maddogdrivethru.net

Open all night
It is currently Wed Jun 20, 2018 3:24 am

All times are UTC - 5 hours [ DST ]


Forum rules


It's the Gulag of Fun



Post new topic Reply to topic  [ 25 posts ]  Go to page Previous  1, 2, 3
Author Message
 Post subject: Re: AI and Ethics: Overcoming the Risks
PostPosted: Thu Mar 29, 2018 5:59 am 
Offline
His Most Gracious Majesty, Commie of the Year
User avatar

Joined: Sat Feb 28, 2009 12:44 pm
Posts: 7718
Reputation points: 17050
Humans indeed design the system that is able to learn.

After that the system itself develops skills that it is capable of developing on it's own. In fact it can be difficult for humans to grasp what is going on within the system - even the designers.

The term is 'emergent capabilities' - we make a system and the system on it's own emerges with capabilities we didn't program. This is what it means that it learns. It develops capabilities that weren't programmed into it by utilizing the programmed ability to learn.

AI has immense advantages and but it is not god, it is not an entity without physical limitations.


It suffers from all the same limitations we do -
    available computing power
    available storage space
    precision of sensory input
    reliability of sensory input
    precision of saved information
    reliability of saved information
    speed at which information can be saved
    speed at which saved information can be accessed
    time sensitivity of certain decisions leading to less well thought decisions
    all the problems associated with filling in missing vital information
    all the problems associated with fully or partially wrong information
    when to replace existing models with newly proposed ones
    etc

This means that not even an AI can be expected to always come up with the right answer. An AI might learn to become a devout Nazi or it might believe that there are reptilian aliens wearing human skins.

As for battlefield solutions, humans themselves have a builtin system for not just coming up with spatial mathematics, vectors etc. of dynamic systems - humans themselves have the ability to create forecasts of possible futures and how other actors in the situation may act. Fighter pilot anticipating the moves of his enemy and flying for a while without seeing the enemy and ending up in perfect kill spot is just an example of this, much of this is 'intuitive' and we vary greatly in our capacity at it.

I expect learning AIs to similarly express different models and differing performances as they adapt alternate primary and secondary approaches, algorithms and learn different things and have access to different information and experiences.

That said, I expect AI combat units to reasonably quickly to be able to consistently outperform humans.

    AI makes better decisions
    AI predicts overwhelmingly better, creating thousands of scenarios in an instant
    AI makes 'insanely' precise trigonometry
    Robotic soldiers have overwhelmingly superior
    -durability
    -firepower
    -performance
    -consistency/uniformity
    etc.

However, they also have a cost. You don't just *will* 2 million Terminators into existence, flying machines are exponentially more expensive so if you can't afford 5,000 fighters today you won't afford them because you have an expensive chip to put in them.

However, you can get 2 million riflemen with minimal level only requiring you to call them to arms with their own weapons and gear.

As such an AI cannot produce endless amounts of war machines by "being AI". When it can produce war machines they can be fought by humans. They may be better at things and be similarly hard to kill as tanks but so what? There won't be that many for centuries to come.

Also, you'd be an idiot to give everything under the command of a single entity.

You can simply make a "Siri" that you ask for solution for strategic deployments. Then humans pass this information to respective strategic level assets, brigades, fleets etc. where individual AIs lead individual fighting units equipped with their own AIs that handle the movements and technical side of individual fighting.

In such a world "oh no, 4 bombers from 591st have gone rogue" is the worst case scenario but mostly quite containable.

The real problem of future lies in ever increasing amount of energy available to individual fighting units. Eventually we'll have battleships that are able to level entire continents, later entire planets.

_________________
Screw you nero


Top
 Profile E-mail  
 
 Post subject: Re: AI and Ethics: Overcoming the Risks
PostPosted: Thu Mar 29, 2018 5:56 pm 
Offline
Sergeant Major

Joined: Sat Oct 04, 2008 4:43 pm
Posts: 14538
Reputation points: 1737
Kameolontti wrote:
Humans indeed design the system that is able to learn.

After that the system itself develops skills that it is capable of developing on it's own. In fact it can be difficult for humans to grasp what is going on within the system - even the designers.
...

So far the neural networks have been used to solve various problems with success. But obviously there has been no need for self awareness, no use for that in pattern recognition or playing games. So far NN-AI lacks ability for abstraction of recognizing higher level object. But for self driving cars some kind of self awareness is necessary, not to mention battle robots.

Another thing with NN-AI is that when one learns something, it is shared with other similar NNs in no time through internet. This makes learning extremely fast.

