New Program Judges If You’re a Criminal From Your Facial Features

H

HardOCP News

[H] News
Joined
Dec 31, 1969
Messages
0
I'm not saying this study is a load of crap...actually, I think that's exactly what I am saying. A program that judges criminality based on facial features? Seriously? Half the people in the study were convicted criminals. Call me when this program can pick one out of a thousand with one hundred percent accuracy.

The images used in the research were standard ID photographs of Chinese males between the ages of 18 and 55, with no facial hair, scars, or other markings. Wu and Zhang stress that the ID photos used were not police mugshots, and that out of 730 criminals, 235 committed violent crimes “including murder, rape, assault, kidnap, and robbery.” Here’s how it worked: Xiaolin and Xi fed into a machine learning algorithm facial images of 1,856 people, of which half were convicted criminals, and then observed if any of their four classifiers—each using a different method of analysing facial features—could infer criminality.
 
“All four classifiers perform consistently well and produce evidence for the validity of automated face-induced inference on criminality, despite the historical controversy surrounding the topic,” the researchers write. “Also, we find some discriminating structural features for predicting criminality, such as lip curvature, eye inner corner distance, and the so-called nose-mouth angle.” The best classifier, known as the Convolutional Neural Network, achieved 89.51 percent accuracy in the tests.

I think that this study has merit if you talk to everyday people discussing police shootings in my area. An 89.51% probability of being a criminal is easily enough justification to shoot someone around here. It is surely enough to arrest someone and hold them for it. Add in some Stop and Frisk and I bet they could get the percentage to 93%. Inconveniencing innocent citizens with a jail term until they can prove their innocence, or accidently shooting the other 7% has enough risk / reward justification if you were to listen to the people I meet on a daily basis. It would be well received here.

What if they just placed the people that look like criminals into an internment camp? What harm would that do? They would just have to stay there until they went through an extensive rehabilitation program. Even if some went that were innocent; how could this adversely affect them in the long run?


I guess this would really work well in countries like India, Pakistan, China, etc where there is such a large population that nobody even cares if a few family members go missing.


/sarcasm and /truth at the same time. ;)
 
Wait till the algorithm finally decides that only humans can become criminals so it just kills all humans. :)
 
It could be the new Minority Report. You look like a criminal, so we snuff you out before you commit any crime. As a result, crimes have dropped (as well as the population problem).
 
So... Phrenology~ish pseudo science makes another round? Considering all of the other stupidity we see daily, is it really a surprise?

That said, I'm growing more and more intolerant and angry at the ignorant hatred...
 
steakman1971 said:
It could be the new Minority Report. You look like a criminal, so we snuff you out before you commit any crime. As a result, crimes have dropped (as well as the population problem).
Click to expand...

In some countries abortion is illegal even in cases of rape. Could you imagine raising a kid for 18 years for him / her to go to jail on their 18th birthday because they have similar facial characteristics as a known rapist?
 
Let it examine photos of the all government officials. Then you can MAYBE examine the gen pop.
 
Let me guess, I just picks all the black and hispanic people, and just assumes they are criminal, just like the real police? :p
 
Zarathustra[H] said:
Let me guess, I just picks all the black and hispanic people, and just assumes they are criminal, just like the real police? :p
Click to expand...
As long haired white boy, I feel I must alert you to a profile to be added. I get just as many nasty looks as any minority.
 
cageymaru said:
“All four classifiers perform consistently well and produce evidence for the validity of automated face-induced inference on criminality, despite the historical controversy surrounding the topic,” the researchers write. “Also, we find some discriminating structural features for predicting criminality, such as lip curvature, eye inner corner distance, and the so-called nose-mouth angle.” The best classifier, known as the Convolutional Neural Network, achieved 89.51 percent accuracy in the tests.

I think that this study has merit if you talk to everyday people discussing police shootings in my area. An 89.51% probability of being a criminal is easily enough justification to shoot someone around here. It is surely enough to arrest someone and hold them for it. Add in some Stop and Frisk and I bet they could get the percentage to 93%. Inconveniencing innocent citizens with a jail term until they can prove their innocence, or accidently shooting the other 7% has enough risk / reward justification if you were to listen to the people I meet on a daily basis. It would be well received here.

