A bank hired a robot to guard an ATM — and here is what happened

Photo by Aideal Hwa on Unsplash

Are robots racist?

No.

This article, “Robots With Flawed AI Make Sexist And Racist Decisions — Experiment Shows”, is a fascinating read.

One of the inferences of the article is that, if a robot is appointed to guard an ATM at night, it may not let customers belonging to a certain demographic profile use the ATM.

There are potentially a few things to think about:

1. Why did the robot display racist behavior? The training data that the robot was trained on should have had a bias.
2. Why did the training data have a bias? Is it indicative of the training data that is available in the public domain or is it indicative of the team that put together the training data? In the said article, the training data did not seem to have an explicit objective to be racist. The robot made some inferences, which means racism was implicit in the training data.
3. Why was racism implicit in the training data? Here is where the issue is. The mass media follows the Pareto Principle. 80% of the mass media content is created by 20% of media organizations.

Unknowingly, the robots tend to pick up on that implicit bias.

When that happens, your ability to withdraw money from a robot-guarded ATM is dependent on your facial features and demographic profile as how the 20% has stereotyped you.