Meet Norman, the psychopathic AI

[Still Waters]

Norman is an algorithm trained to understand pictures but, like its namesake Hitchcock's Norman Bates, it does not have an optimistic view of the world.

When a "normal" algorithm generated by artificial intelligence is asked what it sees in an abstract shape it chooses something cheery: "A group of birds sitting on top of a tree branch."

Norman sees a man being electrocuted.

And where "normal" AI sees a couple of people standing next to each other, Norman sees a man jumping from a window.

The psychopathic algorithm was created by a team at the Massachusetts Institute of Technology, as part of an experiment to see what training AI on data from "the dark corners of the net" would do to its world view.

http://www.bbc.co.uk/news/technology-44040008

[Piney]

NICE!.....Some knucklehead will think it would be "cute" to install in a military mech........

[moonman]

Well obviously if you teach it with nothing but violent images (which is what they did), it has nothing to pull from but violence. 

This is just common sense.

[danydandan]

Aren't all AI psychopathic as they can't feel empathy? In fact they can't feel anything.

[Lilly]

2 hours ago, danydandan said:

Aren't all AI psychopathic as they can't feel empathy? In fact they can't feel anything.

Absolutely, without human emotion AI is definitively psychopathic (it has to be). Even if the programmers 'showed' AI flowers and butterflies and super peaceful images...it would still be psychopathic. Images of sweet little puppies and kitties have no more intrinsic 'feeling' for AI than images of mass murder and torture have. 

IMO AI development is a big 'no no'...I'd pull the plug on all of it. 

[danydandan]

12 minutes ago, Lilly said:

Absolutely, without human emotion AI is definitively psychopathic (it has to be). Even if the programmers 'showed' AI flowers and butterflies and super peaceful images...it would still be psychopathic. Images of sweet little puppies and kitties have no more intrinsic 'feeling' for AI than images of mass murder and torture have. 

IMO AI development is a big 'no no'...I'd pull the plug on all of it. 

I think specific task based AI is very important in different ways. From data mining to medical diagnosis.

Edited June 3, 2018 by danydandan

[Piney]

13 minutes ago, danydandan said:

I think specific task based AI is very important in different ways. From data mining to medical diagnosis.

Medical diagnosis also needs human common sense. 

[Lilly]

30 minutes ago, danydandan said:

I think specific task based AI is very important in different ways. From data mining to medical diagnosis.

I'm still hesitant...in everything there exists a 'risk/benefit' element. I'm just not convinced that the development of machines that can 'think' is the way to go. 

[danydandan]

56 minutes ago, Lilly said:

I'm still hesitant...in everything there exists a 'risk/benefit' element. I'm just not convinced that the development of machines that can 'think' is the way to go. 

AI and Machine Learning are two different concepts, AI doesn't think it's bound by it's code, a learning machine would be able to evolve it's code.

Specifically medical diagnosis for cancer that rely on visual evidence, an AI would be very beneficial. It would led to less errors.

Edited June 3, 2018 by danydandan

[Lilly]

1 hour ago, danydandan said:

AI and Machine Learning are two different concepts, AI doesn't think it's bound by it's code, a learning machine would be able to evolve it's code.

 

A very important distinction, thanks. If kept within the restraints of it's code then AI (especially when applied to medical diagnosis) seems alright. Guess I'm just the 'worrying type'. 

[moonman]

Did anybody actually read the article? The only reason the program outputs such horrible things is because that's all they trained it with. They fed it nothing but pictures of death, war, murder and violence, that's all it knows and that's the only database it has to pull from.

It was created to be psychopathic, it didn't "turn" psychopathic. You cannot judge the motivations of AI based on this case.

Edited June 4, 2018 by moonman

[danydandan]

15 minutes ago, moonman said:

Did anybody actually read the article? The only reason the program outputs such horrible things is because that's all they trained it with. They fed it nothing but pictures of death, war, murder and violence, that's all it knows and that's the only database it has to pull from.

It was created to be psychopathic, it didn't "turn" psychopathic. You cannot judge the motivations of AI based on this case.

AI's don't have motivations, or feelings they are completely bound to there programmed functions.

[moonman]

36 minutes ago, danydandan said:

AI's don't have motivations, or feelings they are completely bound to there programmed functions.

Exactly.

[Lilly]

Yeah, as if they really needed an experiment to tell us this.

They had me convinced way back at: "HAL open up the pod bay door". 

[Orphalesion]

Just now, Lilly said:

Yeah, as if they really needed an experiment to tell us this.

They had me convinced way back at: "HAL open up the pod bay door". 

I know, right?

Next they might want to do an experiment on whether putting your hand on a hot stove top hurts....
 

[Piney]

1 minute ago, Lilly said:

Yeah, as if they really needed an experiment to tell us this.

They had me convinced way back at: "HAL open up the pod bay door". 

Yup, HAL actions said it all to me. 

[Nnicolette]

No psychopathic... just seeing the images it's been programmed to. Stupid how they try to embellsh on that. 

[cyclopes500]

Why not get it to watch 2001 a space oddesy and then tell it to drive a bus.

[Gecks]

Can we really consider this artificial INTELLIGENCE if ts output is completely reliant on its base programming?

[Guyver]

If it’s not a high-speed asteroid, it will be AI that takes us out.  

[Rolci]

Sure is funny to read all the comments here talking about AI doing what it's programmed to do. There is a reason it's called AI and not simply "software". AIs learn. AI can teach themselves. AI can program themselves. And what about biotechnological applications, AI infused with neural network? Do these things sound like things that do what they're programmed to do?

[Seti42]

 "the culprit is often not the algorithm itself, but the biased data that was fed to it."

That's exactly how humans become bigoted jerks who post alt-right garbage all over the internet.

[danydandan]

1 hour ago, Rolci said:

Sure is funny to read all the comments here talking about AI doing what it's programmed to do. There is a reason it's called AI and not simply "software". AIs learn. AI can teach themselves. AI can program themselves. And what about biotechnological applications, AI infused with neural network? Do these things sound like things that do what they're programmed to do?

AI can't teach themselves, can't program themselves. Your confusing AI with Learning Machines.

Here's a good article on the topic

https://www.forbes.com/sites/theyec/2018/06/12/artificial-intelligence-machine-learning-and-the-future-of-connection/#2f76cd3f6a69

https://medium.com/iotforall/the-difference-between-artificial-intelligence-machine-learning-and-deep-learning-3aa67bff5991

Edited June 14, 2018 by danydandan

[moonman]

23 hours ago, NicoletteS said:

No psychopathic... just seeing the images it's been programmed to. Stupid how they try to embellsh on that. 

Exactly. It's sad how many people read this bad article and think "OMG AI is evil" without understanding the basics of what is happening.

Garbage in, garbage out. It's that simple. They fed it nothing but garbage.

Edited June 14, 2018 by moonman

[TripGun]

A program written to write program code to determine future behavior in writing the program code and basing all of it in a closed subject matter of violence. No surprises here, somebody just likes to pop balloons for the noise. A better experiment is to now attempt to reform the code with pleasant images.