LLM-Driven Autonomous Vehicles Inherit Human Driver Biases in Pedestrian Yielding: Results and Implications From A New Benchmark

Explain Like I'm Normal

Researchers found that using Large Language Models to guide autonomous vehicles causes the cars to inherit human racial biases, specifically yielding less frequently to Black pedestrians. The study introduces new 'All Else Being Equal' benchmarks to audit how 'common sense' AI models might unintentionally recreate systemic societal prejudices on the road. This suggests that simply making a car 'smarter' with LLMs might actually make it less fair.