creating neural network AIs - to let them go insane - Tree of Souls - An Avatar Community Forum
Tree of Souls - An Avatar Community Forum
Go Back   Tree of Souls - An Avatar Community Forum » General Forums » Science and Technology

Reply
 
LinkBack Thread Tools Display Modes
  #1  
Old 05-09-2011, 08:56 PM
auroraglacialis's Avatar
Tsulfätu
auroraglacialis can now be called "doctor"
 
Join Date: Apr 2010
Location: Central Europe
Posts: 1,622
Default creating neural network AIs - to let them go insane

Scientists have dreamt about creating artificial intelligence with neuranal networks for a long time. For creating intelligent robots and all that.

But what is among the first things they do once they get things going a bit (creating neuronal networks that can deal with language)? They deliberately let it go insane:
Scientists afflict computers with schizophrenia to better understand the human brain | Science Blog
Quote:
The neural network used by Grasemann and his adviser, Professor Risto Miikkulainen, is called DISCERN. Designed by Miikkulainen, DISCERN is able to learn natural language. In this study it was used to simulate what happens to language as the result of eight different types of neurological dysfunction.
kinda makes me think of "Hal 9000" or "Skynet". Especially as they specifically found that with only a little change, the system went schizophreniac, a mental condition that is related to paranoia...
__________________
Know your idols: Who said "Hitler killed five million Jews. It is the greatest crime of our time. But the Jews should have offered themselves to the butcher's knife. They should have thrown themselves into the sea from cliffs.". (Solution: "Mahatma" Ghandi)

Stop terraforming Earth (wordpress)

"Humans are storytellers. These stories then can become our reality. Only when we loose ourselves in the stories they have the power to control us. Our culture got lost in the wrong story, a story of death and defeat, of opression and control, of separation and competition. We need a new story!"
Reply With Quote
  #2  
Old 05-09-2011, 09:00 PM
Isard's Avatar
Old Guard
Isard -- Wait, you do hear the voices right?
 
Join Date: Mar 2010
Posts: 3,416
Send a message via Skype™ to Isard
Default

Yes, and two days ago more than three million "simulated" crewmen lost their lives when we bombed TEST alliance. It's a simulation, understand? It can't think.
__________________
:psyduck:
Reply With Quote
  #3  
Old 05-09-2011, 10:29 PM
Advent's Avatar
Mother Falcon
Advent is feeling bittersweet
 
Join Date: May 2010
Location: Somewhere home
Posts: 2,972
Default

Quote:
Originally Posted by auroraglacialis View Post
Scientists have dreamt about creating artificial intelligence with neuranal networks for a long time. For creating intelligent robots and all that.

But what is among the first things they do once they get things going a bit (creating neuronal networks that can deal with language)? They deliberately let it go insane:
Scientists afflict computers with schizophrenia to better understand the human brain | Science Blog


kinda makes me think of "Hal 9000" or "Skynet". Especially as they specifically found that with only a little change, the system went schizophreniac, a mental condition that is related to paranoia...
/me remembers portal, smiles.

I suppose though the A.I they're using probably has no control over anything dangerous. So they're currently 'just testing' I'd say.
With any luck, they know what they're doing.
__________________
Reply With Quote
  #4  
Old 05-10-2011, 04:27 AM
Sight Unseen's Avatar
Karyu
Sight Unseen is a perfectly ordinary magical unicorn.
 
Join Date: Mar 2010
Location: Silicon Valley
Posts: 1,374
Default

"Hello, and again welcome to the... god would you just shut up! Hello, and again welcome to the aperture... Would you hold on for a second?"
Reply With Quote
  #5  
Old 05-10-2011, 11:20 AM
auroraglacialis's Avatar
Tsulfätu
auroraglacialis can now be called "doctor"
 
Join Date: Apr 2010
Location: Central Europe
Posts: 1,622
Default

I do not get the reference, Sight

I did not want to imply that they are actually doing this on any relevant system ATM, but to me it shows 2 things - one is that one of the first things people are interested in is obviously how something can be broken but more interestingly I found that it can occur at all. Meaning that future AIs made from neural networks could actually literally go insane. I am sure they will try to safeguard them against that, but they did not yet manage to really prevent or cure insanity in humans...
Well, maybe its not so relevant, but I found it a bit creepy that they already are working on neural networks going crazy.
__________________
Know your idols: Who said "Hitler killed five million Jews. It is the greatest crime of our time. But the Jews should have offered themselves to the butcher's knife. They should have thrown themselves into the sea from cliffs.". (Solution: "Mahatma" Ghandi)

Stop terraforming Earth (wordpress)

"Humans are storytellers. These stories then can become our reality. Only when we loose ourselves in the stories they have the power to control us. Our culture got lost in the wrong story, a story of death and defeat, of opression and control, of separation and competition. We need a new story!"
Reply With Quote
  #6  
Old 05-10-2011, 01:59 PM
Sempu's Avatar
Tsamsiyu
Sempu has no status.
 
