UK police say 92% false positive facial recognition is no big deal | Hacker News Reader

UK police say 92% false positive facial recognition is no big deal

arstechnica.com

18 pointsneverminder8 comments

Reminds of similar stats on stop and frisk - where only a small percentage of stops actually resulted in some sort of criminal conviction. And only a tiny fraction of those convictions were related to the original reason for stop and frisk[1] - discovering illegal weapons.

From the cop's perspective they're looking for a needle in a haystack. Public pressures are especially intense when it comes to anticipating and preventing terrorist attacks or curbing violent crime. But predicting rare events is very difficult. Due to the nature of the problem it's going to be mostly haystack (false positives) most of the time.

So I can understand why the UK police remain unperturbed by 92% false positives. And why the NYPD was largely indifferent to a similarly high rate of failure with stop and frisk.

The problem though is we are the haystack. There's a bit of the snake eating it's own tail here. In order to keep you safe we need to own your privacy. In order to protect your person from violent crime must be able to search your body at any time. Whatever freedom might be I can be sure freedom is not that.

[1] https://www.theguardian.com/world/2013/nov/14/stop-and-frisk...

Whatever freedom might be I can be sure freedom is not that.

Yeah, it's all kinds of throwing the baby out with the bathwater.

It's common for parents to tell kids stuff like "If someone other than this short list of names wants to touch your private parts, scream." And I wasn't comfortable telling my kids that. I felt that would be robbing them of the very innocence I desired to protect.

Instead, I taught them that affection required permission and if anyone did not respect their answer of "no," come get me. That worked remarkably well.

I don't know how to translate that to adult security issues out in the world. But there are definitely inherent problems with the current course we are on that increasingly reminds me of what I heard about East Germany: that everyone was reporting on everyone just to try to cover their own butt because everyone lived in fear of the government and it's surveillance state tactics.

The adult response is to acknowledge the trade-offs involved and to learn to live with a certain amount of risk. A free society is always going to be vulnerable to terrorist attacks to some degree. Reasonable policies might reduce this risk by 25%, 50%, 75%, 90%, etc. But they'll never reduce the risk to zero. So the conversation can't be how do we make sure this never happens again. Instead it must be an analysis of the trade-offs involved in securing against rare events while maintaining the freedom of movement and privacy that a free society requires.

Facial recognition in the UK, the mass phone metadata collection in the US and abroad, stop and frisk in NYC; public support of or indifference to these measures depends on the perception of the trustworthiness of government. That is, for populations that perceive that government as "on their side", the tradeoffs these policies require appear innocuous or beneficial.

So what if the NSA is storing my metadata? They're the good guys trying to catch terrorists. From this perspective the risk of bad government actors is smaller than the risk of terrorism.

So what if cops go hands-on with someone walking down the street? They're finding illegal guns and contraband, and bad guys with warrants. The inconvenience of being stopped is a cost worth paying for safer neighborhoods.

But the fulcrum of government can quickly swing from trust to oppression. There are decades where nothing happens; and there are weeks where decades happen. We cannot keep evil people from attaining power. But we don't have to give them rope with which to hang us. Privacy protections, respect for bodily autonomy, adherence to the rule of law, and the decentralization of economic and political power are all bulwarks against tyranny.

Again, as mentioned above, read up on East Germany. All your questions about "why is this bad?" have already been lived by tens of thousands of hapless people.

"So I can understand why the UK police remain unperturbed by 92% false positives."

Automating bad judgement. Just think of the savings.

I saw something similar with my local election administrators. They enthusiastically embraced automated signature verification (postal ballots), even though it didn't work very well, ultimately creating more work.

I just couldn't wrap my head around it. I think it was mostly about funneling money to cronies.

The problem isn't the inaccuracy, it's that people will inevitably treat it as far more accurate than it is.

As long as it's just flagging people and not arresting/harassing them, I don't think the error rate matters that much.

Even a system that says "there's a 10% chance this guy is a wanted criminal" seems useful.

The probability of any positive, and the probability of a false positive are quite distinct things. It makes a big difference if, of that 10%, 99% are false positives, or 1% are.

If 99% are false positives, then there is another form of harm being done even if the police are not abusive. Time and energy are being spent to vet those positives. It's like medical tests that say you have cancer. If there are a lot of false alarms, people waste effort and don't catch the real issues.

The number of false negatives also matters, and is also not determined by your 10% figure. That is, out of the 90% that weren't flagged, how many should've been?

Reply on news.ycombinator.com