hard work is currently being done on algorithms that can recognize these errors and ensure that skin color or origin do not contribute to the punishment. "I think that will work, and then in the future people could be tried by a computer," he said to the newspaper.