neural networks alone will never be able to show/simulate intelligence on their own. There needs to be something on top of them that probably works in a different way to simulate actual reasoning.
Natural neural networks can do it so in principle it should be doable. It's probably about a higher-level structure/composition of various NN modules working together. Human brain is composed of many different functional parts as well.
Maybe you should be more specific about what exactly you mean by AI. Generally, intelligence is an ability to solve a problem / accomplish a task / reach a goal. Under this general definition literally any machine created by humans is AI.
Also it's not necessary to know HOW a system works for it to be called intelligent. The important part is that it is likely to reach its goal. In fact the more intelligent the system is the more likely it is to reach its goal, and the less we will understand it. That's why this type of research is actually interesting. It attempts to explain how those things work..
Rogue One was quite good. Completely different from TFA and TLJ, and generally geared towards more adult audiences. It has a slow start but the last 1/2 hour has some of the best scenes in all SW movies, and some original music too.. https://www.youtube.com/watch?...
Otherwise I agree with you, definitely not gonna wacth episode 9 after what they did in 7 and 8...
Yep. it's most likely not an organized conspiracy to "take over and destroy Western culture" but that doesn't mean some stupid socialist ideas aren't on the rise. It might also be part of a real hybrid warfare campaign where the goal isn't to push particular ideology but to destabilize enemy states by dividing its population, lowering economic productivity, general health, etc. Research cold war era Soviet anti-NATO propaganda. It may easily have some long-lasting effects until today..
Obviously it depends on how a particular instance of AI is implemented. Your comparison to a person's development is very limited. For example unless a person is born a psychopath you can engrain some values during their mental development phase. There will remain mostly fixed for the remainder of their life. If an AI is implemented like that - that it is only suggestible during some limited period of time when it is learning - it might be ok. But the ultimate problem is that there are no general limits on how AIs can be implemented. The result will be determined by competition, an evenutal "arms race" of many parties trying to employ AI to their advantage..
IMO some form of regulation is prudent and necessary. But that wil postpone the onset of intelligence explosion only for a while. In the long term the process is still pushed forward by an intelligence "arms" race. Filtered by natural selection it will end in behaviors based on game theory. Any individual or group of individuals (such as a government) is just one of many players in the game..
And that is just it. There is no known way to implement the spark of intelligence and insight in a machine. None at all.
Are you sure? Although it might be difficult I see no conceptual problem with trying to mimic what the human brain does. Things like object recognition, abstraction, classification, deduction and higher-level reasoning. Obviously this cannot be implemented by manually editing ontologies in LISP rules. It needs to be automated.
symbolic reasoning needs to take place outside of the Neural Net being used for Object Identification
Why? We do all of this inside our neural nets. Object recognition, identification, analysis, abstraction, classification, and "symbolic" reasoning.. Sure the network needs to be more complicated, probably composed of many different functional "modules" working together to solve complex problems. But I see no reason to have to go outside..
I disagree with this reckless approach.. AI safety must be the first thing to be considered before any R&D, the "goal alignment" problem being the most important of all.
"When you have a machine that can program itself, it is no longer a machine."
What? Why? (Of course it is still a machine. People are machines too, as well as all other living organisms.)
I am threatened by SJW callout culture and people who claim I need a scientifically proven don't-rape seminar.
You're objecting to both sensitivity training seminars AND to calling out known harassers. I agree with you on the first one, you can't train someone to not be a sleazeball. But the second one?
That's completely besides the point. Why should he be required to attend these seminars in the first place? Are you saying he's a proven predator (whatever that means) who should be shamed?
Interestingly, it took the two destroyers a good half an hour to get all of these in the air, so the early units actually loitered above the target, doing laps until the rest of them could catch up, and then all were used on their targets within just a couple of minutes.
That's weird, reading they're trying to ban this practice just now. Isn't eavesdropping on private communication already illegal in the US? I'm pretty sure that would be against the law in EU countries.. not to mention FB terms and conditions.
Nice! I can imagine that many variations of this model could be simulated and tested virtually much faster (3D scene with ray-tracing). Evolutionary algorithms could help find the most efficient constructions which would be then tested in reality.. What a thesis that would be..:)
Some facilities are more important than others. This is a very specific list of infrastructure and resources which the USG considers critical for ensuring public health and economic and military security. There is a clearly stated goal to protect these facilities and prevent their destruction, particularly by terrorist threats.
It would be naive to think that US enemies can't identify important targets but why make their work so much easier? Why hand out this information to them on a silver plate? I don't it's the public who will benefit from this particular publication most and I'm really starting to doubt the real motivations and goals behind this project...
Until my country's copyright law will be amended, I am still entitled to...
The thing is ACTA is an international agreement and if approved at EU-level it would be obligatory for the member countries to adapt their laws accordingly. In case of EU, this so called "trade agreement" is basically a cheap way to circumvent legitimate democratic proceses of individual countries.
Slashdot Mirror
neural networks alone will never be able to show/simulate intelligence on their own. There needs to be something on top of them that probably works in a different way to simulate actual reasoning.
Natural neural networks can do it so in principle it should be doable. It's probably about a higher-level structure/composition of various NN modules working together. Human brain is composed of many different functional parts as well.
Maybe you should be more specific about what exactly you mean by AI. Generally, intelligence is an ability to solve a problem / accomplish a task / reach a goal. Under this general definition literally any machine created by humans is AI.
