Normalized crash data shows Autopilot is much less safe than Tesla claims (twitter.com)

Article summary

A study has found that Tesla's Autopilot crash data is not as safe as the company claims, as it does not account for the safety of the operating domain. The study's findings suggest that Tesla's comparison of raw Autopilot crash data to average human driver crash data is deceptive. The issue lies in the fact that not all crashes are reported, and the data may not accurately reflect the true safety of Autopilot. This has led to concerns about the validity of Tesla's safety claims.

Main themes

• Autopilot safety
• Tesla's methodology
• Automated driving regulations
• Liability and contracts
• Human driving behavior
• Road safety norms

What commenters say

• Tesla's safety claims are exaggerated and misleading due to flawed methodology.
• The company's decision to program illegality into their products, such as rolling stops, opens them up to liability.
• Automated cars should be programmed to follow the law exactly, including speed limits and stop signs.
• Human drivers often break the law, such as rolling through stop signs, and automated cars should be able to make similar judgments.
• The expectation that automated cars will always follow the speed limit and come to a complete stop at stop signs is unrealistic and may not be safe in all situations.
• Tesla's contracts may protect them from liability in cases where their cars are involved in accidents while operating in autonomous mode.
• The fact that not all crashes are reported and the data may be incomplete undermines the validity of Tesla's safety claims.
• Programming automated cars to break the law, even if it is common human behavior, is a bad idea and can lead to accidents and liability.