The super-recognisers of Scotland Yard | Hacker News Reader

The super-recognisers of Scotland Yard

newstatesman.com

98 pointspepys39 comments

I have prosopagnosia. It's not that I can't recognize people at all; however, I have to rely mostly on haircut styles, mimics and movement patterns, and, above all, voice. I almost never identify actors in the movies (as they change their appearances professionally and I have trouble in remembering actual facial features), and it is out of question for me to remember anyone's face after several months. Especially when the haircut changes.

I thought it is relatively common condition (around 2% of people). But other people seem to be genuinely surprised by it every time: "how could you not recognize me? I have passed directly in front of you?" Sorry.

Studies show that people have trouble recognizing someone out of context about half the time. In other words, you work with John, you run into him elsewhere. You may not recognize him at all or you may realize you know him, but struggle to say from where.

People who are giving you a hard time about this likely are overestimating their ability to recognize faces. Context significantly influences this.

Same here - I don't even recognize my immediate family in a lot of situations. I've had people accuse me of being a snob and pretending not to remember them, but once I know their name I can usually remember a ton of stuff about them and that usually makes them feel better about it.

It's not super uncommon, but a lot of people make it to adulthood without realizing they have it - and some lifestyles make it easier than others.

I wonder if it is actually on a spectrum. I have difficulty, but also can work at it with some success. It was a running joke in our family that I would mis-identify actors on the screen. But eventually I got to be pretty good at spotting certain actors- by studying their facial characteristics.

> “Caballero was gobsmacked,” Rabbett told me. “He then got a bit arsey, saying, ‘I’m not speaking to you until I’ve had a cigarette, I’m getting out of here, you can’t do this to me.’ And then it was: ‘Yeah, it’s gonna be a full hands-up. Let’s get it done.’”

I love English slang. Great article as well, I was surprised that the Scotland Yard group was the only one in the world.

> DCI Mick Neville reckons that it will be ten to 20 years before software is advanced enough to be a useful tool, and even then super-recognisers will still be needed to analyse the results and identify the suspects.

Perhaps augmentation is a better approach than replacement. By that I mean it might be really useful for them to be able to search a database of mugshots or photos for 'people with bigish ears, smallish nose and a spot on the left cheek'.

Are you 'aving a laugh?

> I love English slang.

I had to read that a few times to spot the slang.

I grew up in a poor part of London, on a rough estate, crime everywhere. Aside from the slang (which is typical) the thing I always liked about the interactions with police here was the "hands-up" moment. Obviously not literal as our police don't have guns, but once identified very few people flee, and "it's a fair cop guv'" is also a typical response and rational too.

I'm betting that the system they have is an old pre-NN thing. Certainly not state of the art. The big gains in image recognition has happened fairly recently. Here's a survey of results on the Labeled Faces in the Wild dataset:

https://people.cs.umass.edu/~elm/papers/LFW_survey.pdf

They use L1 Identity Solutions ABIS Face Examiner

> Porritt’s unit has its own software but this has been responsible for only one of the 2,010 identifications made since May 2015. DCI Mick Neville reckons that it will be ten to 20 years before software is advanced enough to be a useful tool, and even then super-recognisers will still be needed to analyse the results and identify the suspects

And yet so much surveillance is based on using facial recognition software to identify 'people of interest. Here's an empirical study showing it doesn't really work.

That's no empirical study. It's anecdote from one police department, and who knows what software they use.

In the best case, maybe some 5-6 year old academic project that got commercialized. In the worst case, maybe something their nephew who's really good with computers wrote ("its own software" suggests some bespoke thing, which makes the latter more likely).

I know. It's L1 Identity Solutions ABIS Face Examiner

I'm amazed that someone (who committed 40+ thefts that) they worked so hard to catch got less than 4 years in jail. Perhaps they're wasting their time on petty criminals.

I suspect that computers will help out quite a bit. Similar to other topics, computers will first outperform most people, then ultimately everyone, but computers plus people will outperform just computers.

In UK in the general case none-violent crimes don't carry sentences like in some other countries.

It's a policy I largely agree with.

4 years in prison is approximately 5% of a typical lifespan. That seems like a lot to me.

He will likely serve 2 years at most, and after those two years one could reasonably predict he'll re-offend. In which case I'm reminded of porridge:

"You are an habitual criminal, who accepts arrest as an occupational hazard, and presumably accepts imprisonment in the same casual manner."

Now, apparently the guy never thought he'd go down for all of his crimes, but he knew arrest was a real risk. Same with prison. I'm afraid I don't have a strong opinion on what the best choices are that a criminal justice system can make. Clearly you aren't going to scare people straight with hardnosed punishments though, because their evaluation of risks are not what you think they should be.

Not sure if you're talking about the individual or reoffending in general, but according to HM prison service the reoffending chance would be 36℅:

http://open.justice.gov.uk/reoffending/prisons/

I was basing it entirely on the guy as described by the article (as in, I was making wild assumptions).

