Android is now the world’s largest earthquake detection network

[Pickles von Brine]

Quote

Quote from Google - All smartphones come with tiny accelerometers that can sense earthquakes. They’re even sensitive enough to detect the P-wave, which is the first wave that comes out of an earthquake and is typically much less damaging than the S-wave which comes afterward. If the phone detects something that it thinks may be an earthquake, it sends a signal to our earthquake detection server, along with a coarse location of where the shaking occurred. The server then combines information from many phones to figure out if an earthquake is happening. We’re essentially racing the speed of light (which is roughly the speed at which signals from a phone travel) against the speed of an earthquake. And lucky for us, the speed of light is much faster!
 

In California, Android's earthquake detection will be a collaboration between Google and the same coalition that handles ShakeAlert, the back-end system that the MyShake client surfaces to users. California's ShakeAlert combines smartphone readings with a network of traditional seismometers, and now Android will be just another ShakeAlert client, pumping data into the system and showing alerts from it.

For everywhere else in the world, which doesn't have such an advanced earthquake-detecting infrastructure, Google's in-house Android Earthquake Alerts System will be on earthquake watch. The company says that "to start, we’ll use this technology to share a fast, accurate view of the impacted area on Google Search. When you look up “earthquake” or “earthquake near me, you’ll find relevant results for your area, along with helpful, credible resources on what to do after an earthquake."

The feature is being distributed through Google Play Services for every Android phone running version 5.0 and up. Unlike major system updates, which take years to reach the majority of Android phones, Google Play Services is centrally distributed by Google and can hit every single active Android phone (excluding non-Google devices in China) in a matter of weeks. The Android 5.0 and up requirement means 94 percent of the 2.5 billion Google Play Android devices will have access to the feature.

Wait wait wait. What? Huh? Now this is pretty cool. The ability to use acceremeters in phones to detect earthquakes? I had no idea they were that sensitive. You could use this to create an early detection system! Especially if you are in an affected area or not sure if you are in an affected area. The other part is how it is being rolled to anything running android 5.0 and up! So many devices. So much data! If someone was a seismologist, I am sure they would be excited to have access to this info. Plus, it looks like this is in near real time, so 10s of thousands, if not hundreds of thousands of devices detecting things could be interesting. 

Then again, this is just me rambling. Either way, some cool stuff!
Sauce Source

[Skipple]

It's shit like this that gets me so amped for the future. Such a cool idea. 

[Pickles von Brine]

Just now, Skipple said:

It's shit like this that gets me so amped for the future. Such a cool idea. 

Totally!

[eeeee1]

intresting...

[SpaceGhostC2C]

14 minutes ago, Pickles - Lord of the Jar said:

Quote from Google - All smartphones come with tiny accelerometers that can sense earthquakes. They’re even sensitive enough to detect the P-wave, which is the first wave that comes out of an earthquake and is typically much less damaging than the S-wave which comes afterward

But will that work regardless of where your phone is at, if it's based on the accelerometer? What I mean is, there are many ways in which a phone can be vibrating or at least partially isolated from external vibrations. Or the key is in large numbers, i.e., each individual signal is extremely noisy and could be anything, but the aggregate pattern will only emerge if there's actually an earthquake?

 

17 minutes ago, Pickles - Lord of the Jar said:

 We’re essentially racing the speed of light (which is roughly the speed at which signals from a phone travel) against the speed of an earthquake. And lucky for us, the speed of light is much faster!

That's most definitely not true? As in, waves may travel at the speed of light, but if you take the time between the phone sensing something and the server somewhere decoding it as an alert, putting it together with other signals, and concluding there is an earthquake, and you divide the distance between the earthquake and the server by that figure, you don't get anywhere near the speed of light? I mean, once you account for all the latencies at both ends, and in all intermediate hops, not to mention the time involved in actually doing something useful with those signals, I'd think you're way behind again. But maybe I'm overestimating "the speed of the earthquake". Still, it's certainly not a race against the speed of light...

 

23 minutes ago, VegetableStu said:

also xkcd related

 

That's good 

Although I still think it merits a Willy Wonka saying "tell me again about your 0ms ping, 0 latency wireless connection" or "so you were saying your wireless mouse causes input lag?" 

A guy saying "we have an earthquake" over AM radio will probably outpace twitter many times over 

[Pickles von Brine]

1 minute ago, SpaceGhostC2C said:

But will that work regardless of where your phone is at, if it's based on the accelerometer? What I mean is, there are many ways in which a phone can be vibrating or at least partially isolated from external vibrations. Or the key is in large numbers, i.e., each individual signal is extremely noisy and could be anything, but the aggregate pattern will only emerge if there's actually an earthquake?

