Quote:Original post by random_thinker I don't think that 'intelligence' can be achieved by a computer unless we have the following conditions:
a) Code that can produce its own, and more complex, code. b) A group of competing intelligent entities. c) A value system to encourage development. d) Mutation and selection of the most fit entity e) Collective learning of all entities.
--random
Basically, you want a genetic programming engine. We have those, though I can't say they've been applied to general computation (IME they're always targetted at a very specific problem, because it makes them much, much faster).
You wan't something that can adapt quickly and precisely... Like a neural network. The problem with current neural networks is that all those we see are feed-forward networks. People also have to realise that the brain has some inherent programmation coming from its design (its not just a big bunch of neurons, they have a very precise organization). They wouldn't be what actually does the computations, however.
Where genetic algorithms could be used, in my opinion, is to evolve an efficient neural network layout (ie: evolve an artificial "brain").
Quote:Original post by random_thinker I don't think that 'intelligence' can be achieved by a computer unless we have the following conditions:
a) Code that can produce its own, and more complex, code. b) A group of competing intelligent entities. c) A value system to encourage development. d) Mutation and selection of the most fit entity e) Collective learning of all entities.
--random
Basically, you want a genetic programming engine. We have those, though I can't say they've been applied to general computation (IME they're always targetted at a very specific problem, because it makes them much, much faster).
You wan't something that can adapt quickly and precisely... Like a neural network. The problem with current neural networks is that all those we see are feed-forward networks. People also have to realise that the brain has some inherent programmation coming from its design (its not just a big bunch of neurons, they have a very precise organization). They wouldn't be what actually does the computations, however.
Where genetic algorithms could be used, in my opinion, is to evolve an efficient neural network layout (ie: evolve an artificial "brain").
I am more inclined to think that computing at a quantum level is much more likely to achieve AI than the von neumann machines. You can certainly build very intelligent applications on PCs though, to the point it is illusive enough it is intelligent, and that is my goal to build I(ntelligent) A(pps) and leave AI to someone else ;-)
I really don't consider I care if my PC is sentient, what I care is it behaves intelligently and does the right thing. In a game environemnt, if it presents the illusion it is intelligent and it is believable, it is good enough for me.
Quote:Original post by random_thinker I don't think that 'intelligence' can be achieved by a computer unless we have the following conditions:
a) Code that can produce its own, and more complex, code. b) A group of competing intelligent entities. c) A value system to encourage development. d) Mutation and selection of the most fit entity e) Collective learning of all entities.
--random
Basically, you want a genetic programming engine. We have those, though I can't say they've been applied to general computation (IME they're always targetted at a very specific problem, because it makes them much, much faster).
You wan't something that can adapt quickly and precisely... Like a neural network. The problem with current neural networks is that all those we see are feed-forward networks. People also have to realise that the brain has some inherent programmation coming from its design (its not just a big bunch of neurons, they have a very precise organization). They wouldn't be what actually does the computations, however.
Where genetic algorithms could be used, in my opinion, is to evolve an efficient neural network layout (ie: evolve an artificial "brain").
I am more inclined to think that computing at a quantum level is much more likely to achieve AI than the von neumann machines. You can certainly build very intelligent applications on PCs though, to the point it is illusive enough it is intelligent, and that is my goal to build I(ntelligent) A(pps) and leave AI to someone else ;-)
I really don't consider I care if my PC is sentient, what I care is it behaves intelligently and does the right thing. In a game environemnt, if it presents the illusion it is intelligent and it is believable, it is good enough for me.
Quantum computing is interesting, but AFAIK, its not possible to create a turing complete computer using only quantum technology, which might very well make it impossible to create AI that way. I've discussed with someone who is very involved in quantum computing research at my university, and trust me, its a little overhyped.
I've read up on quantum computing a very little bit, and what I don't understand is why it can't be easily simulated on regular computers. In QC, ir seems that the idea is that there are X states which start out qith equal probability, and the algorithm adjusts the probabilities so 'right'/'good' answers are more likely to be the result after you 'collapse the state' to a single answer. Sounds like it could be easily approximated using a simple array of probabilities.
