This is an archive of a previous version of Sodarace.

forum  |  »» sodaplay

»» forums  »» sodarace league forum subject: AmeobaRace 271 replies on 19 pages. most recent reply: 13-Aug-07 22:55 by edlinfan

»» back to topic list

This topic has 271 replies on 19 pages [ « | 1 2 3 4 5 6 7 8 ... 19  | » ]

»» previous topic   »» next topic

Veret »» models Re: AmeobaRace   posted: 28-Nov-03 13:41   »»  Yeah solomon you are really god at this, but i'm not sure anyone can really compete in the AI league, since if you give it enough time they'll just come out with a model that is as good as that model type can be, and there would be no point in trying to beat it. Anyway, chirag, i have been proven wrong about your methods, but from what i can tell you still can't make an algorithm that will create a whole new model. This is the field in which computers are lamentably far behind humans, and i think it will stay that way for a while. Oh, and climbing stairs isn't that hard, you just need to be a little creative. This particular model is the product of about 10 minutes worth of fiddling with someone else's motor, so i'm sure you could optimize it better than me, chirag. The point, however, is that you would have to start with what a human already created in order for your algorithm to do its thing. race: http://sodarace.net/upload/Veret/Sodarace_StairZ.jnlp

the_duck »» models Re: AmeobaRace   posted: 28-Nov-03 14:05   »»  Well, Veret, if you don't want to race AI models, just ignore them. You don't need to rant on and on about how humans are better that computers.

Chirag »» models »» homepage Re: AmeobaRace   posted: 28-Nov-03 14:29   »»  im sure i mentioned this before but if u look at the models in waans thread http://sodarace.net/forum/thread.jsp?forum=16&thread=1637 he actually makes whole new models from scratch without starting from a human made model. Thats why they look so weird. The computer doesnt know the difference between a beautiful model and an ugly one. The only thing that matters to it are how fast it goes. We'll be writing our own algorithms that create models from scratch soon.

Veret »» models Re: AmeobaRace   posted: 28-Nov-03 15:11   »»  I don't think you understood what i said, duck. Chirag and his colleagues are making an algorithm to create the fastest possibl model, and i'm just stating the facts. I think AI modeling is a great thing, and i'm certainly not ranting about it. Anyway, chirag, you have proven me wrong yet again, but those models are still not fast enough. Incidentally, this model that wins is the very first one I built. Also, I'm curious whether this program can make any models more complicated than just a bunch of (very fast) vibrating sticks. Anyway, i eat them all: race: http://sodarace.net/upload/Veret/Sodarace_humanbeatsall.jnlp

Veret »» models Re: AmeobaRace   posted: 28-Nov-03 15:12   »»  sorry...i meant to say i BEAT them all :)

Chirag »» models »» homepage Re: AmeobaRace   posted: 29-Nov-03 01:04   »»  clearly then the algorithm must be improved. we're just starting out. warming up as it were. we're relatively new to this whole sodarace thing and have alot to learn. being able to create a model form scratch is challenging for an AI. though u seem to think that AI has to be able to make something from scratch for it to valid. This is not the case. All we're doing is using AI to "intelligently" search for a solution to a problem within a given search space. Wether that space be the combination of parameters that make up an Amoeba or the parameters that make any type of model. We couldve used a systematic brute force method which goes through every combination of parameters, which would eventually find the optimum , but that would take far too long, it takes an intelligent algorithm to find the solution more quickly. I mean, when you make your amoeba by hand, you obviously tweak the settings to make it faster. When u change a parameter u dont really know wether thats gonna have a positive or negative effect. You just do it and hope it does. trial and error. Would you consider that an intelligent way of thinking on your part? my alogrithm does the same thing only better in this case. Its very difficult for anything man or machine to instantly know the solution to a problem. Human constructors may look at a terrain and know what kinda model would best suit it. and thats not through any great intelligence. its because they draw on the past experiences they 've had making models and use the knowledge they've learned to approximate a solution and even then it'd take alot of trial and error to come up with a good solution. If you really want something to think about. next time you solve a problem. any problem. think about the thought processes that are going through your head. how do humans solve problems? through trial and error and past experience. but the knowledge you've gained from past experiences is mostly from trial and error itself. you know that expression ? "you learn from your mistakes". If you look at it that way , you'll see that the human mind is not so great. Its just good at learning things and has had years by the time it gets to your age to do so. Once you give a machine a similar ability to learn things the way humans can i think in my opinion humans will be out matched. race: http://sodarace.net/upload/Chirag/hmm.jnlp

KingSolomon »» models Re: AmeobaRace   posted: 29-Nov-03 01:12   »»  Computers may be be faster at number crunching and mathematics, but the human mind will always have a creative aspect that computers lack. How's the site going?