 

That's most definitely not true? As in, waves may travel at the speed of light, but if you take the time between the phone sensing something and the server somewhere decoding it as an alert, putting it together with other signals, and concluding there is an earthquake, and you divide the distance between the earthquake and the server by that figure, you don't get anywhere near the speed of light? I mean, once you account for all the latencies at both ends, and in all intermediate hops, not to mention the time involved in actually doing something useful with those signals, I'd think you're way behind again. But maybe I'm overestimating "the speed of the earthquake". Still, it's certainly not a race against the speed of light...

 

That's good 

Although I still think it merits a Willy Wonka saying "tell me again about your 0ms ping, 0 latency wireless connection" or "so you were saying your wireless mouse causes input lag?" 

A guy saying "we have an earthquake" over AM radio will probably outpace twitter many times over 

I have no idea. i am just the messenger. (I am being facetious)

[Bananasplit_00]

So without asking me, Google will now be sending off my accelerometer data and location to some third party? Yikes

[Kisai]

30 minutes ago, Bananasplit_00 said:

So without asking me, Google will now be sending off my accelerometer data and location to some third party? Yikes

https://www.theverge.com/2020/8/11/21362370/android-earthquake-detection-seismometer-epicenter-shakealert-google

Quote

Stogaitis says that the information collected as part of this program is “de-identified” from users and that Google only needs “coarse” location information for it to work. Both the earthquake alerts and the detection system are opt-in, as well. “What we really need for this is just these little mini seismometers that are out there,” Stogaitis says. “We don’t need to know anything about the person itself that’s sending it because that doesn’t matter.”

It's opt-in.

[Pickles von Brine]

1 minute ago, gabrielcarvfer said:

They're not. I don't want to enter into statistics, but they're basically doing a glorified Monte Carlo simulation using real data instead of randomly generated ones. When you have a large enough and uniformly spaced population (in time or in space, depending on what you're trying to measure) and sample them, your results will end up following a normal/Gaussian distribution with the mean value approximating the real unknown value you're trying to measure (in this case the vibrations).

English please. Thanks. 

[RejZoR]

Android is also Google's biggest data hoarding network... Yeah, no thanks, no matter how cool it sounds.

[GOTSpectrum]

2 hours ago, SpaceGhostC2C said:

But will that work regardless of where your phone is at, if it's based on the accelerometer? What I mean is, there are many ways in which a phone can be vibrating or at least partially isolated from external vibrations. Or the key is in large numbers, i.e., each individual signal is extremely noisy and could be anything, but the aggregate pattern will only emerge if there's actually an earthquake?

Think about it like google traffic, if you have one person zooming down the motorway on a bike between cars, the potential hundreds of people who are sitting in the jam will overwhelm the few individuals that are outliers. 

 

Basically they are looking for patterns in the highly noisy data, it is an interesting concept to say the least, they will probably be doing many things with the data in real time. Probability density functions will be a must here. If you have enough data points, you can compute the likelihood of them being the true value by using various averages and other such statistical wizardry that I don't honestly comprehend.  

 

The interesting thing here is once this has been deployed and it has 'seen' real earthquakes the data from that could most probably be used to help improve the model. As more data is available of real earthquakes the likelihood of it finding smaller earthquakes increases. This can be done by using any number of methods but training some kind of neural network would probably be quite interesting. 

[Kisai]

5 minutes ago, Caroline said:

They always say that.

They only need a few dozen opt-in's per sq km for it to be of any value, it's not like the covid contact tracing which requires 100% opt-in to work, and even 50% opt-in is too coarse to really do anything but generalize.

[Commodus]

8 minutes ago, RejZoR said:

Android is also Google's biggest data hoarding network... Yeah, no thanks, no matter how cool it sounds.

Can we please have a more measured approach to Google data collection?

 

Yes, there are reasons to be concerned about Google's approach to some data collection, but if we don't distinguish between responsible collection like this and the unscrupulous kind, we're not acknowledging when it does things right.

 

Besides, what is Google going to do with anonymized and deliberately rough earthquake data? It's not about to target you with quake insurance ads.

[GOTSpectrum]

Just now, gabrielcarvfer said:

You can't use maps without letting them track you, you can't use gmail without letting them look through your stuff, you can't use voice assistant without letting them hear and transcribe whatever you said. The "but you can opt-out" thing is completely fake, it just means that you can request them not to show your own data to you, but they obviously use it to train their models, extract insights, etc.

Trying not to derail this thread too much but I have a real world example of this that affects me personally.

 

I have 'Web and App activity tracking' disabled on my google account, and due to this my google assistant has lost almost all of it's functionality. I can't even ask the assistant to call someone from my contacts for example. Now I don't know why Google needs to track my web activity to be able to call my mum when I ask but here we are. 

