Philosophy

5

LECTURE: The Turing Test and Searle’s Chinese Room

Artificial intelligence

The computer model of the mind suggests the possibility of artificial intelligence (or AI). If the mind is just a kind of software, then wouldn’t a machine running the right kind of software have a mind, no less than we human beings do?

Philosopher John Searle distinguishes two approaches to artificial intelligence. The first he calls Weak AI, which is simply the method of studying the mind by creating computer simulations of various mental phenomena. Searle has no objection to this. It is like studying weather systems or digestion by writing computer simulations. It has no special philosophical implications.

What Searle calls Strong AI is much more ambitious and controversial (which is why it is called “strong” – it makes a stronger or bolder claim). Strong AI holds that a mind really is nothing more than a kind of computer program, so that an appropriately programmed computer not only simulates the mind, but literally has a mind. This is the view that follows from what we have called computationalism. As we’ll see, it is a view that Searle has famously criticized.

We have real intelligence! Artificially, of course.

The Turing Test

Suppose you try to create a machine that can think. How would you know whether you’ve succeeded? Think of the stilted, canned sort of responses that an AI program like Alexa or Siri gives. We don’t think of them as genuinely intelligent, because they don’t respond the way that human beings do, and human conversation is our “gold standard” of intelligent behavior.

But suppose a machine could be programmed in such a way that it did respond the way human beings do. This is the idea behind what is called the “ Turing Test” for determining whether a machine is intelligent – named for Alan Turing, who we talked about last time and who came up with the concept. The basic idea of the test is that if, in conversation, a machine could produce answers indistinguishable from those a human being would give, then it can be said to think.

To modernize Turing’s original scenario a little, suppose you are engaged in an internet chatroom discussion with two other conversation partners. You are told that one of them is a human being, and the other one is a computer program designed to mimic the behavior of a human being. You are asked to try to figure out which is which based on what each of them says in the course of your conversation. Let the conversation go on as long as you like. Suppose that, no matter how long it goes on, you simply cannot figure out which one is the machine running the program. Its answers are not stilted or canned like those that Alexa or Siri would give. They are instead just the kind that a human being would give.

The Imitation Game: Which is the real Turing?

If this is what happens, Turing argues, then we have all the evidence we could need in order to justify concluding that the machine is genuinely intelligent.

Searle’s Chinese Room scenario

There has been a tremendous amount of controversy about whether it is likely that we will ever build a machine that can pass the Turing Test in a rigorous test situation. But also controversial is the question whether it matters. In an extremely influential argument, Searle claims that it does not. He argues that, even if a machine could pass the Turing test, that wouldn’t show that it is really intelligent. And for that reason, he also argues, Strong AI and the computationalist claim that the mind is a kind of software are false.

John Searle

The argument begins with a thought experiment. Searle asks us to imagine that he, Searle, has been given the task of sitting in a room and answering questions that are written in Chinese and slipped to him through a window in the door. However, Searle also has no knowledge at all of Chinese. To him, the symbols on the paper just look like meaningless squiggles. So how is he supposed to answer the questions?

The way he does it is as follows. He is given a box full of multiple copies of cards with all of the characters of the Chinese alphabet written on them. He is also given a rulebook written in English which tells him how to combine these symbols into answers to the questions he is given through the window. The rulebook does not tell him the meanings of the symbols. Rather, it tells him things like: “When you see such-and-such a series of symbols, reply with this other string of symbols.” He then has to tape the appropriate series of cards down on a piece of paper and give that out through the window as a response to the question he was asked. The rulebook (which is massive) tells him how to do this for every possible question.

Suppose further that Searle eventually gets very good at this, flipping through the rulebook quickly and taping the answers down so as to provide a speedy response to every question put to him. Imagine he gets so good at it that the people asking him the questions assume that he speaks the language, and does so very well.

But it’s just an illusion. In fact, even though Searle behaves as if he understood the language, he still has no idea what the symbols he is manipulating mean. In effect, he passes the Turing Test for intelligent use of the language, but he doesn’t really know it at all.

Cartoon of Searle in the Chinese Room. (Courtesy of the Internet.)

