On the problems with automated contact tracing | Hacker News Reader

On the problems with automated contact tracing

twitter.com

69 pointsdotcoma44 comments

This is spot on and I think the reason why no country has been successful at digital contact tracing. The human factor is so important. In the UK we have recently scrapped our big, centralised contract tracing service because, among other reasons, people have been conditioned by years of robocalls to ignore out-of-area calls. So they have switched to smaller, local contact tracing centres embedded into the community. Plus, it's easy to send someone round if they don't pick up the phone.

Except people have absolutely no obligation to any authority aside from the actual legislation in terms of lockdowns and what you are allowed/not allowed to do. The cummings story was basically the end of anyone actually respecting the government and their policies.

Do I need to take a test? No

Do I need to talk to a contact tracer? Nope

Do I need to share any information? Also no.

Because when the policy is against you every step of the way, you deincentivise people sharing valuable information, or encourage them to lie and perhaps maliciously/falsely claim you were near X person so they need to isolate too or just give them bad data, which is absolutely useless and meaningless.

The government haven't built a society that will respect what they want you to do, as if you do, you end up losing out in the process, such as missing weeks off work without 100% pay which is simply unacceptable.

People have absolutely no faith or trust in the government, and this pandemic has only highlighted many instances where they have gotten policy wrong, or why we shouldn't trust their "apps" (by releasing a trial app without doing a proper DPIA) or follow their policy as you end up getting fucked over.

People are already coming to the UK from abroad and not filling in their 'locator forms', nor are all of them being checked, nor are people being checked if they're isolating [1], which is probably for the best as even with a fine, it's likely better to just not bother giving them your information that will be stored for 7+ years (according to their privacy policy).

So until they fix that, nothing will change.

[1]: https://old.reddit.com/r/ukpolitics/comments/iexgg5/daily_me...

> people have been conditioned by years of robocalls to ignore out-of-area calls.

This is so true!

In a somewhat recent local election, polling was showing the incumbent party winning easily. Of course, on election night, the left leaning party won and displaced the party that lead my province for 40 years, a huge upset that no one predicted.

It was also an election full of robocalls, mainly by the incumbent party. My thoughts on why polling was so off is that the only people answering their phone were those that supported the incumbents.

If you are a person supporting Party X and they are calling each day with a pre-recorded message, you may be more likely to answer. You are more likely to continue to answer your phone and respond to a pollster. If you support Party Y and the only phone calls you get are robocalls from Party X, you stop answering your phone, including those calls from the pollsters. Therefore the pollsters are actually measuring who is willing to answer their phone.

Link without all those ads.

https://mobile.twitter.com/doctorow/status/12962481673194823...

Don't really follow his logic there. Contact tracing is a manual process, thus digital contact tracing apps bad because of that one Michigan app that's actively stupid and a few other failed attempts? That whole argument seems to make major leaps and ignores more sensible approaches to the degree that I'm unable to contact trace whatever point he was originally trying to make. Sure, some people took the "just throw tech at it without thinking" approach but that happens everywhere these days.

The point he's making is that "just throw tech at it without thinking" is this industry's only approach, and it's an approach that, in a situation like this, not only can't help but can't even help but be actively harmful.

I have some relevant experience here. Back almost a decade ago, while I was working as a staff engineer for a local medical school's genomics institute, I was briefly involved with some tooling work for a project that aimed to characterize the spread and prevalence of various STDs among Baltimore's at-risk youth. Contact tracing was of course a major part of that, but where today's (or 2012's!) tech industry would just shrug and stick New Project.xcodeproj in front of an unsecured Mongo store, the doctors and public health professionals running this project built free clinics. Yes, there was an app involved - a tablet app for doctors to use, in documenting what their patients told them about whom they'd been with, when, and so forth. But, unlike some, they knew better than to assume that any app could by itself do the job for them, rather than they themselves needing to go out and serve the community they wanted the chance to learn more about.

