Apple Executives Offer Closer Look at iPhone 14's Crash Detection

[szw-mapple fan]

If it's not a "straight equation" why offer it? At this point, it will be best to offer users to manually toggle the crash detection off more easily. It seems like this feature is still in the pre-mature stages. It needs work!

Because it takes into account multiple factors that are relative each other, like the rest of the quotes says? Because some false positives is worth it if it could save more lives? I don’t understand this mentality of if it‘s not 100% we shouldn’t have it.

Just because a few edge cases came up doesn’t mean it doesn’t work. Sometimes these can only be discovered once people use it in more places. That‘s just the nature of these things. Would you think to test it on a roller coaster if you were developing this feature for the first time?

 

[theluggage]

Crash detection is pretty cool but I have concerns about false positives (roller coasters) and what about the bad press when it doesn't detect a crash and someone doesn't get help in time?

Beyond that, what is the extent of the problem its meant to solve? I.e. how frequent are crashes that:

(a) Incapacitate the driver and any passengers so they can't call but don't kill them
(b) Aren't witnessed by anybody
(c) Can be reliably detected by the phone
(d) Happen in places with reliable cellphone coverage (see (b))

... and its no good just finding a few anecdotes where alerting the emergency services a minute or two earlier would have saved lives (that's undoubtedly true) but it needs numbers that can be weight against:

(1) The effect of false positives wasting the emergency services' time (there will be cases where that costs lives)
(2) The "cry wolf" effect of (1) (if you have to prfioritize, you're gonna take the live call over the robocall)
(3) The "false security" effect (no. I'm not going to tell people I'm off for a drive through Dead Man's Pass - my phone will save me)

...and Apple have a responsibility to do the math before introducing such a feature.

 

[szw-mapple fan]

3. Apple could automatically disable the feature while a user is in an amusement park. Apple knows where all the amusement parks are (because Apple Maps) and location services are required to be enabled to use crash detection, so this seems an easy way to eliminate many false positives.

You can still have car accidents in amusement parks so that wouldn’t be ideal. Larger parks can still have high speed traffic going through them. It would also require Apple to constantly collect everyone’s GPS data just for the sake of identifying just the people who are in those amusement parks, which is terrible for privacy.

It‘s more likely that the engineers never thought to test the feature in amusement parks during development. Now that they know rollercoasters are triggering it, it should be easier to retrain their model so it can detect rollercoaster rides and sounds and avoid false positives than to precisely geofence every amusement park in existence.

 

[WiiDSmoker]

Because it takes into account multiple factors that are relative each other, like the rest of the quotes says? Because some false positives is worth it if it could save more lives? I don’t understand this mentality of if it‘s not 100% we shouldn’t have it.

Just because a few edge cases came up doesn’t mean it doesn’t work. Sometimes these can only be discovered once people use it in more places. That‘s just the nature of these things. Would you think to test it on a roller coaster if you were developing this feature for the first time?

Many of us immediately thought 'what about roller coasters' as soon as the functionality was revealed - so yes.

 

[tdar]

Did they explain why it goes off as a crash on a roller coaster?

 

[Jim Lahey]

What happens if I crash my phone into a rollercoaster? 🤔

 

[bizack]

If it's not a "straight equation" why offer it? At this point, it will be best to offer users to manually toggle the crash detection off more easily. It seems like this feature is still in the pre-mature stages. It needs work!

ML is not a 'straight equation' by definition. Modeling specific inputs to trigger an output, often times with hundreds or thousands of inputs, isn't an A + B = C process. As more use cases and inputs are added, and the model is updated, the output increases in accuracy. In the case of crash detection, you may want to initially bias towards a false positive under extreme scenarios, like rollercoasters. Also, tradeoffs.

 

[CarAnalogy]

There was so little for Apple to promote as a reason to buy the iPhone, they were grasping at straws and a half-baked crash detection feature seemed like a good idea at the time.

It was definitely on theme with this year’s “Buy an iPhone or Literally Die” message.

 

[citysnaps]

Another race to the bottom with inane juvenile comments.

 

[RodThePlod]

I mean - I saw this story where crash detection did work as intended. So... pros and cons, I guess...