Super cool site. That data would be pretty fun to play with though, age/offense count/previous offences/family etc would all be cool buckets to play with. Plus all the things I don't think of off the top of my head

If your argument is that it takes a lot of resources to apprehend and convict someone, only to send them away for a mere four years, by far the largest cost in all of this is that of keeping them in jail.

It costs around £40k/year to keep someone in jail, not including the opportunity cost of lost productivity.

Will the prospect of an eight year prison sentence make someone less likely to commit the original offence(s) than a four year sentence? Probably not.

Will it reduce the risk of reoffending once they are released? It will probably have the opposite effect.

Could super-recognizers be used for parallel construction? Seems like it would be a great fit.

https://en.wikipedia.org/wiki/Parallel_construction

I suspect the "super-recognizers" are a good cover for some new algorithm for face identification..

"trawling through an image database with more than 100,000 stills.." sounds quite incredible to me.

I have seen organisations manually perform larger, more tedious tasks, for more money, with less useful results, so I'm happy to take it at face value.

Pun intended? If so, nicely done...

A similar effort was taken to successfully identify MLK's killer from his fingerprints, back in 1968.

I don't see the connection. Could you explain what you mean?

Police have information that a suspect will be at a certain place at a certain time. Maybe from wiretap, maybe from informant. They want to arrest the suspect. They do not want the suspect to know about source of information, or even existence of information. They want the arrest to appear as if it was just bad luck. This way suspect does not switch phones or performs mole hunt within their organization.

Police force including an officer who is known as super recognizer is told to be somewhere at some time. Suspect shows up doing nothing suspicious, giving no reason to be stopped and questioned (identification requested) by police. Officer known as super recognizer "recognizes" suspect from book of wanted criminals with 100,000 photos, giving reason for police to approach, apprehend suspect. Officer gives sworn testimony successful arrest was pure good luck.

> To cover up the fact that the Allies were reading Japanese code, American news agencies were given the same cover story used to brief the 339th Fighter Squadron, that civilian coastwatchers in the Solomons observed Yamamoto boarding a bomber and relayed the information by radio to American naval forces in the immediate area.

https://en.wikipedia.org/wiki/Operation_Vengeance

For some reason I thought of Tron.

http://tron.wikia.com/wiki/Recognizer

In that universe, recognizers are "capable of capturing fugitive programs and vehicles"

It'd be fun to try that test - I find myself noting the similarities between people "X looks like Y", and people nod along.

Maybe I'm not material for their special unit, though: one that completely escaped me is that George from Blackadder is one and the same as Dr. House: Hugh Laurie.

Here are some tests for talent (or lack of talent) in this area:

https://greenwichuniversity.eu.qualtrics.com/jfe/form/SV_e3x...

http://www.abc.net.au/cm/lb/6744428/data/david-white27s-face...

http://www.testmybrain.org/SupersRecruitment.html

http://www.faceblind.org/facetests/ff/ff_intro.php

http://www.bbk.ac.uk/psychology/psychologyexperiments/experi...

I'm not totally face blind, but I'm definitely impaired. I recognize people I see regularly just fine if they are in a familiar context. In other situations, rely more heavily on voice, gait, and haircut. The idea of choosing movies based on "favorite actor" makes no sense to me.

On tests such as these, I'm happy when I get the "do you understand the instructions" training questions right. Sometimes I can pick a face that I've just seen out of a short lineup based on verbal description of a single feature ("bushy eyebrows" or "full lips"), but usually I'm just purely guessing, hoping that my subconscious has the fabled "blind sight" that I can't consciously access.

I just did the greenwich one. That was fun - thanks. One complicating factor is that they re-use faces in the lineups, so you end up recognizing people from two faces ago. (intentional, no doubt).

I did that one, as well. 9/14, which is a lot better than I expected - I did "cheat" though. I picked details on their faces, and concentrated on remembering them.

I have aphantasia (to a degree). For example, I had difficulty recognizing my mother when she dyed her hair a few years ago.

Thanks for introducing me to aphantasia, which I hadn't heard of. For others that haven't, it means not being able to imagine images or sounds in your mind. It was only discovered recently, and I think it was popularised by this blog post by Blake Ross, which is well worth a read.

https://www.facebook.com/notes/blake-ross/aphantasia-how-it-...

Thanks for these.

I have realised that I am perfectly average at recognising faces getting 79% on the bbk test and 10 out of 14 on greenwich, however I don't recognise people that well. I hadn't realised there is a difference.

Anybody come across this?

Ouch, I got 31% correct in the first one (average is 85%). I always knew I was bad with faces, but I guess I might have prosopagnosia.

As someone who can barely remember faces, i find this quite amazing.

Another thought is - how long before AI recognition would eliminate this job. 20 years they give sounds like an enormous over-estimate.

What is up with this site? The tab rapidly climbs to 1.8 GB of memory and 100% CPU utilization, on the latest version of macOS and Chrome.

I'm surprised face recognition software isn't used in the high-end stores.

maybe humans do better in facial recognition than NEC software. But I found no statistical evidence for that.

Just a note on the title: it appears to be an homage to Warren Buffet's classic speech "The superinvestors of Graham-and-Doddsville"

Reply on news.ycombinator.com