You know. Contact tracing. As opposed to, slapping together a pile of React Native and throwing it on an app store in hopes of grabbing some VC bucks and making a get-rich-quick exit - or even Apple and Google getting together and actually engineering something that's technically very sound and still totally useless because it can only ever be purely quantitative in a single dimension, physical proximity, and - as I learned when I was asked to help out with building data models for it - contact tracing is not and never can be a one-dimensional problem.

That's the point Doctorow is making here. And for all that I'm not a big fan of the guy, in this case he's nailed it dead square in the bullseye.

> The point he's making is that "just throw tech at it without thinking" is this industry's only approach...

More subtly, his point is that tech is choosing to do what is easy, not what is necessary.

Tech has had some great successes doing this in the past (see also Worse is Better). It isn't a very good approach though when the relationships are mandatory and the data sensitive.

Sure.

Looking at it in the most generous fashion possible, it's an example of what happens when the low-hanging fruit lasts so long that people forget what ladders are for.

Looking at it with the experience of two decades spent in and around this industry, it's the latest in a long and depressing series of reasons why unicorn factory farming needs to die.

> The point he's making is that "just throw tech at it without thinking" is this industry's only approach

Maybe that's the point he intended to make, what he actually wow is really just a collection of anecdotes about troubles with location based tracking. That's like realizing that it hurts if you hit yourself with a hammer and thus vowing to never again use a screw.

The closest thing be has to an argument against the decentralized proximity approach is a hand-wavy mention of the idea that there might be a possibility of false positives in a traffic jam.

The thing is that the conflict between manual tracing vs electronic exposure notification is a complete fantasy because the resources required to do one have zero overlap with the resources to do the other. Sure, automated exposure notification is wildly imprecise. But so is viral spread, it's a game of probabilities with or without the uncertainties of Bluetooth beacons. But beacon tracing scales where manual tracing doesn't. In a high incidence situation manual tracing will quickly run out of call center agents and in a low incidence situation social distancing and manual location logging (e.g. documentation of restaurant table allocation) will quickly erode to the point where a very, very large fraction of contacts cannot be manually reconstructed. And even in the moderate incidence band where manual tracing works best it has blind spots where beacon logging can fill in, not only anonymous exposures like public transit but also the entire field of "awkward" contacts that Doctorow spent an entire paragraph on. "And then there's this survival sex worker I occasionally meet, here's his number" surely isn't a line that comes up often in a manual tracing conversation.

But yeah, it must be fun to drum up negativity about exposure notification apps, it's a self-fulfilling prophecy when nobody is using them.

You seem to know a lot about the problems of contact tracing! From what demonstrably successful work, or at least training or experience in the field, do you derive your perspective?

And it's not just a simple question of opportunity cost - or, rather, it is, but you're wrong about in which direction. The same people who are building useless junk that nobody uses absolutely could be building tools, like the one I worked on years back, that actually empower clinicians and public health workers to produce datasets susceptible to useful analysis. Doctorow's argument, with which I agree, is that they aren't doing that because it's difficult, exacting, unglamorous work, exactly the sort toward which our industry absolutely sucks at incentivizing anyone. So that's not what happens, and instead we get a bunch of pointless data about physical adjacency that's totally devoid of the kind of context it needs for any meaningful analysis to be possible - but which can be misused to any number of harmful ends.

> The same people who are building useless junk that nobody uses absolutely could be building tools, like the one I worked on years back, that actually empower clinicians and public health workers to produce datasets susceptible to useful analysis.

I disagree: the last thing we need right now is giving the impression that manual contact tracers are building extensive shadow profile databases. Which can be misused to any number of harmful ends. Unfortunately many people would happily focus only on that, promoting the idea that . And they wouldn't even be entirely wrong about it, as illustrated by dozens of cases in supposedly privacy-sensitive Germany where police has been accessing even the most innocuous form of manual contact tracing datasets, paper attendance lists held offline at restaurants. Decentralized rolling beacon tracing is orders of magnitude more frugal with PII. And yes, it's still no substitute for manual tracing. Just like manual tracing isn't a substitute for beacon exposure (and yes, this wouldn't be true at all if we were talking about STD).