What if they just placed the people that look like criminals into an internment camp? What harm would that do? They would just have to stay there until they went through an extensive rehabilitation program. Even if some went that were innocent; how could this adversely affect them in the long run?


I guess this would really work well in countries like India, Pakistan, China, etc where there is such a large population that nobody even cares if a few family members go missing.


/sarcasm and /truth at the same time. ;)
Click to expand...

In the paper, I don't find their false negative rates, but that false positive rate should be considered alarming. Your misunderstanding is a common one - that the test shows an "89.51% probability of". Instead, it correctly rejected the null hypothesis in 89.51% of cases. The Type 1 error chance (that is, where the person is not a criminal but gets classified as one) means that the conditional probability that someone is actually a criminal given that the test identified them as one:

P(actual criminal | identified as criminal) = P(actual criminal & identified as criminal) / P(identified as criminal)

The sample set they used for training had a population of 1,856. Of these, 730 are criminals. That means the data set is trained with the assumption that ~39% of the population is criminals. This is the first major issue I see, as that would be a considerable portion of the population. However, it's their stats, we'll use them.

P(actual criminal) = 0.3933
P(actual criminal & identified as criminal) = 0.8951

Code: 
                           |  criminal |  non-criminal  |  Totals
identified as criminal     |  3,520    |  742           |  4,262
not identified as criminal |  413      |  6,325         |  6,783
Totals                     |  3,933    |  7,067         |  10,000

Out of the population that is identified as criminal, 17.4% are not criminals. Of those not identified as criminals, 6% actually are.

If you maintain the same level of accuracy, but change the P(actual criminal) to be 10% of the population, the numbers change as follows:

P(actual criminal) = 0.1000
P(actual criminal & identified as criminal) = 0.8951

Code: 
                           |  criminal |  non-criminal  |  Totals
identified as criminal     |  895      |  944           |  1,839
not identified as criminal |  105      |  8,056         |  8,161
Totals                     |  1,000    |  9,000         |  10,000


Now, 51.33% of people identified as criminals by the test that is "89.51%" accurate are not criminals. Only 1.287% of those not identified as criminals are actually criminals. In other words, if the population is 10% criminal, 18.39% of the population would be identified as a suspected criminal, but the chance they actually are criminals after that identification is worse than a coin flip.

Keeping in mind conditional probability and the actual population chance of something being true is very important, but this paper skips right over that.
 
DrLobotomy said:
As long haired white boy, I feel I must alert you to a profile to be added. I get just as many nasty looks as any minority.
Click to expand...

Well, I guess police are the same everywhere. I get dirty looks all the time and frisked all the fucking time. Never seen them stop and bother anyone with a suite, though. Ever

Police = face control of the steets
 
dgz said:
Well, I guess police are the same everywhere. I get dirty looks all the time and frisked all the fucking time. Never seen them stop and bother anyone with a suite, though. Ever

Police = face control of the steets
Click to expand...
I'm actually ok with profiling some things.

Lets say that you are a black guy walking down the sidewalk. On one side of the street there are 5 white guys in suits going towards you on the sidewalk. On the other side there are 5 white guys going towards you with sleeveless leather vests and ss lightning bolt tattoos. Which side of the street would you want to be on? Was it because they were white or was is due to other clues?

You choose your clothes and how to project yourself by those choices. You don't choose a certain color or race.
 
/usr/sbin said:
I'm actually ok with profiling some things.

Lets say that you are a black guy walking down the sidewalk. On one side of the street there are 5 white guys in suits going towards you on the sidewalk. On the other side there are 5 white guys going towards you with sleeveless leather vests and ss lightning bolt tattoos. Which side of the street would you want to be on? Was it because they were white or was is due to other clues?

You choose your clothes and how to project yourself by those choices. You don't choose a certain color or race.
Click to expand...

Let's say you're a street patrol in that situation. Your options are to determine what is the minority person guilty of if they go with the suits or watch them get fucked up by the skin heads
 
You must log in or register to reply here.
Back
Top