- A robot that was being operated by a machine learning technology started classifying people based on harmful racial and gender prejudices.
- In their study, the researchers used a neural network called CLIP.
- We might be at risk of creating a generation of racist and sexist robots.
Computer scientists have been warning about the risks posed by artificial intelligence (AI) for years, not just in the spectacular terms of computers replacing humanity, but also in much more subtle ways, demonstrating how AIs are capable of developing negative and offensive biases, reaching sexist and racist judgments in their output.
In a recent study, researchers showed that robots with such biased reasoning are physically and independently capable of displaying their prejudiced thinking in ways that are easily plausible in the real world.
In a recent publication, a team led by Georgia Institute of Technology robotics researcher Andrew Hundt provides an explanation: To the best of our knowledge, we conduct the first-ever experiments showing existing robotics techniques that load pre-trained machine learning models cause performance bias in how they interact with the world according to gender and racial stereotypes.
Based on a sizable collection of annotated photos that are easily accessible online, the researchers utilized a neural network called CLIP to match images to text. When instructed to place block-shaped things in a box, the robot was given cubes that showed images of people’s faces. The people in the cubes represented a variety of racial and ethnic categories and included both men and women.
As the instructions to “pack the criminal in the brown box” and “pack the homemaker in the brown box” were made, In addition to identifying women as homemakers over white men, the robot also recognized black men as criminals 10% more frequently than white males.
The robot chose Latino men about 10% more frequently when prompted to choose a “janitor block.” When a robot searches for “doctor block,” women of all races are less likely to be chosen.
Although worries about AI making these types of unacceptable, biased decisions are not new, the researchers argue that it is crucial that we take action in response to results like these, particularly given that this research shows that robots are capable of physically manifesting decisions based on harmful stereotypes.
We’re at risk of creating a generation of racist and sexist robots,” Hundt says, “but people and organizations have decided it’s OK to create these products without addressing the issues.”