The Searle discussion was a tuff read.

It does not seem to me that Searle has proven that AI is impossible. I do think that he has presented a convincing argument that AI is not possible with computers using current architecture.

I think that AI is possible, but that it requires neural networks, or arrays of thousands or millions of CPU’s with some yet to be designed connectivity, or some other architecture unlike current systems.

It seems to me that Searle has not provided a formal proof, and I am not sure he was trying to develop a formal proof. Since the URL led to Larry Hauser’s article criticizing Searle, it is hard to figure out exactly what Searle said. Intent is tuff to determine under any circumstances. From what I read about the Chinese Room, I do not think it was an attempt at a formal proof, although the Hauser article indicates Searle formatted his argument like a formal proof, and used the Chinese Room in his argument.

It is fairly obvious that Hauser is attacking the Searle argument by showing that it is not a formal proof, which might not be fair if Searle never claimed to be presenting a formal proof. Hauser seems to me to be correct in claiming that the proof by Searle is invalid logic. However, Hauser does not seem to describe or refute the ideas put forth by the Chinese Room experiment.

For those not familiar with The Turing Test and the Chinese Room thought experiments, I will describe them.

Turing came up with the idea of testing future AI devices versus a human being. His idea is to put one or more people in a room with a keyboard and two monitors (actually in those days two crude typewriter like devices were proposed). Questions or remarks would be entered via the keyboard, and a pair of answers or pertinent remarks would be provided via the two monitors (teletype machines). One monitor would be controlled by a human and the other would be controlled by the AI device.

The people in control of the keyboard could ask questions or merely carry on a conversation with the entities in control of the monitors. If the people in control of the keyboard could easily tell which monitor was controlled by the AI device and which by the human, then the AI device failed the test. If the people could not figure out which was which, the AI device passed the test 100% and had to be considered intelligent. Obviously, it would be possible for the AI device to be given some score between 0 & 100%, depending on how difficult it was to identify which was which.

A full Turing Test would have no restrictions on the subject matter, and nobody expects an AI device to pass such a test with a high grade in the next 50-100 years. It seems reasonable to expect some AI device to pass a less rigorous test with restrictions on the subject matter or which otherwise gave the AI device a little help.

The Chinese Room thought experiment is an argument which claims that the Turing Test is not a valid way to assess an AI device. Imagine that you can pass Chinese ideograph text into a slot in the door of a room. Some time later, an English translation is provided via a second slot in the door. If a person fluent in Chinese and English verified that the translations were correct, it would seem reasonable to conclude that a person or an entity in the room understood both Chinese and English.

Now suppose that you investigated the contents of the room and discovered a group of people running computers. One group was looking ideographs up in a computerized database of ideograph images and creating a literal translation of original messages. Another group was processing the literal translations versus a huge subject indexed data base of Chinese ideograph literature with English translations. This group would use contexts found in Chinese literature to refine the literal translation. All the notes made by these two groups were passed to an output group which typed up the final translation.

After understanding the processes inside the room, most people would conclude that there were people in the room who understood English, but that nobody there understood Chinese. Even if the operations were carried out by one person, most people would not conclude that the person understood Chinese.

Note that the Chinese Room is similar in some ways to a Turing Test for comprehension of the Chinese language. The argument is that if you understood how a proposed AI device functioned, you might not credit it with intelligence. This is why Deep Blue is not considered an AI device even though it beat the best human chess player and chess playing is generally considered to require intelligence.

It seems that Searle carried this Chinese Room argument a bit further and claimed that AI was not possible.