Follow Slashdot stories on Twitter

 


Forgot your password?
Close
typodupeerror
×
Google Apple

Google and Apple's Contact-Tracing API Doesn't Work on Public Transport, Study Finds (vice.com) 43

Posted by msmash from the closer-look dept.
Covid-19 contact-tracing apps that rely on an API developed by Apple and Google and bluetooth technology cannot accurately measure the distance between users on public transport, a recently released study from Trinity College Dublin has found. From a report: The researchers of the study first tested the API on a group of volunteers who switched seats every fifteen minutes in a Dublin tram. They then ran the collected data through the detection rules of the Swiss, German, and Italian contact-tracing apps to see how often they correctly identified contact between users. Based on this, they found that the chance of an accurate detection was "similar to that of triggering notifications by randomly selecting from the participants in our experiments, regardless of proximity."

So, no better than random. This finding is the latest example of mounting skepticism among experts regarding the effectiveness of the technology underlying the apps which have been widely released -- but less widely used -- by governments across Europe and more recently the United States. Most contact-tracing apps in Europe and the United States use Apple and Google's exposure notification API, which in turn relies on in-built wireless Bluetooth technology to estimate the distance between two users and whether they've been in contact. What actually constitutes 'contact' is set by the developers of whichever app calls the API, but it's usually defined as being within 2 meters (~6.5 feet) of another user for at least 15 minutes. Once a user uploads a positive test result to a contact-tracing app, it notifies all contacted users and lets them know that they've been at risk of infection.
This discussion has been archived. No new comments can be posted.

Google and Apple's Contact-Tracing API Doesn't Work on Public Transport, Study Finds More

Google and Apple's Contact-Tracing API Doesn't Work on Public Transport, Study Finds

Comments Filter:

  • The researchers of the study first tested the API on a group of volunteers who switched seats every fifteen minutes in a Dublin tram. They then ran the collected data through the detection rules of the Swiss, German, and Italian contact-tracing apps to see how often they correctly identified contact between users. Based on this, they found that the chance of an accurate detection was "similar to that of triggering notifications by randomly selecting from the participants in our experiments, regardless of pr

    • in point of fact some viruses do care about race, this pandemic's one does. Even accounting for disparity in health care, it hits blacks and latinos hardest, asians somewhat harder than whites.

      https://www.cdc.gov/coronaviru... [cdc.gov]

      • Considering where this branch is going, I shudder to imagine where the AC FP is.

        I think there's a deeper reason this approach is stupid, though I acknowledge that the details are devilish and damning. The gigantic problem is that the incentives are bass-ackwards. You have to go to a lot of trouble, you can't tell if the app is going to work correctly even if the details are correct, it's probably draining your battery, too, but then, even after all of the hassles, you get punished if you succeed in using th

        • Re:Incentives to use the app are all wrong (Score:4, Insightful)

          by CaptQuark ( 2706165 ) on Saturday October 10, 2020 @01:51AM (#60591030)

          ...you get punished if you succeed in using the app. You don't get any reward, but rather you might get sentenced to 2 weeks of solitary confinement. But it won't matter anyway. If you get sick then you still get sick and suffer and if you don't get sick, then you wasted time and probably lost a lot of money on the false alarm. With success like this I'd rather fail

          Of course I want to know if I've been exposed. Knowing that I can adjust my habits to perhaps wait to go visit my kids and grandkids. I can reschedule a dental appointment until after I know if I've been infected. I can attend a business meeting virtually instead of going into the office. I can avoid going to restaurants until I know. Saying it won't matter anyway means you view it as not mattering to YOU but it could matter to a lot of other people.

          Hmm... I just thought of an interesting wrinkle that might help, though it could make the app less simple. Periodic sound samples. The basic idea is that when I'm in quiet places without noisy crowds of people around me, then I'm probably safer from Covid-19 and I should try to adjust my lifestyle to spend more time in such places.

          Yeah, and when you put your phone in your purse, you've just rendered the comparative noise data worthless.

          ---

        • You need an app to figure if you are surrounded by a lot of people?
          You need an app to track times when this happens? Or places?

          Well, I'm under a bridge, I hack you the app in a few hour, but you have to pay me handsomely, especially for the bridge, too.

      • What they've found is that it's not really race, but rather poverty that makes people at higher chance of getting the virus. Because black people are more likely to have less access to healthcare, more likely to live in population dense areas, more likely to live in a home with multiple people, etc, they're more likely to get COVID. Compared to white people in poverty, they see equal levels of risk to contracting it. Seems the fact was largely reported before they looked at what factors may contribute to su

      • Re: (Score:2)

        by tlhIngan ( 30335 )

        in point of fact some viruses do care about race, this pandemic's one does. Even accounting for disparity in health care, it hits blacks and latinos hardest, asians somewhat harder than whites.