Police say iPhone 14 crash-detection feature alerted them to fatal crash where a Honda Accord plowed into a tree, killing 6

The new iPhone and Apple Watch include a crash-detection feature that can automatically alert emergency services after a severe crash.

www.businessinsider.com

 

[citysnaps]

Because it takes into account multiple factors that are relative each other, like the rest of the quotes says? Because some false positives is worth it if it could save more lives? I don’t understand this mentality of if it‘s not 100% we shouldn’t have it.

Just because a few edge cases came up doesn’t mean it doesn’t work. Sometimes these can only be discovered once people use it in more places. That‘s just the nature of these things. Would you think to test it on a roller coaster if you were developing this feature for the first time?

Finally, an intelligent assessment in an abundance of juvenile comments from others.

Apple will gather reports and adjust their algorithm. Life goes on.

 

[tomtad]

Came expecting Phil Schiller on a rollercoaster, left disappointed

 

[WiiDSmoker]

Another race to the bottom with inane juvenile comments.

Did you miss the one where the OP is the one who started all this? Then had to delete his post. Then had to delete his news story.

 

[citysnaps]

ML is not a 'straight equation' by definition. Modeling specific inputs to trigger an output, often times with hundreds or thousands of inputs, isn't an A + B = C process. As more use cases and inputs are added, and the model is updated, the output increases in accuracy. In the case of crash detection, you may want to initially bias towards a false positive under extreme scenarios, like rollercoasters. Also, tradeoffs.

Hat-tip for another thoughtful and intelligent assessment!

 

[citysnaps]

Because it takes into account multiple factors that are relative each other, like the rest of the quotes says? Because some false positives is worth it if it could save more lives? I don’t understand this mentality of if it‘s not 100% we shouldn’t have it.

Just because a few edge cases came up doesn’t mean it doesn’t work. Sometimes these can only be discovered once people use it in more places. That‘s just the nature of these things. Would you think to test it on a roller coaster if you were developing this feature for the first time?

Absolutely. Having many millions of phones out in the field all of a sudden will provide Apple with valuable information, leading to a refinement of their crash-detection algorithm.

 

[macfacts]

Typical apple, using the public to beta test features.

 

[Sportcat]

"You're crashing it wrong."

 

[Expos of 1969]

Another race to the bottom with inane juvenile comments.

Not sure why you continue in these forums when you seem to get bothered by insightful, humorous and topical comments exposing Apple flaws which don't jive with your pro-Apple fandom.

 

[NMBob]

Is there going to be feature creep to cover the stock market?

 

[JosephAW]

„We put iPhones in many different places throughout the car — on the dummies and the car itself and mounts and so forth.“

Yeah, but not in a roller coaster!
Yeah Tim, you could have tested it on the Oktoberfest, instead of drinking 🍻 all day long.

View attachment 2092061

No that's fall detection.

 

[compwiz1202]

Crash detection is pretty cool but I have concerns about false positives (roller coasters) and what about the bad press when it doesn't detect a crash and someone doesn't get help in time?

Yes an error either way will make them look bad. Either it doesn't work with only an unconscious driver, or too many false positives stealing resources from a legit emergency

 

[compwiz1202]

Beyond that, what is the extent of the problem its meant to solve? I.e. how frequent are crashes that:

(a) Incapacitate the driver and any passengers so they can't call but don't kill them
(b) Aren't witnessed by anybody
(c) Can be reliably detected by the phone
(d) Happen in places with reliable cellphone coverage (see (b))

... and its no good just finding a few anecdotes where alerting the emergency services a minute or two earlier would have saved lives (that's undoubtedly true) but it needs numbers that can be weight against:

(1) The effect of false positives wasting the emergency services' time (there will be cases where that costs lives)
(2) The "cry wolf" effect of (1) (if you have to prfioritize, you're gonna take the live call over the robocall)
(3) The "false security" effect (no. I'm not going to tell people I'm off for a drive through Dead Man's Pass - my phone will save me)

...and Apple have a responsibility to do the math before introducing such a feature.

2 is a great point

 

[fontman]

I don't know if this feature is ready.

 

[Ashbash75]

Obviously not certain executives, i.e. :

"I race cars, play tennis, and fondle women, BUT! I have weekends off, and I am my own boss."

 

[tomtad]

I don't know if this feature is ready.

I don’t know if this feature is needed.