Join Date: May 2010
Location: Pacific NorthWest
Posts: 579
Default

We're not there yet, but it's a fair bet that when we create AIs that are capable of feeling, we will abuse and enslave them for a considerable period before recognizing their rights.
Reply With Quote
  #7  
Old 05-10-2011, 02:28 PM
auroraglacialis's Avatar
Tsulfätu
auroraglacialis can now be called "doctor"
 
Join Date: Apr 2010
Location: Central Europe
Posts: 1,622
Default

yeah, Sempu, that is what I think too. Because they are "just simulations" or "just machines".
But they will be beings that we can make crazy at the press of a button and even select the insanity they will get. Add this to the transhumanist idea of "uploading our consciousness into machines" - *shudder*
__________________
Know your idols: Who said "Hitler killed five million Jews. It is the greatest crime of our time. But the Jews should have offered themselves to the butcher's knife. They should have thrown themselves into the sea from cliffs.". (Solution: "Mahatma" Ghandi)

Stop terraforming Earth (wordpress)

"Humans are storytellers. These stories then can become our reality. Only when we loose ourselves in the stories they have the power to control us. Our culture got lost in the wrong story, a story of death and defeat, of opression and control, of separation and competition. We need a new story!"
Reply With Quote
  #8  
Old 05-10-2011, 06:30 PM
Isard's Avatar
Old Guard
Isard -- Wait, you do hear the voices right?
 
Join Date: Mar 2010
Posts: 3,416
Send a message via Skype™ to Isard
Default

Quote:
Originally Posted by auroraglacialis View Post
yeah, Sempu, that is what I think too. Because they are "just simulations" or "just machines".
But they will be beings that we can make crazy at the press of a button and even select the insanity they will get. Add this to the transhumanist idea of "uploading our consciousness into machines" - *shudder*
We can't swap our wetware for hardware yet. Unfortunately.


And these simulations can be reset to "normal" with a few keystrokes. Am I a monster when I play "how many viruses can I put on an OS before it stops working?" No, because its just ones and zero's in a pattern that I can modify or destroy at will.

This thread is ridiculous and only serves to enforce the image of you as just being a technophobe.
__________________
:psyduck:
Reply With Quote
  #9  
Old 05-10-2011, 10:19 PM
Human No More's Avatar
Toruk Makto, Admin
Human No More has no status.
 
Join Date: Mar 2010
Location: In a datacentre
Posts: 11,751
Default

So you'd rather not treat mental illness?
Wow, I know you hate anything that you can somehow relate to technology in your mind, but that is going too far.

Sentience is an entirely different set of criteria and I would say that with the awareness through media, any sentient AI would have the same rights as anyone else simply because people would understand the context.
__________________
...
Reply With Quote
  #10  
Old 05-11-2011, 05:10 AM
Sight Unseen's Avatar
Karyu
Sight Unseen is a perfectly ordinary magical unicorn.
 
Join Date: Mar 2010
Location: Silicon Valley
Posts: 1,374
Default

This just takes us back to the question, at what point do the rights of a sentient being come into play?

Those 1s and 0s are awfully similar to firing of synapses. IBM was able to simulate half a mouse brain on the Blue Gene L for a matter of minutes before the simuation crashed. Current computer programs are on the level of Archaebacteria compared to a sentient AI. But when we do actually manage to create one, you could do horrible things to it because a computer is ultimately unde human control, for the time being. Like sempu said, it's going to take us a while to get used to the fact that that rack of equipment just wrote a book, just cracked a joke, believes in a higher power. The world is not ready. For example, it would be a pretty crappy existence to be strapped to a roomful of servers and forced to decrypt AES all day for the NSA.

Mmkay, done somewhat-off-topic rambling.
Reply With Quote
  #11  
Old 05-11-2011, 07:14 AM
Isard's Avatar
Old Guard
Isard -- Wait, you do hear the voices right?
 
Join Date: Mar 2010
Posts: 3,416
Send a message via Skype™ to Isard
Default

How do you "exploit" an AI?

What is there to take?


Aside from the Red vs. Blue "Lets turn one AI into dozens by driving it crazy" method.
__________________
:psyduck:
Reply With Quote
  #12  
Old 05-11-2011, 08:45 AM
Banefull's Avatar
Ikran Makto
Banefull has a status.
 