You're not wrong about the perceived and actual risk of real, high-quality, contact tracing data falling into untrustworthy hands - in this context very much including those of police; where on Earth did you get the idea I had suggested their involvement in contact tracing work? I said "clinicians and public health officials"; I did not say "cops".

Before being allowed to even access the code, much less the data, for the STI project, I did a week's training on HIPAA and related regulation whose acronyms I've since forgotten, both in the large and in specifically how I could and could not treat patient data and remain in compliance. At no other point in my career, which aside from that one role has been entirely in industry and outside academia, have I received any training on safe handling of PII. The closest anyone ever bothers to come is PCI, which is both not directly related and, as anyone who's ever done it knows, a total joke of a requirement honored entirely in the breach and directed solely at deflecting liability. Which is why our industry has such a shit record with data security, and why the academic institution where I previously worked has had, to the best of my knowledge, exactly one patient data leak in at least the last ten years - and that one was last month.

All of that goes to trust, and trust is a major part of why I emphasized the specific and ongoing effort of providing free sexual health services in my initial comment. Such activity builds significant trust between the research group and the population whose behavior is under study, and that kind of trust is required to get worthwhile data. Don't assume that need for trust only exists in the context of STI tracing among vulnerable communities, either - even your own offhand comment, about "oh by the way" and a sex worker, goes very much to the heart of what I'm saying. If I've engaged the services of such a person, but I'm not willing to mention that to whoever's interviewing me, then the resulting graph will be missing what may prove to be a highly consequential edge. If I am such a person, and likewise decline to disclose, the graph will miss a lot of such edges.

That's why it matters to put in the work of building trust, and that's not something that any app - or, given its extremely sketchy history with PII protection, our entire industry - is able alone to do. That takes people, doing the work, over a sizable span of time.

We haven't had that here in the US, so we were never going to make contact tracing work for this pandemic no matter what else happened. But we might still have a chance to make it work in the next one. If we do, though, it won't be because some Silicon Valley huckster came up with "Facebook for coronaviruses". It'll be because we got it together and did the work. The ecosystem of Silicon Valley hucksters and their enabling VC parasites has no place in that process. Some of the people involved might; there is a role for good, careful, thoughtful engineers, if they're willing to stop wasting their lives on lottery tickets and start doing work that might actually count for something.

Unfortunately, the incentives of the moment remain all in favor of lottery tickets. But I see no reason why that can't, or shouldn't, change. And in the meantime, the very least I can do is explain to people why this isn't at all the simple problem they imagine it to be.

Taking one step back, this applies to coronavirus and data in general. We don't have accurate data on case numbers, because testing capacity has varied with time and between countries, and the precise specificity and sensitivity of the PCR and serology tests, in the real world, are not known. There are biases is reporting of the illness, deaths, those tested. Never before have we had so much data to deal with such a challenge to our society, and yet, the data we do have is not representative of the natural process, and often doesn't help us to understand a model of what is going on.

Getting data on mobile phone adjacency is just adding to the large amount of poor quality data.

It's one of the fields where I confidently claim parity with Cory Doctorow. I wish I could say the same about writing.

Parity of subject matter expertise with Cory Doctorow might mean something in an argument with Cory Doctorow. He's not here right now, at least as far as I'm aware.

To me, the important point is #6

6/ The thing is, contact tracing is high-touch/low-tech because it is a social science intervention. Social scientists have always understood that if you only gather the data that's easy to reach, you'll come to bad conclusions skewed by defects in your collection.

> thus digital contact tracing apps bad because of that one Michigan app that's actively stupid and a few other failed attempts?

Well, what successful attempts should we balance these "few" data points against?

What app can compensate for the difference between

a) I have been a few feet from this person for hours because they slept in the same bed as me and

b) I have been a few feet from this person for hours because they live next door, behind a brick wall.