Sure, it would be slower, but something is surely better than nothing, and I don't think the simulator would cost quite as much as the real thing right now =-)
"Walk not the trodden path, for it has borne it's burden." -John, Flying Monk
Quote:Original post by Extrarius I've read up on quantum computing a very little bit, and what I don't understand is why it can't be easily simulated on regular computers. In QC, ir seems that the idea is that there are X states which start out qith equal probability, and the algorithm adjusts the probabilities so 'right'/'good' answers are more likely to be the result after you 'collapse the state' to a single answer. Sounds like it could be easily approximated using a simple array of probabilities.
Sure, it would be slower, but something is surely better than nothing, and I don't think the simulator would cost quite as much as the real thing right now =-)
The whole point of quantum computation is to use quantum tricks to obtain better running time on a few algorithms. Certainly, you can simulate quantum computers, but then the speed advantage is nullified (you even lose in speed), and so there is no point.
I sometimes think that the AI community seeks to find a deterministic soution to a non-deterministic problem. Human intelligence is absolutely non-deterministic.
Indeed genetic programming must be a part of the solution because the ability to mutate randomly is crucial to 'true' intellectual and physiological development. In my view, this abilty to mutate randomly must be at the heart of creativity. However, learning that is driven by a communal value system must be used to guide the species in a direction of advancement.
Until we 'program' a community of competing entities that can compete within the framework of a value system then true 'intelligence' will not be achieved by the thinking machine.
At the heart of this is a programming language or code that permits the thinking machine to write new code and develop. To date this has been a fundamental stumbling block to machine 'intelligence'.
I would be interested if anyone has any experience on this specific point.
--random
[Edited by - random_thinker on August 28, 2005 12:51:34 PM]
--random_thinkerAs Albert Einstein said: 'Imagination is more important than knowledge'. Of course, he also said: 'If I had only known, I would have been a locksmith'.
This randomness is at the heart of intelligent development; anyone who has studied stochastic processes can testify that there is an infinite number of paths that can be taken from any starting point.
For example, if we consider Darwin's Theories are valid, then our planet has followed a particular stochastic path that has led to the diversity of species that we observe here today.
However, what this implies is that on any other planet in the Universe with the same single-celled starting point as ours had billions of years ago, there is almost complete statistical certainty that the diversity and type of species will be entirely different, and utterly unrecognizable, from that of our planet earth. This would also apply to the development of intelligence on that planet.
Human intellectual development and therefore the development of machine intelligence must follow the same process, in my view.
--random
[Edited by - random_thinker on August 28, 2005 12:00:05 PM]
--random_thinkerAs Albert Einstein said: 'Imagination is more important than knowledge'. Of course, he also said: 'If I had only known, I would have been a locksmith'.
The problem with using neural networks is that you must have a large body of data with which to train them. There have been studies where genetic programming algorithms have been used to accelerate the solution of neural network problems (see 'Genetic Algoritms in C++' by Ladd) but the same fundamental problem of initial datasets exist.
Creativity and mutation however, are responsible for the advancement of human intelligence. To date I have not successfully found a method for simulating this feature. However, I think that the answer to this is not distant.
--random
Neural network approaches can be used to develop 'knowledge' from a given dataset, but as Albert Einstein said: 'Imagination is more important than knowledge'. Of course, he also said: 'If I had only known, I would have been a locksmith'.
[Edited by - random_thinker on August 28, 2005 4:16:20 AM]
--random_thinkerAs Albert Einstein said: 'Imagination is more important than knowledge'. Of course, he also said: 'If I had only known, I would have been a locksmith'.
Regarding creativity, I believe that the process must be very similar to this: The somewhat chaotic nature of neurons in our brains, a product of the Whatever Works (tm) phylosophy enforced by natural evolution, is constantly producing erratic (random) patterns in our neural networks that have never showed up before. Most of these are automatically shunned away, which must be mostly an unconscious process, but some of them pass through these layers of anti-gibberish protection to produce a "hey! that's actually a good idea" moment.
Extremely creative people must have their brain circuits modeled in a way that encourages this process. It is however understandable, from an evolutionary point of view, that this process doesn't happen more often -- just a little bit of chaos is needed, too much and it would interfere with the normal functioning of the brain.