Also it's not necessary to know HOW a system works for it to be called intelligent. The important part is that it is likely to reach its goal. In fact the more intelligent the system is the more likely it is to reach its goal, and the less we will understand it. That's why this type of research is actually interesting. It attempts to explain how those things work..
That's a terrible idea. Who wants to live next to a production facility?
Rogue One was quite good. Completely different from TFA and TLJ, and generally geared towards more adult audiences. It has a slow start but the last 1/2 hour has some of the best scenes in all SW movies, and some original music too.. https://www.youtube.com/watch?... Otherwise I agree with you, definitely not gonna wacth episode 9 after what they did in 7 and 8...
Yep. it's most likely not an organized conspiracy to "take over and destroy Western culture" but that doesn't mean some stupid socialist ideas aren't on the rise. It might also be part of a real hybrid warfare campaign where the goal isn't to push particular ideology but to destabilize enemy states by dividing its population, lowering economic productivity, general health, etc. Research cold war era Soviet anti-NATO propaganda. It may easily have some long-lasting effects until today..
That's my problem with "liberalism". At the core it really is hypocritical.
I don't understand why Americans so often conflate socialism with liberalism..
Obviously it depends on how a particular instance of AI is implemented. Your comparison to a person's development is very limited. For example unless a person is born a psychopath you can engrain some values during their mental development phase. There will remain mostly fixed for the remainder of their life. If an AI is implemented like that - that it is only suggestible during some limited period of time when it is learning - it might be ok. But the ultimate problem is that there are no general limits on how AIs can be implemented. The result will be determined by competition, an evenutal "arms race" of many parties trying to employ AI to their advantage..
So who defines these norms?
Natural selection. Intelligence arms race. Game theory.
I think there already are technolgies that can do all above-mentioned tasks to some degree. The trick is to wire them together in a way that works..
IMO some form of regulation is prudent and necessary. But that wil postpone the onset of intelligence explosion only for a while. In the long term the process is still pushed forward by an intelligence "arms" race. Filtered by natural selection it will end in behaviors based on game theory. Any individual or group of individuals (such as a government) is just one of many players in the game..
And that is just it. There is no known way to implement the spark of intelligence and insight in a machine. None at all.
Are you sure? Although it might be difficult I see no conceptual problem with trying to mimic what the human brain does. Things like object recognition, abstraction, classification, deduction and higher-level reasoning. Obviously this cannot be implemented by manually editing ontologies in LISP rules. It needs to be automated.
We already have it. It's called automation. There are many problem domains where AI is already orders of magnitude better than human being.
symbolic reasoning needs to take place outside of the Neural Net being used for Object Identification
Why? We do all of this inside our neural nets. Object recognition, identification, analysis, abstraction, classification, and "symbolic" reasoning.. Sure the network needs to be more complicated, probably composed of many different functional "modules" working together to solve complex problems. But I see no reason to have to go outside..
I disagree with this reckless approach.. AI safety must be the first thing to be considered before any R&D, the "goal alignment" problem being the most important of all.
"When you have a machine that can program itself, it is no longer a machine." What? Why? (Of course it is still a machine. People are machines too, as well as all other living organisms.)
Intelligence is an ability to achieve a goal / solve a problem -- in this case steering a car. Intelligence IS the control mechanism.
I thought laser weapons were among the forbidden weapon categories. Is China not a signatory to the Geneva Protocols?
I am threatened by SJW callout culture and people who claim I need a scientifically proven don't-rape seminar.
You're objecting to both sensitivity training seminars AND to calling out known harassers. I agree with you on the first one, you can't train someone to not be a sleazeball. But the second one?
That's completely besides the point. Why should he be required to attend these seminars in the first place? Are you saying he's a proven predator (whatever that means) who should be shamed?
You may fire when ready.
Interestingly, it took the two destroyers a good half an hour to get all of these in the air, so the early units actually loitered above the target, doing laps until the rest of them could catch up, and then all were used on their targets within just a couple of minutes.
http://www.latimes.com/world/m...
Looking at the video they fire 1 missile per 15 s which means the whole salvo from 2 ships would take about 7.5 minutes.
That's weird, reading they're trying to ban this practice just now. Isn't eavesdropping on private communication already illegal in the US? I'm pretty sure that would be against the law in EU countries.. not to mention FB terms and conditions.
Nice! I can imagine that many variations of this model could be simulated and tested virtually much faster (3D scene with ray-tracing). Evolutionary algorithms could help find the most efficient constructions which would be then tested in reality.. What a thesis that would be.. :)
Some facilities are more important than others. This is a very specific list of infrastructure and resources which the USG considers critical for ensuring public health and economic and military security. There is a clearly stated goal to protect these facilities and prevent their destruction, particularly by terrorist threats. It would be naive to think that US enemies can't identify important targets but why make their work so much easier? Why hand out this information to them on a silver plate? I don't it's the public who will benefit from this particular publication most and I'm really starting to doubt the real motivations and goals behind this project...
Until my country's copyright law will be amended, I am still entitled to...
The thing is ACTA is an international agreement and if approved at EU-level it would be obligatory for the member countries to adapt their laws accordingly. In case of EU, this so called "trade agreement" is basically a cheap way to circumvent legitimate democratic proceses of individual countries.