        Is it strictly race based, or is it because of other factors?

        After all, blacks and latinos generally have worse health insurance, if they have health insurance, and are generally lower on the economic scale, which puts them in worse health from the get-go. Thus, they aren't starting from a great place to begin with..

      • Re: (Score:2)

        by dwpro ( 520418 )
        What is accounted for? That link states that race is a marker for other underlying health conditions or even occupations that makes one more at risk, signifying that they didn't control for those variables. That's not accounting for anything useful.
    • The US CDC has defined "close contact" as = 15 minutes, since the thing started here.

      I have no idea where it came from or hat it is based on.

    • Re: (Score:3)

      by dgatwood ( 11270 )

      Secondly, why would the detection rules of different countries vary? Viruses do not care about countries, races, gender, age, etc.

      Detection rules don't exist. What you have is a giant pile of identifiers and distance estimates that gets periodically uploaded somewhere. Then you have different classification rules that are applied to that data after the fact. The classification rules are set by whoever wrote the app that is requesting the data. Different countries set their own rules for who they want their contact tracers to have to call, likely based in part on science and in part on not overwhelming the contact tracers. :-)

      • Re: (Score:2)

        by amp001 ( 948513 )
        Just a clarification, the only thing that gets uploaded is a set of the most recent 14 daily identifiers your device used after you test positive. Other devices download those and use them to generate the more ephemeral identifiers (each daily id is used to generate a new identifier that gets updated every 15 minutes, and those are the ones that actually show up in the Bluetooth protocol) that they have kept in the on-device logs. So, the proximity detection after-the-fact is generated on-device only (using

        • Re: (Score:2)

          by dgatwood ( 11270 )

          Ah. My bad. I assumed they would provide an estimate of proximity and distance from both ends on every identifier, because that would give you significantly more accurate results. (Signal strength is not always the same on both ends of communication; knowing that either end thought that it was within the specified range, then, is probably less likely to miss actual positive results.) But sure. Either way.

          • Re:Bad API/specifications or bad detection rules? (Score:4, Informative)

            by amp001 ( 948513 ) on Friday October 09, 2020 @05:14PM (#60589908)
            Yeah, the protocol level is actually really simple. Chirp your own ephemeral/rotating identifier periodically. Listen for other chirps and record them along with their received signal strength in a log. That's about it until you download the daily identifiers from people who've tested positive and walk your own log. So, yeah, without an actual exchange of information (including signal strength), the proximity detection isn't as accurate. But, the efficiency is quite good and the privacy preservation is excellent.

    • First of all, where did that 15 minutes period come from? What is based on medical findings? Was it the length of time recommended by doctors or virologists?

      Secondly, why would the detection rules of different countries vary? Viruses do not care about countries, races, gender, age, etc.

      There are not many exact answers to be found in a pandemic, only educated guesses. 15 minutes is indeed an arbitrary number, but still based on the application of known science and statistics so far. In the end they had to pick a number. They picked one they decided was a good guestimate for the purpose. If you wish to argue a different number (14, 22, 9 perhaps?) you should send them your credentials and suggestions.

      And it's not the same from country to country because - surprise - not all countries ar

    • Yeah a test in which they move everyone every 15 minutes and then impose a 15 minute threshold is, of course, going to lead to random outcomes because the detection is entirely based on the random timing of the last ping in the 15 minute interval. But that's because the dataset is artificial. Real people don't watch their watch and move after exactly 15 minutes. The point of the software was always to SUPPLEMENT normal contact tracing. The 6ft, 15 minute thing was NEVER intended to be a rigid guideline

      • I always say that the situation that the app is for is where you went to the grocery store for a long trip and didn't notice that you were next to the same person for nearly the whole time. And it was a week ago so you don't remember. Or you went to talk to someone briefly but it ended up taking lot longer, but you don't remember that a week later. It's to catch connections that might be overlooked in standard contact-tracing interviews. Also the app can make exposure notification a LOT faster -- which

      • I never heard about such a 15 minutes rule.
        Makes not really sense anyway.

