‘AI self-driving cars have accident risk between dawn and dusk’

Which is safer: self-driving cars driven by artificial intelligence (AI) or cars driven by humans? Contrary to the common perception that self-driving cars are safer, recent research indicates that human-driven cars are safer under certain conditions. With current technology, the safety comparison between AI and human drivers results in a 'draw.'

On Wednesday, Professor Muhammad Abdel-Atty's team at the University of Central Florida announced findings that, while self-driving cars are generally safer in normal driving environments, human drivers outperform AI in dark conditions such as dawn or dusk. These results were published in the scientific journal "Nature Communications."

The research team analyzed accident data from 2,100 self-driving cars and 35,000 human-driven cars. Their study shows that self-driving cars are safer in typical driving situations, such as clear weather and maintaining constant lanes. Even in rainy weather, self-driving cars have only one-third the probability of accidents in human-driven cars.

However, the accident risk for self-driving cars was 5.25 times higher than for human-driven cars in dark environments such as dawn or dusk. The research team attributed this to the slower response of autonomous driving sensors and cameras to changes in light. For instance, long shadows at dusk can confuse sensors, making it difficult to recognize objects or hazards.

Additionally, self-driving cars are 1.98 times more likely to be involved in accidents when turning at intersections. This is due to the limited recognition range of autonomous driving sensors, which restricts their ability to detect overall road conditions. The reliance on pre-programmed algorithms also makes it challenging for self-driving cars to handle unexpected scenarios effectively.


Teuk-Gyo Koo kootg@donga.com