[NaughtyShiba]

It’s probably easy (in terms scale of all that detection) to fix, and for what it’s worth, this feature can definitely be a lifesaver.

[fasthands9]

I am very curious if they did assessment of other situations which could trigger similar sensor reactions that they had to rule out. Theme parks could actually be geofenced off without not that much work - there are under 1000 roller coasters in the US.

Did they have to consider people dropping phones out of car windows (or off a boat ?) What about ski crashes or skateboard crashes?

[bpeebles]

There are traveling carnivals to things like state/county fairs, as well as just big empty parking lots of malls/etc for a few weeks, with significant g-force rides. So well over a thousand locations in the US.

[mdaniel]

> Theme parks could actually be geofenced off without not that much work - there are under 1000 roller coasters in the US.

I'm cognizant of the 80/20 rule, and "perfect is the enemy of good," but if a run-of-the-mill coaster is setting off that detection, then I'd wager the traveling fair rides will be even more violent, based on my experiences growing up

I have no idea where this falls on the privacy spectrum, but I'd guess if there are 5 or 10 triggering events within some timespan and some geo boundary, that's an indicator of a themepark-ish setup

Clicking "report false positive" on the phone would also likely go a long way

[NobodyNada]

There was an article a couple weeks ago about the feature triggering when someone's iPhone fell off his motorcycle: https://news.ycombinator.com/item?id=32968229

[MBCook]

To be fair it didn’t fall off a motorcycle. It was thrown (due to a hard bump) from a moving motorcycle. If it had been in the person’s pocket at the time because they had fallen off too, calling 911 would have been very fitting.

[fasthands9]

I guess the question to me is what the final false positive rate on this feature is. Its obviously a huge net win if the proportion is small - but if it is high enough I'd imagine that dispatchers would be less likely to want to send officers to the automated calls.

[NaughtyShiba]

Hard to answer. However it’s also probably hard to come up with all the edge cases.

[Tempest1981]

Or NFL quarterbacks!

[postalrat]

It could also be a life-taker when it calls 911 and takes resources that could have been used for saving a life.

[P5fRxh5kUvp2th]

yep.

It's the hubris of tech people to think everything can be solved (or made better) with technology.

[tomcam]

Let me try playing devil's advocate (with zero inside info). Isn't it possible that more lives are saved such that it's worth the false alarms?

It is obvious to me that these decisions will become more and more difficult. I am certain that at some point (not sure if 10 years from now or 50) driverless tech will be safer than people--by a margin of say 99%. I am equally certain that this will be accompanied by a set of accidents unique to the driverless tech. I think of it as similar to compilers. When I started programming PCs assembly was always king. C got better and better. 10 years into my career C was mostly better. Now C/C++/Rust etc. almost never need code rewritten in ASM.

[P5fRxh5kUvp2th]

Driverless vehicles don't go on rollercoasters, if they want to automatically phone in an emergency call they'll probably be able to do so reliably.

a phone in your pocket does not have that context.

[NaughtyShiba]

Aren’t new cars supposed to have that feature, which can automatically call emergency services on crash?

[P5fRxh5kUvp2th]

I would imagine so, but my newest vehicle is 2004 so I have no firsthand experience :)

[jeffreygoesto]

It isn't. They hit the ROC like every other classifier on the tails.