3 ms·
Tesla clearly needs to hire a "Red Team" to find weaknesses of their autopilot (and other systems). The odd thing is that even a randomly stupid AI for self dr
by mikewarot 5y ago
Tesla clearly needs to hire a "Red Team" to find weaknesses of their autopilot (and other systems).
The odd thing is that even a randomly stupid AI for self driving is statistically safer than most drivers. Clearly there is a ton of room for improvement in both AI and Humans.
- Closi 5y ago> The odd thing is that even a randomly stupid AI for self driving is statistically safer than most drivers. This hasn’t been shown yet at all. Statistics showing autopilot have less crashes per mile always ignore that Autopilot is doing the type of driving that has the least accidents per mile (motorway driving).
- bigdict 5y ago> Red Team I'm sorry, did you mean customers?
- falcolas 5y agoHe didn't. Telsa did.
- ModernMech 5y agoWhy hire a team to beta test your AI self driving car when you have customers willing to shell out $10k for the privilege?
- Syonyk 5y ago> The odd thing is that even a randomly stupid AI for self driving is statistically safer than most drivers. A strong claim, lacking actual evidence for it. All we have to go on are some Elon tweets (rather the definition of a biased source) and the actual crash rate. Without a lot more data (which Tesla steadfastly refuses to release) about environments, corrections, etc, it's quite impossible to make that sort of statement with any confidence. The Tesla hardware is a weird combination of capable and insanely dumb, and it's far from obvious which it will be in any given situation until it's gone through it. If an honest statistical analysis of the data indicated that Tesla's automation was better than human drivers (or better than other driver assist systems), I would fully expect them to have released the values. Since they haven't, and only hint at it and make statements that sound statistical but really aren't, I assume they've done the numbers internally and know it's not nearly as good as they like to imply. If I drove in a city like their "self driving" beta was a few months back, I would be hauled from the car on suspicions of driving while hammered.
- darepublic 5y ago> even a randomly stupid AI ... Is statistically safer than most drivers the fact that people still trot this out every Tesla thread is super annoying. Sorry to break it to you but this is a 100% false claim
- alkonaut 5y ago> The odd thing is that even a randomly stupid AI for self driving is statistically safer than most drivers. Under sunny highway conditions? Perhaps. In a night snowstorm? Probably yes - because the AI would be at the side of the road waiting for the human to drive. I think self driving is a typical 80/20 problem. We won't have "full" self driving because the costs are exponential for each step closer to it. But driving on 80% of roads on 80% of days, with supervision? That could happen. But that said: we won't accept AI that just makes traffic safer "on average". I'm fine with human shortcomings causing accidents. People will not accept car manufacturers cutting corners and causing accidents, even if statistically it's safer. So the very high bar for self driving isn't just "as safe as humans".
- mikewarot 5y agoI was wrong, it's almost as safe, not safer. https://www.forbes.com/sites/bradtempleton/2020/10/28/new-tesla-autopilot-statistics-show-its-almost-as-safe-driving-with-it-as-without/?sh=796a3f661794 https://www.forbes.com/sites/bradtempleton/2020/10/28/new-te...