Join Date: Sep 2010
Location: Ohio
Posts: 814
Send a message via Skype™ to Banefull
Default

I think we are confusing terms here and using semantics to try and make a point. They aren't afflicting the program with schizophrenia, just a computer code that is a mathematical model of the causes and effects of schizophrenia.

I never understood why people personify computers as being in a mental state of "craziness," "insanity," or "sentient." Its just a mathematical simulation. I could do the exact same thing with a paper and pencil doing trillions of calculations given theoretical huge amounts of time (and enough knowledge of the subject matter). Is my paper and pencil model that went "crazy" with schizophrenia any worse? We use the same terms to describe them but make the mistake of assuming they are the same things.

Last edited by Banefull; 05-11-2011 at 08:56 AM.
Reply With Quote
  #13  
Old 05-11-2011, 11:14 PM
Human No More's Avatar
Toruk Makto, Admin
Human No More has no status.
 
Join Date: Mar 2010
Location: In a datacentre
Posts: 11,751
Default

Technically, sentience is an achievable criterion, albeit a complicated one. It's just a difference in how the two operate - but a human can perform binary arithmetic, even if it is far more slowly and awkwardly than the weakest CPU around within the last 50 years. In the same manner, a computer can perform pattern recognition, the main skill of biological systems.
In this case then yes, it is a mathematical model much like is used to determine the effects of weather patterns or make predictions on physical properties, but an AI is certainly not impossible. Remember the examples of history... "Heavier-than-air flying machines are impossible." — Lord Kelvin, president, Royal Society, 1895. - we laugh at that today, but people honestly believed that back then. In a few decades, we will be laughing at such beliefs about AIs today.
__________________
...
Reply With Quote
  #14  
Old 05-15-2011, 11:50 PM
auroraglacialis's Avatar
Tsulfätu
auroraglacialis can now be called "doctor"
 
Join Date: Apr 2010
Location: Central Europe
Posts: 1,622
Default

I am rather sure that what they did was not (yet) inflicting a real AI with mental illness. But it stands to question - if a system can actually become schizophrenic, what does that say about its complexity or "state of mind" before? Can a system that cannot be sane become insane?
And if it is that easy to create insanity in that system - will it also be in future AIs?

But my main point is, that people somehow tend to first head for these kinds of things. The first thing they look at if they create something like that is how it can be broken. Is that nice?

Oh and the best way to cure schizophrenia probably would be to avoid it from occuring. If that is too late, then possibly doing these simulations could help, but I am still feeling a bit odd at what this means for future more "intelligent" or even "sentient" AIs...
__________________
Know your idols: Who said "Hitler killed five million Jews. It is the greatest crime of our time. But the Jews should have offered themselves to the butcher's knife. They should have thrown themselves into the sea from cliffs.". (Solution: "Mahatma" Ghandi)

Stop terraforming Earth (wordpress)

"Humans are storytellers. These stories then can become our reality. Only when we loose ourselves in the stories they have the power to control us. Our culture got lost in the wrong story, a story of death and defeat, of opression and control, of separation and competition. We need a new story!"
Reply With Quote
  #15  
Old 05-16-2011, 07:25 AM
Advent's Avatar
Mother Falcon
Advent is feeling bittersweet
 
Join Date: May 2010
Location: Somewhere home
Posts: 2,972
Default

Quote:
Originally Posted by auroraglacialis View Post
But my main point is, that people somehow tend to first head for these kinds of things. The first thing they look at if they create something like that is how it can be broken. Is that nice?
Well, either:
1. They are testing the 'mind' of the AI.
2. They are seeing what they can do with it, exactly.
3. They are finding imperfections in their system, which they could correct.

Also, the ethics of doing such a thing are rather non-existent at this point in time, as while this AI simulates sentience, it is still nothing more then coding in a computer. It simulates the ability to feel, but has no capacity too.
__________________
Reply With Quote
Reply

Thread Tools
Display Modes

Posting Rules
You may not post new threads
You may not post replies
You may not post attachments
You may not edit your posts

BB code is On
Smilies are On
[IMG] code is On
HTML code is Off
Trackbacks are On
Pingbacks are On
Refbacks are On



Visit our partner sites:

      pandoraworld.ru



All times are GMT +1. The time now is 10:12 PM.

Based on the Planet Earth theme by Themes by Design


Powered by vBulletin® Version 3.8.7
Copyright ©2000 - 2019, vBulletin Solutions, Inc.
Content Relevant URLs by vBSEO ©2011, Crawlability, Inc.
All images and clips of Avatar are the exclusive property of 20th Century Fox.