|
 |
You are here : The Arena
|
|
Welcome to our readers' forums, which we fondly call The Arena. You must be REGISTERED to post here.
|
 |  | |
| | |
| | |
| | |
 | | |
|
|
|
|
the possibility of a thinking machine is limited only by programmers. the reason we haven't been able to create artificial intelligence is simply that we don't understand how (nor really why) we think. computers cannot think because we don't know how to tell them to do so, and as it stands, a computer can only do what it is told. that's the block. that all code resolves to binary may be a limitation too. every instruction, problem, algorithim, what have you, on a computer, is tranformed into yes/no (on/off is probably more accurate). computers don't understand "maybe" or "sort of", and while strides in fuzzy logic have been made, they are made by inputting lists of parameters to match.
the question then, is less can we create a thinking machine. instead, it is can we create a thinking machine without actually understanding what thought is? or without understanding how we learn? part of the problem may also lie in the fact that computers don't have the array of different inputs we have. i watch my son look at things and touch them (try to eat them too) while he is trying to figure out what they are and how they might work. sure, he probably doesn't actually figure the inner workings of his jumper-roo out (tall order for a 6 month old), but it's the curiosity that is significant. curiosity, senses; these are key to learning. learning is integral to thought. how in fuck's name do you program that? when we figure it out, then we'll be able to create AI. |
|
|
|
| |
|
|
|
Twif,
Not fully sure I understand what you are saying. But it seems to me there is a difference between creating a model of thought on the one hand and thinking on the other.
What I have in mind is the history of computers in meteorology. My understanding is that for some time the computer was held to be the key to weather prediction because people thought that the basic problem was the enormous amount of data. Computers were seen as devices for calculation -- if you could get enough readings from around the world and plug them into a computer using known information about weather, you would be able to get accurate predictions.
This turned out not to be the case. The reason being (as I understand it) that weather tends to be a Chaotic system (reference to Gleick book). One consequence is that small differences in data can mean huge differences in outcome at a given moment, but that patterns will be recognizable in the system as a whole. There was thus a shift to computer modeling -- programs that act like a weather system -- which gives insight into the phenomenon of weather without actually predicting the weather.
Not sure if I am making sense. But I am wondering if one way of thinking about the arguments that Keif dissects is a distinction between a model of thought -- an algorithim that tells us something about how thought might work in the brain -- and a thinking machine -- that is, artificial intelligence. You could create a model of curiosity and learning that helped you understanding those processes but which nevertheless frequently erred when given a specific task.
I guess what I'm trying to say is that I'm not sure whether knowing more about thought will actually result in AI -- just as knowing more about weather models doesn't necessarily help predict weather.
I am, however, in over my head. If I understand Keif, I think I'm also agreeing with him in shared bafflement about how Godel's incompleteness helps with consciousness. Seems to me that one could make a much simpler argument -- that distinct neural systems, which orginally evolved in the context of solving particular problems, might at some point interact in a way such that the whole was greater than the sum of the parts.
|
|
|
|
| |
|
|
|
Twiff: big questions, and people much smarter than me have disagreed about them. We ask, is the brain a computer? It seems to follow rules, adjust it's state based on external inputs and its previous internal state. But there's a self in there that's hard to explain.
can we create a thinking machine without actually understanding what thought is?
Heh, I'll ask my kids. Some ideas on that in next part, sort of.
August: the thing with Godel's theorem is a real poser. If the brain is a computer, then it shouldn't generate it (says Penrose). Is it sufficient to say that brains can believe or prove things (what's "belief?" what's "proof?"--maybe ghost doesn't like the article for such imprecision, that's what bothers me) things that are false? Penrose is really bothered by taht idea. I'm lending an engineering background some of the concepts, maybe controversially.
|
|
|
|
| |
|
|
|
Well, I can't decide whether this is stupid or not. (and if not, the same as Dawkins's "meme"?). So I empathize.
http://www.edge.org/3rd_culture/christakis08/christakis08_index.html
[edit for typo] |
|
|
|
| |
|
|
|
the major difference i see is that computers, such as they are, can only do what they are told. whereas humans can overcome that particular hurdle. if you consider the base technical aspects, computers and brains are essentially the same: machines that process instructions as electrical impulses. it is the programing that seems different. a computer can only learn, such as it is, by having a human reprogram it. human learning is an automatic reprogramming. thing is, we really don't know how we do that. given that computers do only what humans tell them to do, and humans don't know how to tell them to, say, question their instructions, how then, can we artifically create thought?
as far as august's mentioning of weather models (i'd say this goes for climate models too), i'd venture to say that part of our problem in prediction is still a lack of understanding. weather predictions are fairly accurate about 3 days out. we can, based on our knowledge, be remarkably accurate in forecasting what tomorrow's weather will be. why then, are longer term predictions less accurate? chaos theory is, to my mind, a euphemism for "we still don't completely understand this". we may understand many of the rules, and the major variables. but, given the complexity of the system, we still don't know all the variables (with weather and climate the include, well, everything) and thus don't know how the rules apply, or if there are still unknown patterns. chaos is a misnomer, as it implies an absence of rules and patterns. they are not necessarily absent, we just don't understand what they are.
