AI can predict who will be criminals based on facial features

[System Error Message]

a lot of faces of what people would think are innocent are infact criminal and vice versa. Unless you can peer into peoples minds you will never know. I prefer to use the alternative, rather than detect crime deter it by influencing people not to do it.

[bigneo]

On 11/19/2016 at 7:15 PM, Ryan_Vickers said:

 when has anything ever been absolutely 100% correct, every time? 

 

Me and Food

[patrickjp93]

5 hours ago, DrMikeNZ said:

I had read it, it has been a great laugh. If it was submitted to either J ACM or IEEE it would seem odd for it to have been released outside of their journals, unless it was rejected. But I would not be surprised if it was submitted unsuccessfully to both.

I cannot imagine any university that would accept that as a masters thesis.

It was successfully submitted to both.

 

Carnegie Melon did, so seriously, shut up and actually read it.

[DrMikeNZ]

35 minutes ago, patrickjp93 said:

It was successfully submitted to both.

 

Carnegie Melon did, so seriously, shut up and actually read it.

Very funny.

[Sauron]

Minority Report anyone?

[MrDynamicMan]

On 11/19/2016 at 10:16 PM, mvitkun said:

Psycho-Pass IRL

Except, like, not way as accurate.

[laminutederire]

14 hours ago, patrickjp93 said:

 

And what face structure has to show about what happens in their head? 

Data science has shown what? Please enlighten me. To me that looks like a bunch of guys overfitting different classifiers to get good scores, then trying to pass it as true by finding a reason of their classification.

Well... lot of crimes were passed on to corporations when their ceo were as guilty. That created bias for white collar criminal who can avoid conviction because of that transfer of responsability, as well as the use of good lawyer to fool jurys.

Since you admit it has no sense to say their criminal because of their face, why would you try to recognize them based on that?

How do you want people to take you seriously when waging something like that without any justification other than "it's better on that point". I looked at your system, and European systems, and frankly I prefer to have a lawsuit in Europe. I'll explain why later.

[laminutederire]

11 hours ago, DrMikeNZ said:

I had read it, it has been a great laugh. If it was submitted to either J ACM or IEEE it would seem odd for it to have been released outside of their journals, unless it was rejected. But I would not be surprised if it was submitted unsuccessfully to both.

I cannot imagine any university that would accept that as a masters thesis.

" Absent, too, is any discussion of the incredible potential for abuse of this software by law enforcement. Kate Crawford, an AI researcher with Microsoft Research New York, MIT, and NYU, told The Intercept, “I‘d call this paper literal phrenology, it’s just using modern tools of supervised machine learning instead of calipers. It’s dangerous pseudoscience.” " Does that resume your thoughts? (It's from the article of the source of this post)

Re read the article more thoroughly, it's damn scary...

Half of the criminals probably aren't criminal since they are only in the wanted list of the Chinese government... so they may just be enemies of the state because they want freedom and say it loud enough, while still being smart enough not to be arrested. That starts well. 

[Real_PhillBert]

I feel like I've seen this movie before. Oh wait, I have!

 

 

[samcool55]

This should be easy to test.

Get identical twins, let one commit a crime, let the thing figure out which one did it.

 

I'm sure it will fail

 

[patrick3027]

Funny how no one has thought of plastic surgery in this thread to 'bypass' such a system if it would be implemented.

[DXMember]

It's not the first time I hear this...

What the hell?

[SpaceGhostC2C]

I've just noticed another issue with the title, that may seem semantic but it's not: it states that the AI can "predict who will be criminals", but it does no such thing. There is a difference between fitting past events and predicting the future. You can do "out of sample" tests of forecast ability, but you are still using data of people who already (allegedly) committed some crime. Ultimately, you would need this AI to predict criminal behavior within a sample of people with clean records, and see how many of those do become criminals at some point (that is, provided all the other caveats were addressed). But if they have a clean record now, the AI would be tuned to predict them as innocent. In other words, to conduct this exercise they had to assume that everyone who is not a criminal yet would never be. Hence, by construction, it's pretty much useless for prediction purposes.