Chirag »» models »» homepage Re: AmeobaRace   posted: 29-Nov-03 01:57   »»  i disagree. human creativity is overrated , in want of a better word. We can look at our own minds and think how wonderful they are. How they're able to solve so many problems. What u dont actually appreciate is how it got to be in that state. All the years of learning it took from the day you were born, and you're not even learning that from scratch. You learn from teachers and books. Whats the source of all human knowledge? science. and how do thing develop in science? trial and error. Just like when most people look at humans in general and think to themselves how wonderfully complex they are. they dont appreciate the millions of years of evolution it took to get to that state, and what is evolution ? trial and error at the basis of it. What im trying to say is that when it comes to problem solving humans aren't creative at all. They use a combination of previous knowledge and trial and error to solve problems. This produces new knowledge that can be used to solve future problems and the process goes on. The advantage you have is that you can use this forum to learn from each other. I may not be right in saying this but i bet most of the great discoveries of soda constructing were not made by creativity, but by accident . The ppl who made them tho still recognised that they had discovered something great and passed this knowledge on to other vias this very forum. My final point is, you can make a computer do this , you can tell it how to do trial and error, and u can tell it to remember the good models and 4get the bad... therefore learning in the same way humans do. PS sorted out the web hosting... give us a few days..

KingSolomon »» models Re: AmeobaRace   posted: 29-Nov-03 02:09   »»  You're basing your whole argument on the ability of the human mind to solve problems. The amazing creativity in the human mind is manifested in how we invent and imagine. Also, learning from teachers and books is articificial learning. We don't learn anything real from them, just useless drills to go over and over with our minds. This kind of learning limits our ability to image and experience.

Chirag »» models »» homepage Re: AmeobaRace   posted: 29-Nov-03 02:42   »»  I used that as an example to prove a point. Even great art is created through trial and error. I know it sounds cold but making great art is simply solving the problem of making something that is pleasing to the human eye. Think about it KingSolomon for a long while before you reply. How do we invent? How do we imagine? How do we decide what are good thoughts and bad thoughts? You're not looking at it from a close enough perspective. Its difficult for a person to accept the real limitations of the human mind. Most ppl reject these thoughts with pure emotional reflex. Probably because it involves admitting we're not as special as we think we are.

KingSolomon »» models Re: AmeobaRace   posted: 29-Nov-03 02:52   »»  Any great artist does not come at his easel as a problem to solve. That kind of art is shallow, bubble-gum art. Real art, inventive art, imaginative art is not a problem being solved, but a creation.

Chirag »» models »» homepage Re: AmeobaRace   posted: 29-Nov-03 09:58   »»  im just sayin that creativity is not something so complex that it cant be simulated by a computer. tho we at the moment lack the computing power to simulate true human creativity. There have already been successful experiments where a computer was taught the difference between good art and bad art and it was able to reproduce the good art as such. Everything you do in life can be boiled down to solving a problem , or if you dont like those words achieving a goal. You take certain thoughtful steps to achieve those goals. The end product maybe some amazing piece of art but the thought processes behind them can be simplyfied to a certain number of emulatable steps. Take a human being for example. Was it the product of some great creativity? No it was the product of millions of years of cold harsh survival of the fittest. If you believe that humans are creative things, then you have to believe that its possible for a computer to simulate the same thing since the process that created humans (evolution) is made of a number of very simple , reproducable , uncreative number of steps. Just as in the same way the human mind has evolved to be creative and thoughful ,a computer brain will eventually be able to do the same thing. The methods we are using simulate evolution. Genetic Algorithms.

Veret »» models Re: AmeobaRace   posted: 29-Nov-03 14:21   »»  great food for thought, chirag. its probably true that someday computers will be able to do just about everthing as well as humans, but certainly not yet. your theory leaves out a few important points, however. while the computers can create a form of accelerated evolution with their models, the computers themselves aren't evolving as quickly. as you pointed out, humans took several million years of trial and error to reach this point, and i see no reason why computers should be any different. if we humans weren't pouring everything we knew into computers, would they eveer get anywhere? as yet, no computer has ever created another computer on its own, let alone create a human. On the other hand, we are able to create computers (which by your views are almost another form of life themselves) at the drop of a hat. my point is, are computers really more capable than we are?

Chirag »» models »» homepage Re: AmeobaRace   posted: 29-Nov-03 21:40   »»  Of course not. Not yet anyway. Probably not for a long time, But i think there is a good chance that they one day will. This would be not by simply giving them vast amounts of knowledge but by giving them the same ability to learn that we have. The disturbing thing is that machines have the potential to develop much faster than us. We evolve very slowly, machines on the other hand can be redesigned and improved at a great rate. Once we get computer designing computers who knows what they'll come up with. We already have AI's writing computer programs by themself. Its a field called genetic programming. Getting back to the point of sodarace, i think the computers today are powerful enough to develop solutions to the relatively simple problems of gettting across a terrain, and once we've done it. Thats it. We wont have to do anything anymore. The AI should be able to learn to deal with any type of terrain or situation as long as its possible and a solution does exist. Getting an AI to be able to learn about the world is just a question of time, complexity, and scaling up.

Veret »» models Re: AmeobaRace   posted: 29-Nov-03 22:49   »»  good point, chirag, but i think kingsolomon touched on something interesting too. computers are certainly better than us at some things, but they still lack creativity. i think that that is something computers may never get. anyway, if your algorithm is looking for some work, why don't you get it to DESIGN (not modify) a model for this race: race: http://sodarace.net/upload/Veret/Sodarace_StairZ.jnlp

This topic has 271 replies on 19 pages [ « | 1 2 3 4 5 6 7 8 ... 19  | » ]

»» previous topic   »» next topic   »» back to topic list  »» top of the page

PLEASE READ THE FORUM GUIDELINES AND ALWAYS PREVIEW TO CHECK MESSAGES BEFORE POSTING...
...Help keep our forums creative and constructive. Thank you.

»» forum home