It's not an issue at all. Including the brick false positive (or drywall) you still get a much better test allocation than blindly guessing. Same for distancing allocation: occasional false positive seclusion would have much better cost/benefit than prolonged blanket closure.

You don't need to be able to perfectly predict the roulette wheel, you'll already find yourself in an interesting conversation with security if you consistently beat the odds between black and red.

A campus map including blueprints solves "b". It's pretty easy to tell whether two Bluetooth broadcasters are in the same room or have a wall between them. Beacon-based asset tracking systems do this all the time.

Campus? I live in a city not a campus.

If we actually wanted contact tracing to work, we'd go with an all of the above strategy, including telling people when they get their positive test to go ahead and immediately call everyone who they can think of that they've had contact with in the past few days.

Analog contact tracing is great because if the tracer is good, they can help evaluate the risk of a situation -- duration of exposure, mask usage, distance, etc.

Digital contact tracing is great because it can potentially reach people who the case doesn't know personally or have contact information for. You're not going to be able to provide a phone number for the person sitting at the next table at the restaurant.

Self contact tracing is great because it can potentially happen faster.

Of course, in order for contact tracing to have much value, you have to have a place for people to quarantine. That seems to be the missing link in many places. There was just a story in the NYT about an immigrant family who lived in a two bedroom apartment with nine people. The mom caught COVID and tried to isolate by sleeping in the closet, but this was not successful. If governments want to get serious, they need to provide out of home quarantine free of charge. This was very successful in Wuhan as a mandatory program, but IMO you would get good compliance if you just told people that they should go to a free hotel room for two weeks and that meals would be delivered.

>> call everyone who they can think of that they've had contact with in the past few days.

How do I call the subway travelers? How do call the people at the pizza parlor? How do I call the people who bumped me in line at the grocer?

Like I said, this is why you want all of the above. Digital contact tracing can do that.

This is, like, just wrong, at least for the implementations I’m familiar with. The intent of the decentralised apps used in Europe is to supplement manual contact tracing; I don’t think anyone has claimed they can replace it. They’re useful for the case of, say, the person who was sitting behind you at the restaurant for an hour, who is at risk but who the manual tracer won’t find.

I would agree that dodgy non-state-sanctioned centralised ones are a problem, and should arguably be banned. But they’re far from the whole story.

Everyone looks at automated contact tracing as an alternative to manual contact tracing, where really they should complement each other.

For a successful app a government needs to weave together both into a seamless user experience.

A lot depend on implementation of the app (is more a concept, is like putting Clippy along some modern well-tought assistant and saying that everyone in every possible situation are bad), and a lot on the stage of the pandemy on a country (or potentially delimited enough region). And also how easily/fast you can get tested and how that scales and how responsible is the average citizen.

In US or Brazil, where the virus is very widespread, it will give you a lot of false alerts (and a few right ones). In countries where the cases are relatively few and spread apart, having good warnings about being close enough with some of those rare cases may be the key to avoid spreading it. Once exposed, you can go around up a lot of days without developing symptoms, an early warning could limit how you interact with others at the very least (and getting yourself tested if in doubt).

But a bad implementation is not a reason to discard the concept, at least, not that alone.

> It doesn't tell you about the circumstances - like, was it one of the people at that eyeball-licking party? Or someone in the next car in a traffic jam?

For most people I suspect that this wouldn't really make much difference.

In my state, approximately 1% of the population is known to have had COVID. That's over the whole course of the pandemic so far. I haven't seen the numbers for number of active cases, but looking at the total case curve, I'd estimate active cases are at least an order of magnitude less.

My chances of actually encountering anyone with COVID are pretty low when I'm out, and I try to minimize the number of times I go out. Even if everyone in my state were using an automated contact tracing app, I'd probably only be getting an alert once a month or so, and that's probably overestimating it.

