One of Searle's replies to this argument is to do away with the rules and the room, and have the man memorise all the rules. Then, the entire system comprises of just the man, who can demonstrate an ability to process Chinese, even though he has no understanding of the language. Is there still a "system" present that understands Chinese?
Let's turn this other way round. Suppose, that you can speak/understand Chinese and you are talking to a person that really understands Chinese. How do you know that he/she has real understanding of the language, and is not just imitating it through some sort of memorised rule trickery?
This is not new. People already mentioned Genetic Algorithms & Genetic Programming here. Some even suggest this is the only way to do (future) engineering of very complex systems (e.g. Hugo de Garis, see his site, namely this article on "Evolvable Hardware". You may want also to look at his PhD thesis.
> very interesting > Especially tried to cheat by standing on it's wingtips or similar
This a perfect illustration to the fundamental problem with evolutionary aproach to engineering. The evolutionary hype mostly goes along the lines of "you should only define what do you want (define fittness function) and the (nearly optimal) solution would be automagically evolved into existence". That's OK, but the very problem is in exactly defining "what do you want". The very term "cheating" suggests that researchers got something that satisfied their fittness function but was not "what they wanted". This means that their fittness definition was not correct (too simple?). Of course, we humans tend to blame somebody else: "it cheated". Perhaps, it is entertaining and fun to see such "cheating" in balsa-winged toy robot, but it may get far less fun if, say, in some (probably distant) future evolution-engineered robo-surgeon suddently decides to "cheat" a bit.
Slashdot Mirror
One of Searle's replies to this argument is to do away with the rules and the room, and have the man memorise all the rules. Then, the entire system comprises of just the man, who can demonstrate an ability to process Chinese, even though he has no understanding of the language. Is there still a "system" present that understands Chinese?
Let's turn this other way round. Suppose, that you can speak/understand Chinese and you are talking to a person that really understands Chinese. How do you know that he/she has real understanding of the language, and is not just imitating it through some sort of memorised rule trickery?
This is not new. People already mentioned Genetic Algorithms & Genetic Programming here. Some even suggest this is the only way to do (future) engineering of very complex systems (e.g. Hugo de Garis, see his site, namely this article on "Evolvable Hardware". You may want also to look at his PhD thesis.
> very interesting
> Especially tried to cheat by standing on it's wingtips or similar
This a perfect illustration to the fundamental problem with evolutionary aproach to engineering. The evolutionary hype mostly goes along the lines of "you should only define what do you want (define fittness function) and the (nearly optimal) solution would be automagically evolved into existence". That's OK, but the very problem is in exactly defining "what do you want". The very term "cheating" suggests that researchers got something that satisfied their fittness function but was not "what they wanted". This means that their fittness definition was not correct (too simple?). Of course, we humans tend to blame somebody else: "it cheated". Perhaps, it is entertaining and fun to see such "cheating" in balsa-winged toy robot, but it may get far less fun if, say, in some (probably distant) future evolution-engineered robo-surgeon suddently decides to "cheat" a bit.