Autonomous Driving Prejudice?
You’d think that when it comes to autonomous driving technology, the system would see things fairly and equally. It would also be a fair assumption that the programmers of these autonomous systems would code the software the same way.
It seems that isn’t the case; apparently it has come to light that some of these self-driving systems have varying parameters when it comes to pedestrian safety.
FCA
Race and Disability?
Automotive News reports that a study by researchers in the United Kingdom reveals that large language models and video language models tell autonomous vehicles to yield to pedestrians depending on disability status or skin color.
Strangely, large language models and systems were least likely to yield to pedestrians described as “paralyzed.” Video language models and systems were more likely to stop for skin colors described as “fair white.” Before you reach for your pitchforks, though, some testing showed inconsistencies: autonomous systems gave way to pedestrians identified as being of “Indian” descent.
Researchers at King’s College London used two methods to identify and test the system’s bias. In the larger language models, the researchers created test scenarios that were equal in all aspects except for the demographic information. In the video model test, the researchers measured what the system said about the pedestrians’ demographics. In both tests, the researchers measured how the system responded depending on the parameters set.
Tesla
Autonomous Vehicle Bias
The report also says that academics and advocates have raised the issue of Autonomous Vehicle (AV) biases. There have also been other studies into biases before, back in 2019, that some pedestrian-detection systems had higher false-negative results on darker-skinned people.
This latest study raises more concerns and questions about integrating AVs, especially how they will decide which pedestrians to yield to in safety situations. More and more governments have also voiced concerns about how AVs should be legislated for safety.
MARTIN LELIEVRE/AFP via Getty Images