 

The issue is we don't have any granularity in what data we do and don't want them to store. And honestly, I'm still not even certain that they don't actually store the information anyway; the sad part is, there is no way to know one way or the other unless there's a dataleak that contains that specific data and i learn about it in some way. 

[Pickles von Brine]

34 minutes ago, gabrielcarvfer said:

It is harder to explain than to demonstrate. Same concept but to calculate the value of Pi.

Generate a bunch of random coordinates (samples) inside of a square.

Count how many of those coordinates are inside a circle of radius equal to the square side.

Divide the number of samples inside the circle by the total samples and voilà, a Pi approximation.

 

Yeah not helping there chief. My brain already broke with your previous comment. This just caused a meltdown into a puddle of goo. Numbers hurt brain. 

[Shreyas1]

How does this distinguish between an earthquake and for example, people on a train? Could the acceremeters be triggered by something else? Could cause a lot of panic if something goes wrong.

 

But still a great idea IMO

[SpaceGhostC2C]

1 hour ago, gabrielcarvfer said:

They're not. I don't want to enter into statistics, but they're probably doing a glorified Monte Carlo simulation using real data instead of randomly generated.

It's not a Monte Carlo simulation if it uses real data, though...

 I agree with @GOTSpectrum, it's probably just filtering, machine learning, and plain statistics more generally. I guess it boils down to finding a statistic that correlates heavily with an earthquake taking place. The way they describe it makes it seem like they are not collecting accelerometer data all the time, then figuring out a pattern among all possible measures, but rather trigger 0/1 reports only if the individual accelerometer detects a particular movement, then make a decision by aggregating the binary data. Then it's just a matter of identifying a threshold for the density of 1s beyond which you attribute it to an earthquake.

 

[SpaceGhostC2C]

3 minutes ago, Shreyas1 said:

How does this distinguish between an earthquake and for example, people on a train? Could the acceremeters be triggered by something else? Could cause a lot of panic if something goes wrong.

 

Well, you have a lot of people on the train every day, so you learn pretty quick how that looks like. Since an earthquake takes place on top of everyday life, you're not looking for movement, but for abnormal movement on top of the "background" routine data.

1,000 people shaking their phones frantically will look very different from 100,000 phones catching P-wave-induced vibrations, 1,000 of which are being shaken frantically  

[Commodus]

1 minute ago, gabrielcarvfer said:

You can't use maps without letting them track you, you can't use gmail without letting them look through your stuff, you can't use voice assistant without letting them hear and transcribe whatever you said. The "but you can opt-out" thing is completely fake, it just means that you can request them not to show your own data to you, but they obviously use it to train their models, extract insights, etc.

First, the earthquake data is opt-in, not opt-out. And again, we need to be adults about this. Screaming "Google is always spying on everything you do" is not only objectively false, but unhelpful. We won't see healthy uses of data if we insist that any kind of data collection is automatically evil. You want Google to mend its ways? Support its good decisions and focus your energy on the bad ones.

[wANKER]

13:02 EMERGENCY BROADCAST: EARTHQUAKE DETECTED SEEK SHELTER IMMEDIATELY  

13:03 EMERGENCY BROADCAST: oh wait no it's valentines day isn't it and pornhub have free premium 

[Commodus]

42 minutes ago, gabrielcarvfer said:

It is not objectively false if you add "on your Android phone/Chrome browser".

It isn't automatically evil, even thought this is actually the case. The lack of transparency in data collection alone is more than enough indication of their evilness.

What exactly Google Play Services do in the background? You have no idea.

 

Sorry, but that's still not true. Even if you allow Android's tracking features, it's not constantly monitoring everything you do. More than you may be comfortable with, but not everything. There are areas where Google is transparent, too, even though it ought to be much clearer.

 

You see the issue here? I'm a big fan of privacy-oriented solutions and use Apple gear partly for that reason, but it's not advancing the discussion if you simply portray Google as a demon... because it's not. I'm frustrated as an Apple user when people hate the company based on simplistic stereotypes; it's not right to do the same against others.

[rcmaehl]

Waiting for Car Crash Detection to be mainstream
 

 

[Curufinwe_wins]

This is in fact very good. The detection thresholds from that many signals random noise shifting by those fractional values definitely can be much much faster and more precise than traditional models, and with earthquakes, even the smallest additional warning has huge knock-on affects for human safety.

[Quinnell]

If you told me in 2007 that everyone's phones would aid in measuring earthquakes, I would've thought you're full of it.  Technology really can do some amazing things.

[Nine Tailed Fox]