Hacker News Reader

Comment by himinlomax | Hacker News Reader

Ah yes, deploying a technology with obvious potential false positives and operating without the assumption that it has false positives. Geniuses right there.

I remember a story where after 9/11 and during the dirty bomb panic that ensued, a network of radiation detectors was deployed around New York. Authorities identified many dirty bombers. It seems Al Qaeda had recruited cancer patients for some reason, and they all congregated around radiation therapy centers for some nefarious purpose.

Those people shouldn't be fired for infringing upon this gentleman's civil rights, they should be carted off to a secure facility for the severely mentally impaired, for the good of society. I don't want any of them to be allowed to operate heavy machinery (including vehicles and elevators), matches or knives; the risk is too great.

View on HN

This has been going on for a while and has nothing to do with AI or tech.

> Drug tests generally produce false-positive results in 5% to 10% of cases and false negatives in 10% to 15% of cases, new research shows.

Lots of innocent people were probably jailed because of that. Afaik, most countries still relies on these tests to at least detain people. The system will implement something (anything?) as long as it is "good enough" or appear to fix the problem.

> Drug tests generally produce false-positive results in 5% to 10% of cases and false negatives in 10% to 15% of cases, new research shows.

Most often, some tests have high sensitivity (no false negatives) whilst others have high specificity (no false positives). (Rarely both.)

So, for example, this is the reason why a positive RT-PCR test has more weight than a negative test because of the test’s high specificity but moderate sensitivity.

Lumping together false positive stats (for low-specificity tests) with false negative stats (for low-sensitivity tests) is misleading at best.

> > Drug tests generally produce false-positive results in 5% to 10% of cases and false negatives in 10% to 15% of cases, new research shows.

> Lots of innocent people were probably jailed because of that.

Efforts to increase the accuracy of the tests have downsides too, since the tests are already too sensitive. Lots of people have lost their jobs due to erroneous true positive results from eating something like a poppyseed bagel.

The test result is erroneous because no drugs were used, but a true positive because the trace amounts detected from the poppy seeds were actually present.

Improving the test can't fix this.

Like one Florida Man's Krispy Kreme donut crumb that tested positive for methamphetamine? https://www.npr.org/sections/thetwo-way/2017/10/16/558147669...

> deploying a technology with obvious potential false positives

"Okay, let's run this through the system."

"What, no match? This system sucks."

"Let's try the competitor's system."

"56% match! We're in business."

People just are not prepared/don't understand how even very low level of false positives looks like in real world.

If a vaccine kills 1 in a million people, that's still hundreds of people in US and thousands in entire world.

But now, see, how most people just don't understand the concept of very large/small numbers and can't interpret what it means exactly.

There are tests with very high false positive ratio. For example, pregnancy test.

Now, you don't do pregnancy test all the time. If you did it 5 times a day every day, basically at least once a week you will find out that you are pregnant and the test would be useful. And then you go to a real doctor to confirm.

The way to use test like that is to use it combined with some other knowledge/process/situation. In case of pregnancy test you only do pregnancy test when a) you show other early symptoms of pregnancy and b) you had sex with possibility of getting pregnant.

Facial recognition could be useful but as any high false positive test, looses its usefulness when deployed on large scale, indiscriminately.

> But now, see, how most people just don't understand the concept of very large/small numbers and can't interpret what it means exactly.

I think the argument is that this particular form of innumeracy should no longer be acceptable if you're making decisions where this matters. At a minimum, this would be police officers, judges, lawmakers.

But they are normal people and so are our lawmakers.

Unfortunately you can't build the system based on assumption that officers, judges or representatives are Vulcan and only use logic. Any system like that will necessarily be pretty bad.

The safety in the system is general public scrutiny, but as we see it is failing miserably, too.

> But they are normal people and so are our lawmakers.

Okay, then this kind of innumeracy is no longer acceptable in society, full-stop.

So how do you propose to do that?

You know, this might be shocking, but around half of all people have less than average intelligence...

Do we euthanize half of population?

You know, then a significant portion of surviving population will have less than average intelligence...

Most people are not qualified to be doctors. We don't just throw up our hands and conclude genocide would be the only answer and it is unacceptable therefore we need to let anybody do organ transplants.

Statistical literacy isn't so low we cannot make it a requirement to be a judge. The gatekeeping for once would be the start of a solution - make statistical literacy a qualification for well paying jobs and many will steer themselves to it.

Yes, but in a lot of cases facial recognition/drug testing is done without a valid prior reason, just the office pulling the "I smell weed" crap -- there is no incentive for the police officer not to make up an unfalsifiable prior reason.

Soo... what I just said?

The point is, for facial recognition to make sense, you would need to have some additional context.

For example it could make more sense if you followed up any results with an actual person making determination (some variation on a blind or double blind test).

> as any high false positive test, looses its usefulness when deployed on large scale, indiscriminately.

Well of course anything mis-applied is probably bad. But sometimes false positives can be a good thing. For example, in immunology, it's quite common to use cheap tests that have quite high false positives, but low false negatives to screen samples before applying more expensive, but more rigorous tests.

cough cough PCR tests cough

With that cough you shd get one.

Yes, what a coincidence that as soon as the cold/flu season is approaching covid symptoms are equal to cold/flu symptoms

Where does the flu season approach in June?

https://www.nytimes.com/2020/08/29/health/coronavirus-testin...

Reply on news.ycombinator.com