So, if the judges say only Londoners, how can Scotland Yard then distinguish an EU citizen form a Londoner?

Or from any other country including non-Londoner UK citizens?

The starting point is identifying that the suspect is black. From there, the Met can narrow it down to the black Londoner they want to fit up.

I don't think the Reg went far enough in analysing the numbers. If the "0.0003 percent" is not an error, then there were only 9-10 errors in the whole data set and at an error rate of 0.48% in a sample of 2077 LFR would be responsible for all of them. Even if it's a factor erroneously written as a percentage, though, that still represents LFR being 16 times less accurate than other forms of identification.

We also could use a number for how many of the (presumably) 936 people incorrectly recognised by other means than LR are black, to give context. Because if less than 80% of them are black then LFR is an inherently flawed system with a racial bias, and if 80% or more of them are black then the same thing must be said about the Met.

"Overall, the system's performance remains in line with expectations,"

Whose expectations? False positive rates need to be evaluated in the light of the impact on those affected. If the consequence is an arrest and prevention of the victims' going about their daily business, however urgent they should be financially supported in making a claim for false arrest against the officer concerned - or maybe the Commissioner - and libel against the equipment vendor who should have strict liability.

"and any demographic imbalances observed are not statistically significant,"

Unless the cameras are deployed in locations where 80% of the faces are black the figures do not support this claim.

What happened to the principle of policing by consent.

No doubt the report is using the technical measure of "statistical significance" (where the "p-value" has to be less than 0.05, an arbitrary threshold, whose choice, and significance, is much debated). This technical measure does not align with the everyday meaning of "significance", and should not be used as proof of the importance or real-world impact, or otherwise, of the analysis.

In this particular case, a high p-value (i.e. low statistical significance) is probably just a consequence of the small volume of relevant data, combined with the precise framing of the "null hypothesis" used in the analysis.

The way the calculation works means that statistical significance invariable rises as more data is collected. It can also change if the null hypothesis, and its scope, are framed differently, or if a different threshold value is chosen.

The evident racial discrimination of facial-recognition systems, even if unintended, is clearly at odds with the 2012 Equality Act. IMHO, appealing to "p-values" to deflect from this is an abuse of statistics.

"and libel against the equipment vendor who should have strict liability"

Nonsense. The machine is just an identification tool. Would you urge claims against eye-witnesses who wrongly identify people as criminals? Those who cause harm are those in any official capacity who take action against an individual based on uncorroborated identification by a machine. They should face consequences if they knowingly rely on a system that is not 100% reliable.

And that's the real problem.

Yes the current human rights are a joke. They are stopping us getting things done.

> And of the false positives, 80 percent of them were made on Black people.

Is that the correct summary value to extract? Surely the comparison needs to be of the percentage of false positives for each ethnic group.

For example, consider if there were 100 flagged white faces, with 2 false positives, and 1000 flagged black faces, with 8 false positives. Black faces would be 80% of the false positives. However, the false positive rate for white faces would be 2÷100 (2%) while the false positive rate for black faces would be 8÷1000 (0.8%). That indicates a different bias than is being suggested. The provided summary is unhelpful, without further information.

N.B. Simple statistics example - not an extensive list of ethnic groups. Also, it is a argument concerning the statistics and not suggesting, or explaining, that such a disparity of flagged faces between face groups.

I don't have a problem with the technology itself. How does the error rate compare to human misidentification? The problem is the acceptence, by police, that it is 100% reliable. Anyone wrongly identified should be questioned, of course, but the assumption should always be that the machine might be wrong. If action is taken against an individual, based solely on the identification by a machine, whether that be by police or a store, then there should be financial consequences.

As for invasion of privacy: no one has a reasonable expectation of privacy in public.

WTF!?!? Take your Stasi-style all-pervasive surveillance and sick it up your arse!

It is more than reasonable to expect to be able to walk down the street or cross the road or go into a shop or... without being tracked by PC Plod's snoopercams. Those who think otherwise can fuck off to North Korea or China.

I don't have a problem with the technology itself.

Your should. Lots of problems.

Even if the technology always worked perfectly and never made mistakes - and we know it can't - you have to consider how it will get used and abused by the authorities. I for one don't want to live in a dystopia where everyone is monitored 24x7 and there's a Crapita-run YouTube channel showing everything everyone has ever done when they're outside. Or some PFY in the local nick uses the snoopercams to stalk an ex-boyfriend.

The assumption will always be the snoopercams are never wrong and there will be no way to challenge or disprove whatever "evidence" they produce. And remember too whoever supervises the snoopercams will have a vested interest in preventing anyone from questioning Big Brother's all-seeing eye.

How much is a burqa these days?

Not as much as the cost of getting beat up by Nigel's wannabe stormtroopers.

The judges love to treat the population like serfs. In all but the most egregious of cases, it is nanny knows best. They back the police time and time again.

Yet Pikachu face when the Metropolitan Police is found to be institutionally racist, institutionally sexist, institutionally homophobic, and institutionally corrupt.

That institutional corruption happens because the judges let the police get away with it.

UK can have all the stupid tourists, smart ones won't be visiting.

I thought the stupid ones went to the US?

Cue arrests for having the wrong face and or being dark skinned to continue.

I don't wanna say that Nazis did that... but they did.

Ah yes, won't somebody please think of the children. The classic refrain whenever they want to do something the public opposes.

And either they prevented attacks or they didn't. There's no "potentially". That's just more weasel words desperately trying to justify something intensely unpopular.