The Turing Century blog has started a debate about the Turing Test for machine intelligence vs. John Searle's Chinese Room thought experiment. Searle claims that his elegant thought experiment shows that computers can never think no matter how smart they appear as they are just manipulating symbols, which after all is all that a Turing machine does. My take on this is as follow, but first I'll explain the Chinese Room in case you're not familiar with it. (Note: this is an exert from the final chapter of my book The Universal Machine)
Imagine you are sitting in front of room with a locked door. You are a native Chinese speaker and you’re able to ask questions of the room by writing them down in Chinese characters on pieces of paper, which you post through a slot in the door. The room replies in Chinese by sending its written replies back through the slot. You spend time asking the Chinese Room factual questions and then progress to asking its opinions on the economy, politics the arts and you even ask it to tell you some jokes. All of its answers are perfect, just what you’d expect from an intelligent Chinese person. It has passed the Turing Test with flying colors – therefore it must be intelligent.
Then Searle unlocks the door and shows you inside the Chinese Room. Inside waiting to receive questions is an Englishman who doesn’t speak or read Chinese. He takes each received question and identifies the Chinese characters from a big ledger and writes down their corresponding numbers. He then, in a complex and laborious process, cross-references the numbers with other ledgers and indices and eventually obtains more numbers from which he obtains new Chinese characters that make up his answer. He writes these characters down and posts them back through the door.
Does the operator understand Chinese? The answer is obviously no. He’s just following a laborious mechanical process, a program. In fact he needn’t even know the characters make up a language called Chinese. Does the operator understand the questions? Again the answer must be no; he doesn’t even know they are questions because he doesn’t understand Chinese. Therefore the Chinese Room is not intelligent and never can be. It is, as Jefferson [a professor who clashed with Turing over machine intelligence] observed, just moving symbols around; there is no understanding.
AI experts, including myself, have struggled with Searle’s simple thought experiment. It does seem to show that a universal machine by manipulating symbols can never be said to think. My take on this problem is to consider the example of flight. Do birds fly? Of course they do, not all birds, but most do and some do very well. Do planes fly? Yes they do, but they fly in a very different way to birds. Planes don’t have feathers and they don’t flap their wings, but they can fly great distances and carry much more weight than even the largest bird. Therefore, flight is something that birds and planes both do but by using different methods. Birds are living animals that have evolved to fly and planes are engineered artifacts; machines that we have designed to fly.
Computers, like [IBM's] Watson, are machines that we have engineered to think. Watson isn’t made of flesh and bone and it doesn’t have a brain, but it appears to think, just not in the same way that we do. For some reason when it comes to intelligence and consciousness we are much more sensitive about the abilities of our creations. If we engineer a machine that performs as well as birds we proudly claim it flies but if we engineer a machine that performs as well as or better than people in a game show we doubt it’s thinking. I believe in the future the question of “do computers think?” will be one that most people will not even consider. We’ll just all take it for granted that computers act as if they are thinking and that’s good enough. Philosophers will still be arguing about this in the future and the religious will always believe that machines don’t have souls.