So, once a month or so I'd need to get a test, and then stay at home for a day waiting for results. Since periodic testing is actually a good anyway, this would not be very disruptive at all. The only real disruption for me would be that for a test in response to an alert, I should stay at home until I get the result. For a test done as part of general periodic testing, there would be no specific need to stay home waiting for the result.

But with alerts only happening every month or so, it would not be a big deal even if most of the alerts are from traffic instead of eyeball-licking parties.

And that's assuming I respond to each alert by getting tested and staying home a day. Instead, I could think about what I did the last two weeks. If I've only encountered people 2 meters away, in passing, with both of us wearing masks (which is in fact true for most two week periods for me), except for people in other cars on the road, then I could ignore the alert.

Edit: I looked up the data for my county. We have 70 active cases in a population of 276000, so about 0.03%. There have been 923 total cases, so about 0.3%.

The Covid App in New Zealand doesn't use the bluetooth apple/google method. Rather, it is based on QR codes that are printed and posted by every business.

I was initially pretty disappointed in this approach, but I have changed my mind, and now agree with Cory Doctorow, and the New Zealand method.

The app just keeps a (local, temporary) diary of when/where you check in. That's pretty much it. If you talk to a contact tracer, you can send them your diary as a way to augment other sources of information: Your memory, your bank statements, your transit card info, etc.

I was disappointed with the app because I thought there was no way it would be effective as a complete contact tracing solution. I now understand that of course _no_ app could do that. Better to have something modest that fits nicely into a wider strategy.

This could have been a nice little writeup of arguments in the decentralized tracing vs centralized location-based tracking approaches. But it is framed in the introduction as an argument in favor of manual tracing vs apps in general and fails massively to live up to that claim by focusing exclusively on the badness of central/location.

Centralized, geo-based apps have a million ways of failing at privacy and real life implementations will likely strike a few? Who would have guessed, that's why the proximity-only decentralized approach exists.

And those "most at risk students" who would be least likely to actively use a privacy-encroaching location tracker? They'll be invisible to manual contact tracing for exactly the same reasons. The situation of those people would make a very good example for how decentralized, proximity only apps can not only be better than location tracking but also play an important role stepping in where the manual approach has blind spots. But somehow it's used as an argument against apps in general.

> And those "most at risk students" who would be least likely to actively use a privacy-encroaching location tracker? They'll be invisible to manual contact tracing for exactly the same reasons.

Yeah he did a real poor job on that point.

It's Twitter. You were expecting a nuanced, carefully planned and well thought out essay?

This tweetstorm by Cory is one of the reasons why I keep trying to persuade people to get a blog (again) rather than rambling at length via tweets. A blog post demands at least a minimum of planning and upfront thought about what you want to say, as opposed to being a stream-of-conscious sequence of bite-sized quotes, hopefully connected. Now perhaps some people do plan their tweetstorms carefully, but the medium hardly encourages that or lends itself to a process of vetting, rewriting and editing.

Twitter has its uses. Long (for some value of 'long' here!) form writing is not that.

While I don't disagree in general, worth pointing out for anyone who hasn't used Twitter "threads" that you can now write all your threaded tweets before posting the first one - so it's not quite as bad as when you had to write a tweet, send, reply to it, send, repeat etc

What if students swap phones, or pick up a burner phone? They can then go off campus with impunity. Apparently phones are used as student I.D. at Albion, but that won't stop anyone. Hide your phone near the entrance to campus when you leave.

Without surgical implants or home incarceration bracelets, you can't realistically track people who don't want to be tracked.

> Without surgical implants or home incarceration bracelets

Don't give them ideas ...

If only this Twitter user had a blog to write on.... (/S he is a famous author and tech blogger and should know better).

Or if only each tweet had, instead of just a page number, also an indication of total number of pages. Like 10/15 ("oh, only five more tweets") instead of a useless count 5 ("oh I'm on tweet number five, that's of no use to me as I've read the previous four")

In the end I gave up after a few tweets . I would have liked to be able to say I gave up half way but there was no way for me to know that.

