Chapter 4: Thinking
60 - Smart Machines (Part Three)
In 1980, Berkeley philosopher John Searle published his “Chinese room” parable, which would become the classic debunking of classical AI. In it he assailed the Turing test by asking the reader to do the following:
Imagine yourself sitting in a room, provided with a list of Chinese symbols and rules to match them. Your only connection to the outside world is a slot. Through this slot, you receive input in the form of certain Chinese symbols. According to your lists, you replace the incoming symbols with others and pass them back through the slot as output. For the people outside the room, the input was a question in Chinese, and the output was a reasonable answer to it also in Chinese. They would be justified in supposing that a Chinese speaker is inside the room.
But, Searle asked, would you yourself say that you understood Chinese? Clearly not. It was just as clear, he went on to say, that regardless of whether they could pass the Turing test, classical AI programs didn’t know what they were doing.
Berkeley computer scientist Lotfi Zadeh added that in order to confuse a machine in the Turing test, you need only ask it to summarize what you tell it. “No machine will be able to pass this test,” he said. “You need no other test, just ask the machine to summarize what you said or typed.”
By the end of the 1980s the AI landscape was like a beach littered with empty shells, the remains of bankrupt AI companies. As Minsky would later concede, “The main problem seemed to be that each of our so-called ‘expert systems’ could be used only for some single, specialized application…None of them showed any signs of having what we call common sense.” In the early 1990s DARPA announced that it was shifting $500 million in research funding away from classical AI and awarding it instead to a new approach—parallel distributed processing. That approach looked to how information is actually processed in the brain. It was there that the next advances would come.