The lesson of the example

On the basis of this thought experiment, Searle argues as follows. He is, in the Chinese Room, basically doing what a computer does. He is manipulating symbols according to the rules of an algorithm, like a Turing machine. In the case of the Turing machine, it is 0’s and 1’s that are being processed via the rules of its machine table, and in Searle’s case it is characters in the Chinese alphabet that he is manipulating via the rules in the rulebook. But the basic principle is the same. Certain inputs lead to certain outputs, with an algorithm coming in between.

Now, both Searle and the Turing machine can do what they do because they are able to detect the physical properties of the symbols, such as their size and shape. They are in no way sensitive to the meanings of the symbols. The Turing machine does not know that a round shape is a zero and a straight shape is the number one, and Searle doesn’t know what the symbols he is processing mean either. In both cases, it is only how the symbols look (either to Searle’s eyes or to the electronic eye of the Turing machine) that matters, and not the meaning.

Searle puts this by saying that computers are sensitive only to the syntax of symbols, but not to the semantics of symbols. Syntax has to do with the rules governing the order of symbols, whereas semantics has to do with their meanings. And what the Chinese Room example shows, Searle says, is that mastering the syntax of symbols even perfectly is not enough to give you knowledge of the semantics of the symbols.

But in that case, he concludes, computationalism and Strong AI must be false. The reason is this. To have genuine understanding or intelligence requires grasp of the semantics or meanings of symbols. But running a program can only give something mastery of the syntax of symbols, and never of their semantics. So, running a program is not by itself enough to generate genuine intelligence. There is simply more to the mind than just that.

The argument in a nutshell

In summary, Searle sums up his Chinese Room argument as follows:

1. Computer programs are purely syntactic: they are sensitive only to the “shapes” or physical properties of the symbols they process.

2. But genuine understanding and intelligence involves grasp of the meanings or semantics of symbols.

3. And syntax is insufficient for semantics (as is illustrated by the Chinese Room example).

4. So, running a computer program is not sufficient for having genuine understanding or intelligence.

The most a computer could ever do, then, is mimic or simulate intelligence, but not really possess it. And so, if all the brain was doing was what computers do, then we wouldn’t have the intelligence we do. Turing machine functionalism is mistaken.

Not as smart as it thought it was.

It is worth adding that Searle is not a dualist. He does not think that the mind is something immaterial. He thinks it depends entirely on the brain, though he doesn’t think we know at this point how that works. He just thinks that the computer model of the mind does not in fact provide a good materialist explanation of intelligence.

The Systems Reply

Searle’s argument has generated an enormous amount of controversy. Let’s consider one of the objections raised against it, known as the “ Systems Reply.” The objection begins by noting that what runs a program is a computer system as a whole. It isn’t your keyboard, or flash drive, or even the CPU of your computer that runs software, but the computer as a whole operating as a system that does so.

Now, in the Chinese Room example, Searle is part of a larger system – the room as a whole – that can be said to be running the program or algorithm embodied in the rulebook. Searle is like the CPU, whereas the window is like an input/output device (with the questions passed through the window being analogous to what is typed into a keyboard, and Searle’s responses being like what appears on a monitor).

What Searle’s argument shows, the Systems Reply says, is only that Searle, who is merely a part of the system, does not understand Chinese. But maybe the system as a whole – consisting of Searle, plus the rulebook, plus the cards with the symbols on them, plus the room, etc. – does understand.

Searle responds to the Systems Reply as follows. Imagine that Searle memorizes the rulebook and the symbols, and answers the questions put to him just by consulting his memory and writing down the sequences of symbols in response to the questions (while still not understanding their meaning, but just doing what the book tells him). In this case, the system is entirely internalized in Searle himself. He is the system as a whole, and yet he still doesn’t understand the language. So, the Systems Reply fails.

“Bro, I am the System!”

Here are some optional YouTube videos, including a lucid summary from Searle himself:

TED-Ed video on the Turing Test:

https://www.youtube.com/watch?v=3wLqsRLvV-c

Searle on the Chinese Room argument:

https://www.youtube.com/watch?v=18SXA-G2peY

Minds vs. Machines: The Turing Test and the Chinese Room:

https://www.youtube.com/watch?v=qrTqmWbYD3Q