Edits: there are 23 tweets.

I get what you are saying, but there's something to be said for going where your audience is. I'm just glad this wasn't on Tik Tok.

Edit: Am I hallucinating? I could have sworn the first time I opened this, it went to https://threadreaderapp.com/thread/1296248163653660672.html

Same content on his blog.

https://pluralistic.net/2020/08/20/dubious-quantitative-resi...

Yeah... this is beyond me why people want to write blog posts on twitter using atrocious structure.

Is it because people can start providing feedback while he is writing and he comes up with content on the spot? Or is it because some reading that on twitter is more engaging?

Anyway, I just quit immediatelly.

I think the whole idea of 'contact tracing' isn't really that valuable in the long term anyway. New Zealand recently broke their 100-day free streak, and Australia has implemented strict lockdown measures, as well as South Korea.

Most important is understanding the virus in the community - getting people tested (which they don't have to do), and spotting trends (positive person lives in "X" community, we get data on how many people also test positive in the same place) is more important than knowing whether you were in close contact with your friend Betty or where you were at 13:37 5 days ago.

It's hilarious Google/Apple thought "hmm yes, Bluetooth! The perfect protocol to help prevent the spread of a pandemic", and it has spectacularly, and utterly failed. Their APIs are an embarrassment and should've been removed a long time ago. Their arrogance in thinking that crappy protocol can have any part to play in human health/wellbeing is just mind mindbogglingly idiotic.

Apps will not help a pandemic, apart from in the US where they masturbate over your data and want to know your location every second, who your contacts are, who you talk to, what messages you send them, how they can profile/target you better for ads, and how they can feed "big data" machines to later fuck you/others over with more measured/accurate data or AI systems (like clearview) for law enforcement. Hence the whole no US-wide "GDPR" thing because they really want your data.

If your testing capability is low, and if people are not being tested, then everything else is a waste of time. Then you add in societal failures like not getting sick pay, or paid time off if you've got the virus, or incentives to work regardless and you realize 'contact tracing' is just another data gathering excuse.

Because we don't 'contact trace' the flu, or other sicknesses, we accept them as they happen and are a part of life. As the Coronavirus dies down, it'll be in the same bucket, where people just have it, get over it, and that's all there is to it.

Contact tracing has failed in any substantial form, as the recent outbreaks across the world have demonstrated.

I hate to rain on a good rant, but do you have any actual data to back up all these strong assertions?

Contact tracing (the old fashioned kind) works quite well when it's done aggressively and case counts are low. However, for most of the Western world neither is true, and throwing Bluetooth at the problem isn't going to help. Neither can it magically prevent infections from people who don't know they're contagious, which we've come to realize is a significant feature of how COVID spreads.

Agreed, Australia and NZ are doing aggressive manual contact tracing (in addition to the app that was released in Australia in April).

Last I checked NZ had managed to link all but two of the cases back to the one source.

Victoria (AU) likewise has managed to trace the majority of cases to a breach in the quarantine hotel, because security guards weren't wearing PPE and were sharing a cigarette lighter (if I am correct).

It's part of a tool chest, as is the lockdown, the masks. As much as I am glad not to be locked in Australia/NZ right now, I am proud of how they've been handling it in general.

> If your testing capability is low, and if people are not being tested, then everything else is a waste of time. Then you add in societal failures like not getting sick pay, or paid time off if you've got the virus, or incentives to work regardless and you realize 'contact tracing' is just another data gathering excuse.

This.

I'm sick and tired of people thinking that some clever Bluetooth API is going to somehow magically solve the very real social and political issues that are preventing certain countries from getting the virus under control.

[For a start: what about people who, for all sorts of perfectly valid reasons, don't have a (compatible) smartphone with them at all times? There are a lot more of them than you might think.]

You do what Singapore does and pass out bulletproof hardware tokens (no charging needed/possible) that do the same job.

https://www.straitstimes.com/singapore/10000-seniors-get-fir...

Reply on news.ycombinator.com