Some people believe that automatic contact tracing apps will help contain the Coronavirus epidemic. They won’t.

Sorry to bring the bad news, but IT and mobile phones and artificial intelligence will not solve every problem.

In my opinion, those that promise to solve these things with artificial intelligence / mobile phones / apps / your-favorite-buzzword are at least overly optimistic and “blinder Aktionismus” (*), if not naive, detachted from reality, or fraudsters that just want to get some funding.

(*) there does not seem to be an English word for this – “doing something just for the sake of doing something, without thinking about whether it makes sense to do so”

Here are the reasons why it will not work:

  1. Signal quality. Forget detecting proximity with Bluetooth Low Energy. Yes, there are attempts to use BLE beacons for indoor positioning. But these use that you can learn “fingerprints” of which beacons are visible at which points, combined with additional information such as movement sensors and history (you do not teleport around in a building). BLE signals and antennas apparently tend to be very prone to orientation differences, signal reflections, and of course you will not have the idealized controlled environment used in such prototypes. The contacts have a single device, and they move – this is not comparable to indoor positioning. I strongly doubt you can tell whether you are “close” to someone, or not.
  2. Close vs. protection. The app cannot detect protection in place. Being close to someone behind a plexiglass window or even a solid wall is very different from being close otherwise. You will get a lot of false contacts this way. That neighbor that you have never seen living in the appartment above will likely be considered a close contact of yours, as you sleep “next” to each other every day…
  3. Low adoption rates. Apparently even in technology affine Singapore, fewer than 20% of people installed the app. That does not even mean they use it regularly. In Austria, the number is apparently below 5%, and people complain that it does not detect contact… But in order for this approach to work, you will need Chinese-style mass surveillance that literally puts you in prison if you do not install the app.
  4. False alerts. Because of these issues, you will get false alerts, until you just do not care anymore.
  5. False sense of security. Honestly: the app does not pretect you at all. All it tries to do is to make the tracing of contacts easier. It will not tell you reliably if you have been infected (as mentioned above, too many false positives, too few users) nor that you are relatively safe (too few contacts included, too slow testing and reporting). It will all be on the quality of “about 10 days ago you may or may not have contact with someone that tested positive, please contact someone to expose more data to tell you that it is actually another false alert”.
  6. Trust. In Germany, the app will be operated by T-Systems and SAP. Not exactly two companies that have a lot of fans… SAP seems to be one of the most hated software around. Neither company is known for caring about privacy much, but they are prototypical for “business first”. Its trust the cat to keep the cream. Yes, I know they want to make it open-source. But likely only the client, and you will still have to trust that the binary in the app stores is actually built from this source code, and not from a modified copy. As long as the name T-Systems and SAP are associated to the app, people will not trust it. Plus, we all know that the app will be bad, given the reputation of these companies at making horrible software systems…
  7. Too late. SAP and T-Systems want to have the app ready in mid June. Seriously, this must be a joke? It will be very buggy in the beginning (because it is SAP!) and it will not be working reliably before end of July. There will not be a substantial user before fall. But given the low infection rates in Germany, nobody will bother to install it anymore, because the perceived benefit is 0 one the infection rates are low.
  8. Infighting. You may remember that there was the discussion before that there should be a pan-european effort. Except that in the end, everybody fought everybody else, countries went into different directions and they all broke up. France wanted a centralized systems, while in Germany people pointed out that the users will not accept this and only a distributed system will have a chance. That failed effort was known as “Pan-European Privacy-Preserving Proximity Tracing (PEPP-PT)” vs. “Decentralized Privacy-Preserving Proximity Tracing (DP-3T)”, and it turned out to have become a big “clusterfuck”. And that is just the tip of the iceberg.

Iceleand, probably the country that handled the Corona crisis best (they issued a travel advisory against Austria, when they were still happily spreading the virus at apres-ski; they massively tested, and got the infections down to almost zero within 6 weeks), has been experimenting with such an app. Iceland as a fairly close community managed to have almost 40% of people install their app. So did it help? No: “The technology is more or less … I wouldn’t say useless […] it wasn’t a game changer for us.”

The contact tracing app is just a huge waste of effort and public money.

And pretty much the same applies to any other attempts to solve this with IT. There is a lot of buzz about solving the Corona crisis with artificial intelligence: bullshit!

That is just naive. Do not speculate about magic power of AI. Get the data, understand the data, and you will see it does not help.

Because its real data. Its dirty. Its late. Its contradicting. Its incomplete. It is all what AI currently can not handle well. This is not image recognition. You have no labels. Many of the attempts in this direction already fail at the trivial 7-day seasonality you observe in the data… For example, the widely known John Hopkins “Has the curve flattened” trend has a stupid, useless indicator based on 5 day averages. And hence you get the weekly up and downs due to weekends. They show pretty “up” and “down” indicators. But these are affected mostly by the day of the week. And nobody cares. Notice that they currently even have big negative infections in their plots?

There is no data on when someone was infected. Because such data simply does not exist. What you have is data when someone tested positive (mostly), when someone reported symptons (sometimes, but some never have symptoms!), and when someone dies (but then you do not know if it was because of Corona, because of other issues that became “just” worse because of Corona, or hit by a car without any relation to Corona). The data that we work with is incredibly delayed, yet we pretend it is “live”.

Stop reading tea leaves. Stop pretending AI can save the world from Corona.