Chinese Room Experiment
Imagine a windowless room with a clerk inside. The clerk only speaks English. The job of the clerk is to provide answers to visitors, who only speak Chinese. How can this be possible? Well, our clerk has a big book of rules, something like a very complex phrase book, which describes how one can manipulate given strings of Chinese characters to construct valid responses. Visitors pass written notes with their questions to the clerk. Our clerk does not understand any of the questions, but he can look each one up in the book, and copy the answers, and pass them back to visitors. To the visitors it appears that someone in the room knows Chinese, because they always get correct answers.
John Searle, who proposed this thought experiment, concluded that intelligent behavior might be observed where there's no real intelligence ("intelligence" = "understanding"?). Therefore, it can be argued that some computers which display intelligent behavior (for example, passing the Turing Test), possess no more intelligence than is required to look up the answers in the phrase book.
Perhaps I'm missing some key detail, but let me analyze this argument.
Let's say I'm given a written note which says: "1+1=?" A mental process required to answer this question is no more complicated than a table lookup. If I have a large enough brain, and a perfect recall, I can memorize answers to an unlimited number of questions given enough time. I can even do it for questions in Chinese - I don't think the process would be any different - it's still just a table lookup. So, when I answer "2" when asked "1+1=?", do I have the understanding of the question? Am I intelligent, or am I simply doing a mechanical table lookup?
You might say, well, this was a really simple question, which didn't require much "understanding". OK, then when the question becomes complicated enough for an "understanding" to become necessary? Let's take "What is love?" That's surely a harder question, but is there really more going on in my brain than triggering more pattern recognizers? Can't this be modeled as a table lookup which finds a pointer to another table, or perhaps multiple tables, then choosing the strongest association?
I could be given a note asking me to read some text, and then answer a question based on "understanding" of that text. But does "acquiring understanding" involve anything fundamentally different than building connections between pattern recognizers on multiple levels, and then choosing the words and expressions most strongly associated with the triggered patterns? Of course, the process must be much more complicated than that, but is there a reason to believe it's not just a series of biophysical events which could be observed, recorded, recreated, and automated?
These pattern recognizers, do they have to be physically located inside my brain? Why? Why can't we construct them with pen and paper, and store them in a book? I don't see why this is impossible in principle. Then we can describe how to construct them, in English, so our clerk can follow the instructions to answer questions in Chinese.
I believe that when the clerk is manipulating Chinese symbols to construct an answer, he is doing real "understanding" - only this understanding is happening outside his brain.
John Searle, who proposed this thought experiment, concluded that intelligent behavior might be observed where there's no real intelligence ("intelligence" = "understanding"?). Therefore, it can be argued that some computers which display intelligent behavior (for example, passing the Turing Test), possess no more intelligence than is required to look up the answers in the phrase book.
Perhaps I'm missing some key detail, but let me analyze this argument.
Let's say I'm given a written note which says: "1+1=?" A mental process required to answer this question is no more complicated than a table lookup. If I have a large enough brain, and a perfect recall, I can memorize answers to an unlimited number of questions given enough time. I can even do it for questions in Chinese - I don't think the process would be any different - it's still just a table lookup. So, when I answer "2" when asked "1+1=?", do I have the understanding of the question? Am I intelligent, or am I simply doing a mechanical table lookup?
You might say, well, this was a really simple question, which didn't require much "understanding". OK, then when the question becomes complicated enough for an "understanding" to become necessary? Let's take "What is love?" That's surely a harder question, but is there really more going on in my brain than triggering more pattern recognizers? Can't this be modeled as a table lookup which finds a pointer to another table, or perhaps multiple tables, then choosing the strongest association?
I could be given a note asking me to read some text, and then answer a question based on "understanding" of that text. But does "acquiring understanding" involve anything fundamentally different than building connections between pattern recognizers on multiple levels, and then choosing the words and expressions most strongly associated with the triggered patterns? Of course, the process must be much more complicated than that, but is there a reason to believe it's not just a series of biophysical events which could be observed, recorded, recreated, and automated?
These pattern recognizers, do they have to be physically located inside my brain? Why? Why can't we construct them with pen and paper, and store them in a book? I don't see why this is impossible in principle. Then we can describe how to construct them, in English, so our clerk can follow the instructions to answer questions in Chinese.
I believe that when the clerk is manipulating Chinese symbols to construct an answer, he is doing real "understanding" - only this understanding is happening outside his brain.