They’re often portrayed as the future of transport, but self-driving cars may be racist , a worrying new study has revealed.
Researchers from the Georgia Institute of Technology have conducted new research into how good self-driving cars are at spotting people of different skin tones.
In the study, the team applied the Fitzpatrick scale to a dataset of images, which classified them based on their skin tone.
The team then fed these images into object-detection systems used in self-driving cars, to see how often the people were detected.
The researchers also accounted for the time of day, lighting and other objects in view.
Worryingly, the results revealed that on average, the systems were 5% less accurate when detected people with dark skin.
The researchers highlighted that the issue is likely to lie in the way that the algorithmic system is trained.
Read More
Autonomous vehicles
-
Self-driving BAR delivers cocktails
-
Jaguar builds trust in self-driving cars
-
10 years away from self-driving cars
-
Hyundai reveals WALKING car concept
These systems learn based on examples that they’re fed, so it may be that during the learning stage, not enough images of dark-skinned pedestrians were shown.
In the study, published on arXiv , the researchers, led by Benjamin Wilson, wrote: “We hope this study provides compelling evidence of the real problem that may arise if this source of capture bias is not considered before deploying these sort of recognition model.”
Source: Read Full Article