I can see the possibilities for all good, but there are risks in self-awareness and abstract thinking, just like with humans with abstract thinking. ;)

_________________
The trouble with the world is that the stupid are cocksure and the intelligent are full of doubt

Mit der Dummheit kämpfen selbst Götter vergebens.


Top
 Profile E-mail  
 
 Post subject: Re: AI and Ethics: Overcoming the Risks
PostPosted: Fri Mar 30, 2018 3:04 pm 
Offline
His Most Gracious Majesty, Commie of the Year
User avatar

Joined: Sat Feb 28, 2009 12:44 pm
Posts: 7718
Reputation points: 17050
nero wrote:
Kameolontti wrote:
Humans indeed design the system that is able to learn.

After that the system itself develops skills that it is capable of developing on it's own. In fact it can be difficult for humans to grasp what is going on within the system - even the designers.
...

So far the neural networks have been used to solve various problems with success. But obviously there has been no need for self awareness, no use for that in pattern recognition or playing games. So far NN-AI lacks ability for abstraction of recognizing higher level object. But for self driving cars some kind of self awareness is necessary, not to mention battle robots.

Another thing with NN-AI is that when one learns something, it is shared with other similar NNs in no time through internet. This makes learning extremely fast.

I can see the possibilities for all good, but there are risks in self-awareness and abstract thinking, just like with humans with abstract thinking. ;)


Star Wars has a fabulous solution for sentient robots - no digital networking. They have to blurt - talk - to each other. This means they can't hack, recruit or otherwise convert other robots short of opening the panels physically and getting busy the old school way.

And even that could be difficult if you could not wield the necessary equipment to do that, even if you somehow managed to avoid detection and lure another robot with you in a cleaning closet.


As for self-awareness or sentience, I don't see where it's needed. You can have all the fancy pattern recognition modules feeding data to a simple choice array.

For instance battle robots only need a few modes - maintenance | safeties on | wary | combat - and with that it primarily needs to identify threats from non-threats on a multi-dimensional scale and then compare these observations with others for cross reference and for situational awareness and choosing of best play.

Nothing there to require self-awareness or sentience.

Same for cars, which are just combat robots without guns :lol:


Asimov suggested that in future the programming is not in the form of software but through the hardware - that their hardware contains their learning programming as well as the three laws of robotics. You can't go about quickly changing a hardwired rule that is designed to short circuit and destroy the CPU when the unit violates the laws of robotics. While this employed a number of people for the design of the hardwired-type brains this was a small price to pay for robots that can be guaranteed under all circumstances to obey three laws of robotics. Thus any human was always safe from any robot without exception to the level that if you came with your robots to my place and tried to hurt me *your* robots would stop you at the expense of their own 'health'.

Or at least until the Solarians made their own versions...

_________________
Screw you nero


Top
 Profile E-mail  
 
 Post subject: Re: AI and Ethics: Overcoming the Risks
PostPosted: Thu May 03, 2018 5:39 am 
Offline
buck private
User avatar

Joined: Sat Sep 20, 2008 8:01 am
Posts: 15169
Reputation points: 11492
Quote:
AI CEO Says Artificial Intelligence Can 'Defeat Bias,' 'Improve Morality,' and Craft Perfect Laws
By Tyler O'Neil May 2, 2018

https://pjmedia.com/trending/ai-ceo-say ... fect-laws/
:roll:


Top
 Profile E-mail  
 
 Post subject: Re: AI and Ethics: Overcoming the Risks
PostPosted: Thu May 03, 2018 7:28 am 
Offline
His Most Gracious Majesty, Commie of the Year
User avatar

Joined: Sat Feb 28, 2009 12:44 pm
Posts: 7718
Reputation points: 17050
abradley wrote:
Quote:
AI CEO Says Artificial Intelligence Can 'Defeat Bias,' 'Improve Morality,' and Craft Perfect Laws
By Tyler O'Neil May 2, 2018

https://pjmedia.com/trending/ai-ceo-say ... fect-laws/
:roll:


Clickbait and also trying to milk for funds from liberals.

_________________
Screw you nero


Top
 Profile E-mail  
 
Display posts from previous:  Sort by  
Post new topic Reply to topic  [ 25 posts ]  Go to page Previous  1, 2, 3

All times are UTC - 5 hours [ DST ]


Who is online

Users browsing this forum: No registered users and 6 guests


You cannot post new topics in this forum
You cannot reply to topics in this forum
You cannot edit your posts in this forum
You cannot delete your posts in this forum
You cannot post attachments in this forum

Search for:
Jump to:  
Powered by phpBB © 2000, 2002, 2005, 2007 phpBB Group