        The longer you stay at one place, the less people to trace, unless all the others follow a silly 15 minutes rule :D

    • Re: (Score:2)

      by Luckyo ( 1726890 )

      In addition to statement blow by iggymanz that is cares about race, it cares even more about gender. Men are twice as likely to die of this virus as women.

      https://www.nature.com/article... [nature.com]

      It's another reminder of the popular anti-scientific claim of "gender is in your head" is completely anti-scientific in nature. Your gender is biological and fixed. Whatever goes on in your head is irrelevant. This virus kills human males at approximately twice the rate it kills human females. It's not going to care if you

    • Re: (Score:2)

      by Luckyo ( 1726890 )

      And obviously, this virus cares about age. That is the single best predictor of fatal outcomes of infection with it. You have effectively zero death rate for people under 20 and over 10% for men over 80.

      • While the death rate below age of 20 is low, it is not ZeRO.
        And "sex" might be a contributor because of the general bad life style of males (e.g. more smoker - especially in Asia) in relation to woman. Not sure if it has anything to do with a double X versus an XY.

        https://www.worldometers.info/... [worldometers.info]

        Again two of your completely unscientific and WRONG posts :P

        • Re: (Score:2)

          by Luckyo ( 1726890 )

          You know what else is not zero when it comes to death rate?

          Everything. Water poisoning's death rate is not zero.

          Which is why science, unlike the nonsense that you use to base your assumptions on likes to address relevancy. Death toll of under 20s is so small, it's effectively irrelevant. Death toll of over 80s is so high, it's extremely relevant. Realistically, death toll begins to enter relevancy for large scale behavioural change around the age of 60.

          Oh and the fact that this virus kills males at far high

  • US companies employing high salaried developers didn't thoroughly test their API on mass transit? It's almost like there's virtually no mass transit usage in the US...

    • Re: (Score:2, Informative)

      by Anonymous Coward

      A lot of this was developed in Switzerland (Google has a giant office in Zürich), and the first app to use it was released in Switzerland, and Switzerland has an incredible public transportation system. This just isn't designed to catch every case, just be another factor to help push that Rt value below 1.0. Unfortunately, the recent giant parties and clubs have done the opposite here-- I suspect public transportation is not a significant vector.

    • Google is extremely bad regarding anything with public transport. Stations are visible depending on zoom level. Zoom a little bit in, they vanish, zoom more in, they show up again. Most of the time the track they go is not show, or not shown correctly.
      If you walk by foot, it is extremely helpful to know: when I "under cross" the track of that "S-Bahn", I have to go left.
      But no, besides the two S-Bahn stations, there is no track on the map. It is even worse with long distance trains, do you think they have

  • Public transport is the only place where I cannot avoid keeping a distance from other people. Especially morons.

    In shops I can always back awaym But on public transport, people WILL enter the vehicle, even if that makes it impossible to keep a distance. And they WILL touch you while passing, wehnever they freaking can.
    And of course, even if they were willing to run ten times the trains and buses, they simply don't have enough of them. Apart from having to rise the prices by the same amount.

    So this is entire

  • This is completely unsurprising (Score:4, Insightful)

    by HuskyDog ( 143220 ) on Friday October 09, 2020 @05:37PM (#60590002) Homepage
    For anyone with a reasonable feel for how radio waves propagate these conclusions (and other similar ones reported elsewhere) will be of no surprise whatsoever.

    Using radio wave propagation loss to measure distance inside what is effectively a metal box is just doomed to failure. The complex multi-path environment is going to produce numerous local minima and maxima in field strength which will be only loosely correlated with transmitter to receiver distance and furthermore they will all move about all the time as both the passengers with the phones and everyone else moves about.

    Now, no doubt there are things that can be done to reduce the uncertainty and I predict that this is mostly going to boil down to averaging but weighing against this is the fact that these are phones we are discussing and not precision RF measurement equipment. The transmit power is going to vary between models and even individual phones of the same model, but even worse will be the accuracy of the receive signal strength reporting. This was presumably only ever intended to drive a signal strength bar graph with four or five segments so there was no need at all for any sort of calibration.

    I don't in any way mean to criticise the phone manufacturers here. They no doubt designed the bluetooth mechanism to fit four requirements: sufficient range for normal peripherals, regulatory compliance, low cost and low power consumption. At no time will "Sufficient accuracy for range measurement" have been a requirement, but even if it had been, I predict that the intense multipath environment would make accurate measurement challenging even with design optimisation.

  • According to the summary, a "contact" is commonly defined as proximity for at least 15 minutes. Then it is highly unsurprising that you don't get any "contacts" if you move around every 15 minutes.

    But to be fair, many tram/metro/bus rides are probably shorter than 15 minutes, so the rules are not designed to catch those.

Slashdot Top Deals

Experiments must be reproducible; they should all fail in the same way.

Close