now, this argument hinges on a belief in an underlying structure to the universe; knowable, but excruciatingly complex. such a belief is faith; just as irrational as the religious variety. yet, if you think about it, there is value in irratitionality. consider it a protective mechanism. without some base, without axioms and beliefs and faith, we would founder and be unable to function. one can question only so much before succumbing to madness. try pondering just what infinity actually is. or true nothingness. or the origins of the universe. can you really wrap your head around it? if the, for instance ,the universe if finite, then what the hell is beyond its borders? what is nothing? we have to simply accept certain things to function. but not everything. there is something in our programming that acknowledges our limitations, while still seeking to push them. we just don't understand how it works. how then, can we tell something else to do that?
|
|
|
|
| |
|
|
|
Hey twif,
I mentioned in another forum an exhibit that you would enjoy -- Design and the Elastic Mind. Google and you'll find a cool website.
Let me say at the outset that I am a fucking idiot -- meaning, that I don't know what I am talking about, that my sources for thinking what I think are things like blog posts and novels, and that I'm writing anyway. Mostly I enjoy the conversation.
That said -- if you think about the brain (or the weather) as hardware and software, I think the analogy is likely too, um, static. My understanding is that the hardware itself changes. Also, depending on what you mean by learning, in our analogy is it best thought of as an outcome produced by the software or as a change in the software itself? (I don't know enough about the workings of the brain to know if that question even makes sense). But basically, "brain as computer" seems to work to me only by envisioning a very different kind of computer, and by the time you get through imagining that computer, it's not clear that the metaphor is helpful.
I do however, like the point that a crucial aspect of learning is understanding limits to understanding. That kind of meta-learning is plausible to me as a component of consciousness. Although there again -- you can program a computer to tell you that it can't solve a given problem. I would imagine you could program a computer to describe itself as a solver of certain types of problems, to seek those problems out, and to distinguish between what it can and cannot solve. I could envision, for example, a web search that identifies common sources of message board conflict, or a robot that seeks out bad sections of a and re-roofs them. Harder for me to imagine a computer that can come up with useful, non-obvious statements about knowledge itself.
|
|
|
|
| |
|
|
|
next post will talk about some neat hardware design, and briefly about learning hardware and so forth. I'm hung up on the ideas that all of our perceptions and intuitions and measurements are approximations of real physics anyway (including the "unassailable" axioms), and that idea plays in there. Non-ideality may be because we're necessarily running things on hardware. Not prepared to say the physical world is uncomputable. In some sense, it computes itself.
Penrose, for what it's worth, puts down machine learning as algorithmical. Adjusting the rules according to some more rules. Sounds right.
The problem with putting Godel's theorem as an axiom of your theoretical structure is that you've now got a new body of theory (that now includes that axiom), that you can apply the same proposition to. That's about the point where this sort of thinking drives you nuts.
I'm running behind--damned employers are so pushy with their demands.
|
|
|
|
| |
|
|
|
what, you think i know what i'm talking about too? i just get a bit more argumentative (and philosophical, i suppose) when i'm stoned. but i agree, the conversation is damned enjoyable.
brain as computer works insofar as both rely upon: electrical impulses fire, shit gets done. is the human brain algorithmic as well? possibly, but i don't really think so. we seem capable of non-binary thought, and our computers are still limited by that (quantum computers, i'd think, would get around this, but i don't know jack about them. and i'm not sure the people working on them quite understand how they might work either).
the knowledge of our limitations helps in learning by allowing us to use axioms with the knowledge that they are simply assumed to be true. computers do this too, only what ever we tell them is true is, as far as the program goes, true. right or wrong doesn't matter. another key to learning is observation. consider this as a test: if we program a computer to believe that 1 + 1 = 3, it will consider that to be true. if, then, via cameras or what have you, we allow it to observe that 1 + 1 = 2, would we be able to program it to override given truths based upon such observations? that would be learning, and to learn, you need to think. moreover, would be able to program it in such a way that the observation is not treated as an exception? can we create a machine that can extrapolate?
have you ever played with those linked metal puzzles (something like this)? i love them: partly because i tend to fidget and they give me something to do with my hands, partly because i'm fairly good at figuring them out. once you've discovered the trick of the puzzle, it becomes remarkably easy to do. moreover, once you've figured one out, figuring any of them out becomes easier, because despite the many variations and permutations, the design of how they work is the same. to me, this sort of behavior seems the core of learning: not only figuring out how A1 works, but realizing that A2 works the same way, despite having never directly observed A2. computers can only do part of this: they can recognize patterns they've been told about. can we teach them to figure patterns out on their own observations? |
|
|
|
| |
| | |
| | |
| | |
|
|
|
|
|
|
| Home | And Now For The News | The Arena | The Comics Page | About Quiblit
|
|
|