3 ms·
The last time I saw the statistics, Tesla autopilot was about average for US drivers, but it's actual usage was biased towards the roads least likely to cause f
by ben_w 4y ago
The last time I saw the statistics, Tesla autopilot was about average for US drivers, but it's actual usage was biased towards the roads least likely to cause fatalities.
I'm all in favour of the vision behind self-driving cars, but the reality isn't quite as impressive as I'd like, especially given the AI has inherent advantages such as "no blind spot" and "can pay full attention to everything all at the same time" and "not distracted by adverts, hunger, and phone" and "doesn't get tired or drunk".
And — at risk of displaying the limits of my domain knowledge — I think the AI can be tested without being in charge, by asking the AI to create a model of the world as it drives, and/or by testing its predictions of what it will observe in the next few seconds.