tech
The False Positive Paradox
One thing more people need to be aware of is the False Positive Paradox.
If you have an app that can identify people with 97.6% accuracy, that’s pretty good right?
No! When looking for a small sample in a large enough population, it’s really bad.
I ran the numbers for a facial recognition app with a 97.6% accuracy rate, looking for a group of people that make up approx 3% of the US population. Of the ~8 million matches it will find, with a 97.6% accuracy rate, around 96% of the matches will be false positives.
When the thing you are looking for is rare, you need a margin of error that is significantly lower than the rarity of the thing you are searching for. Otherwise you get some unintuitive (and potentially alarming) results.
When you deploy those tests at scale, your results become garbage.
That 97.6% accuracy rate is from facial recognition is for algorithms being tested in ideal conditions. Not when the subject is a terrified, trembling person being filmed in a snowstorm. In those conditions, I can guarantee that the accuracy is significantly worse.
Anybody who claims an app can provide a ‘definitive’ identification in the field is lying to you.
Anybody who claims such an app should supersede provided documentation is malicious.
Anybody who buys into those claims, I happen to have a bridge for sale (and I don’t feel bad about selling you it).
Response from my MP
Hot network closets in your area
One of our network node rooms at work had it’s air con die yesterday (Friday) afternoon. It’s not considered critical and we are in a budget squeeze, so the team who are responsible for it have decided it can wait over the weekend. They have however, not turned off the alert emails. So now I’m getting a nice hour by hour view of the temperature in an uncooled network closet until things start turning off. It’s hit almost 42C overnight and is still on the rise… I’m not sure what the shut off point for the kit in there is, but it’s clearly still running…