4 ms·
"Three more doublings, and that number will be bigger than average miles between accidents for human drivers. Then they can ship a product." I am skeptical of
by perl4ever 8y ago
"Three more doublings, and that number will be bigger than average miles between accidents for human drivers. Then they can ship a product."
I am skeptical of that, because I expect that even if disconnects are ultra-rare overall, and the testing is representative of normal usage, there will be enough people that frequently use the technology in ways far off on the tail of the distribution that for them, disconnects will be much less rare. And that might create a PR blowup or worse.
I think what is worrying me may be something called "heteroscedasticity".
https://en.wikipedia.org/wiki/Heteroscedasticity https://en.wikipedia.org/wiki/Heteroscedasticity
- cbhl 8y agoI'm way less worried about this than when Waymo was originally called the Google Self-Driving Car Project. If Uber and Tesla continue with their current plans, they will have enough accidents that self-driving cars being in accidents will cease being headline news. They just become an actuarial math problem, just like regular car insurance is today. (This is my opinion and not that of my employer.)
- est31 8y agoFor self-driving cars, what would that be? You can misuse a non-self driving car by speeding, giving you above-average chances of accidents. You can be old, giving you above average chances of being involved in accidents (per km driven). You can smoke, giving you above average chances of getting lung cancer. You can misuse a partially self driving car by somehow convincing it that you do follow the traffic while actually not following it, and then blame any accidents onto the self-drive feature. But fully self driving cars only have two inputs: starting point and destination. Doesn't matter whether you are old, sleeping, 13 or a dog. Unless some roads are more susceptible to accidents than others, or manufacturers let people also fiddle with how much a car should risk accidents in order to get a little decrease in travel time, I don't see any way possible. And if manufacturers really do give that option, it's truly their fault.
- salty_biscuits 8y agoIn the hands of a consumer, don't forget being diligent with maintenance, driving on unsealed roads, choosing to drive in poor conditions. There are probably a lot more. It is surprising how much some people abuse machines.
- scottlamb 8y agoWaymo's "shipped product" is likely to be a fleet service in several cities, at least at first, so maintenance is their responsibility. Likewise, I'd imagine unsealed roads would be outside initial coverage areas. (Note that while I work at Google, I don't have any particularly relevant inside knowledge. What I'm saying is consistent with their current pilot in Phoenix and public articles.)
- xbmcuser 8y agoThey already have a fleet although in just a part of one city. They will keep expanding the fleet and the area/cities covered over time. Instead of one big announcement that we have selfdriving cars they will keep expanding and improving over decades. And just like people kept harping about driving range of electric cars. They will suddenly be everywhere. Car manufacturers are wary of Waymo so far but the first large manafacturer to join hands with them will probably dominate car/transport industry of the future.
- bsder 8y ago> You can be old, giving you above average chances of being involved in accidents (per km driven). You have to be very old for this to be true. It turns out that 16-25 years old is worse than any group until something like 75-80 years old.
- paulddraper 8y ago> Unless some roads are more susceptible to accidents than others By chance do you happen to live somewhere without icy roads?
- dcomp 8y agoI've now seen the word heteroscedasticity twice in 5 minutes and thank you for providing the link to Wikipedia. Although I think the 'meta-variety' synonym was quite apt on the other page.