h2g2 The Hitchhiker's Guide to the Galaxy: Earth Edition

I was reading an old discussion on the Turing Test, and it described it as "universally accepted." To review, the Turing Test is a method of testing for computer intelligence, which is often construed to mean consciousness. I'm not sure if that's what Turing meant, but that's what writers here seem to think.
In the Turing Test, a human being and a computer are interrogated under conditions where the interrogator does not know when he or she is talking to the human and when he or she is talking to a computer, the communication being entirely by textual messages. Turing argued that if the interrogator could not distinguish them by questioning, then it would be unreasonable not to call the computer intelligent.

Philosopher John Searle has pointed out, completely rightly, that the Turing test is an invalid method of determining whether or not a computer is sentient because all that it shows is "as-if" behavior. The computer is just behaving "as-if" it is sentient, but it might not be. Passing the Turing Test might be evidence of sentience, but it does not constitute sentience. There's a difference between evidential facts and constitutive facts. Looking at your gas gauge in your car and seeing it hitting empty is evidence that your car is out of gas, but it does not constitute it being out of gas: your car's tank in fact being empty consitutes your car being out of gas. Your gas gauge might just be broken (In fact mine is!). John Searle's famous "Chinese Room Argument" shows why the instantiation of a computer program is not sufficient for intelligence. This should not come as a surprise, but at the time when Searle wrote the paper, artificial intelligence engineers were talking like it was, making claims like the following: thermostats have beliefs, they believe it's too hot when the temperature gets to a certain level, they can believe it's an appropriate temperature, or they can believe it's too cold when the temperature gets too low. This is absurd. Consider Searle's argument:
Imagine there is a man sitting in a room with a window. Through the window, he receives pieces of paper, and all of them are Chinese characters. The man is completely ignorant of Chinese (he speaks only English), so the characters are merely squiggle-squaggles to him. In the room, there is also a massive book, as well as stacks and stacks and stacks of pieces of paper with other Chinese characters. The book instructs the man how to deal with the characters that he receives with instructions like this: "When you see (squiggle-squaggle), replace it with (squoggle-squiggly)" The process of dealing with the characters he receives is difficult at first, but with practice, he learns all the rules. Every time he gets a particular set of characters, he knows which characters to replace them with, and he sends these back through the window. As it turns out, the characters he is receiving are questions in Chinese, and the instruction book tells him how to give answers in Chinese. The instructions were actually good ones, and to a Chinese person, it may appear AS IF when asked, say, "Who is your favorite philosopher?" that the man in the Chinese room responds with the intelligent answer "My favorite is Daniel Dennett, but I'm also partial to Thomas Nagel." But the man in the Chinese room really has no idea what he is saying with the Chinese characters, because obviously, by simply following rules of the type described, his understanding of Chinese does not increase at all. The man is using syntactical rules but has no semantics (meaning). Although he's just a man in a room, he falls beneath the definition of a computer program. This is just like what computers do: they have a set of rules that are programmed into them by human beings through computer language that is translated into electrical signals and ones and zeros. They have syntax but no semantics. As far as we can tell, they are not conscious, and they don't have conscious thought processes or beliefs. They don't know what it is like to see red, and attaching a video camera to them isn't going to help.
This is not to say that computers could never be conscious. Searle doesn't deny this either, but some people have put those words into his mouth. However, achieving computer consciousness would require the actual discovery and in depth study of the neural mechanisms in the human brain that produce consciousness (if they can be found), and these mechanisms would have to be somehow mimicked with computer hardware.
The point is, even though your computer sometimes seems as if it harbors a hatred for you, which it expresses from time to time by crashing on you, it doesn't mean that it really does. The Turing Test just shows "as-if" mentality, it does not prove that there is real mentality, and is therefore invalid!

Hi Whubel,

We actually have an Edited Guide entry on the Turing Test at A168365

You may want to post this on that page where others have commented, or alternatively you could start a discussion on The Forum at A1146917

You are more likely to get a response there as more researchers will see your post



Mort

Don't have time to comment in depth right now, but a few quickies.

Are the Turing Test and the Chinese Room to be taken s tests of intelligence or consciousness?

Is there anything that could constitute a test of consciousness?

What do you think needs to be added to syntax to achieve semantics, nd what do you think needs to be added to rule following information processing to achieve intelligence?


Noggin

I quite agree with the assertion in the title. Machines have already fooled many people into believing they were other people; to interpret this as machines having consciousness or intelligence is a huge leap of faith.

<>

This highlights the other flaw in the Turing test. It assumes that true form of intelligence must be indistinguishable from a human. Given that no-one really knows what consciousness is, why do we always assume that any other conscious being must be a) based on the human model and/or b) behave exactly like a human? Surely a conscious machine would have a completely difference set of motivations.

Exactly. Computer consciousness could probably be achieved with some mechanism other than what is responsible for it in humans. In other words, you could have the following three beings at once: first, a super-sophisticated robot who acts just like a human being but who is NOT conscious, standing next to a real conscious human being who is indistinguishable from the first robot in behavior, standing next to a super-duper sophisticated robot who is in fact conscious and behaves differently from the first two (how does it behave? who knows). What follows from this is that the Turing Test is invalid because the first robot and the human are behaviorally indistinguishable but the robot is not conscious, because it lacks the mechanism that makes robots conscious. Additionally, the fact that the third robot behaves differently from the first two demonstrates an important point: consciousness must serve some purpose that is evolutionary significant, otherwise, it wouldn't have evolved. It's two rich and magnificent to be an accident, and it couldn't have just popped into existence when animals started reaching a certain level of neural complexity without some kind of witchcraft.

Also, passing the Turing Test is not a very good test for intelligence, but in a different sense. What I mean is, clearly, if a computer could pass the Turing test, we might consider it intelligent. But wouldn't we also already consider Deep Blue, the chess-playing supercomputer that defeated Grand Master Gary Kasprov to be intelligent, even though it could not even participate in a Turing Test? Maybe even it could, though, we wouldn't consider it intelligent; it depends on the meaning of the concept of intelligence. Would you describe your computer as being intelligent, or just as a computer? If it turned out that Einstein was actually an unconscious android sent from the future, would you think of him as just as intelligent as when you thought he was human? It depends on whether your concept of intelligence is a behavioristic one or a mentalistic one. The dictionary puts it as both:
Intelligence:
a. The capacity to acquire and apply knowledge.
b. The faculty of thought and reason.

When describing people it's both: when we say that a person is intelligent, we imply that he or she both behaves intelligently and also has intelligent, conscious thoughts. When applied to unconscious robots , however, like the robot that they sell now that cleans your floors, we might say that the robot simply behaves intelligently, but it would be a bit inappropriate or confusing to say that it actually is intelligent.

I thought a test of coinsciousness was like pinching yourself to see if you are awake.. Or what of seeing that the lights stay on no matter which way the switch is facing, the pages of books, screens,clock faces,can't be read when looked at in the face and or rooms start to mutate/melt while you find yourself encountering all sorts of strange and fantastic random people and beings?
Guess that would be more of an am I dreaming/hallucinating test, or a "stream-of-consciousness" test. Have any of you folks seen an art- movie titled Waking Life? It is by some guy named Linklater or something like that, and involves an ingenious/extensive use of roto-scoping techniques. The thing is thoughtit takes a while to get used to the crawly backgrounds that occur usually when the main character is "indoors."

The Chinese Room is just a piece of stupidity: all it proves is that Searle is not able to think straight. We should not consider it further.

The Turing Test is debatable - Turing thought that it would show whether a machine is intelligent or not. Not every agrees. It is certainly not universally accepted.