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

I'm a bit dubious about the way the Turing Test description is phrased. It implies that in order to be recognised as sentient, an AI would have to pass the Turing Test.

However, I think it was really intended the other way round; if an AI can pass the Turing Test, then it is regarded as sentient. This doesn't exclude sentient AIs that can't pass the test.

For example, imagine subjecting Mr Spock to the test. There would be no doubt that he is not human, but there would be equally no doubt that he is sentient.

In summary: if it can pass the test, it is sentient. If it can't pass the test, it may or may not be sentient; the test won't tell you.

I'm doing Metaphysics in Philosophy this semester and we're looking at AI at the moment. In fact we just did Turing the other week, so I'm a bit dubious when you say that the Turing Test is the universally accepted test for sentience. In fact, John Searle provided an arguement, The Chinese Room, that pretty much debased the Turing test. (for explainations of this follow the Turing Test links and read the forum responses).

I thought that the universally accepted way of testing sentience on anything was the Mirror Test. Here, the subject is placed in front of a mirror and is allowed to poke around and try to figure out what's going on. The mirror is then removed and a blemish (usually a dot in texta) is placed on the subjects face. The mirror is then returned.

If the subject is sentient, it will see the reflection, see the blemish on the reflection, and will be able to raise their hand (or equivalent) to thier own face (not the reflection) to try and remove the blemish, therefore showing that the subject has an understanding of itself as a seperate/individual entity as it is able to recognise itself from others of its kind.

I believe this test has been used to prove that dolphins, some of the great apes and some chimps have sentience.

I don't see how that sort of behavior-based test could work with computers. You could simply program them in advance to do whatever it is they needed to do to fulfill the test qualifications. A good enough programmer could hide this instruction set from the testers, thus providing the illusion of sentience where none exists.

For any theory of intelligence, there is a counter-argument that debunks it. Sentience is a bit like the term "pornography." Nobody can agree on what exactly constitutes pornography as opposed to erotica or even romance novels and R-rated films. Different people seem to have wildly different interpretations of the word. But we all maintain the illusion that we know what it is, and claim that "we know it when we see it." Sentience is much the same.

The Turing Test has retained its place among programmers as the ultimate test of sentience because it was the first test devised, and it seems no worse overall than the others which have come after. Philosophers, of course, may have different favorites. But in the end, they are not the ones programming the computers.

Personally, I think that the Chinese Room Argument is a load of bunk, and I'd place good money on there not being a single programmer who thought it had any relevance whatsoever to anything. This is the problem with people philosophising about stuff they don't understand...

There's an entry somewhere in h2g2 on the chinese room argument - we should take this there...

That definition of Sentience is basically self-awareness, and it is (in my opinion) a load of bunk. Self-awareness has already been achieved in the limited sense - computers already auto-detect their processor speed, memory size, mice, peripherals, etc. They are aware of when they are about to 'thrash', which is basically equivalent to panic in humans. They can take steps to avoid thrashing. They can work out how their hard disks are being loaded, and move stuff around to compensate. They can deal with processes that go astray, though currently this is a little all-or-nothing, including the process that deals with processes that go astray. I'd say if you add all that stuff up, you get a pretty complete concept of self emerging.

Being self-aware is, of course, trivially simple if there's not an awful lot of self to be aware of. If you wrote a program which does little, it can easily be self-aware - for example, consider the program which waits a second and then busy-loops forever. This program has only two states it can be in, and can be made completely self-aware by the addition of a boolean which starts false, and is set true when the program starts busy-looping. But intelligent it ain't.

Anyway, I thought the Turing Test tested intelligence, not sentience?

it's at http://www.h2g2.com/A399549

Just to clarify: I didn't mean that I thought the Turing Test was a valid way of determining sentience. I was talking about what the Test is *designed* to do (rather than what it actually does, which is another argument).

My point is that there may be sentient beings (or AI) that do not pass the test because they do not behave in a *human* way, even though they would be regarded as sentient.

So the Test states that any being that passes it must be sentient. However, it is clearly not the case that any being that fails the Test is definitely *not* sentient.

I'm fairly sure that any being that passes the test has to be intelligent, but may or may not be sentient, depending on the definition. (IE, some people define sentient as human, which makes the question rather pointless). Turing himself claimed to be answering the question "can machines think" (after defining 'machine' and 'think').

What I don't understand is how any test can test sentience with complete objectivity. Any answers that we recieve are interpreted with a distinctly human viewpoint towards the concept, and therefore any sentience which falls outside our limited perspective is considered nonsentient. We also have expectations as to what AI can do, based on our thinking of computers. For instance, consider the following dialogue:
-A: What is the square root of 16 plus 3?
-B: I don't know, I'm not good at math.
Is B a machine or not? We would assume not, because our concept of AI as being housed in a computer assigns specific functions of said computer to the AI. Figuring the answer is a relatively simple operation for a computer, so the AI should be able to do it to, right? Wrong. The human mind can do differential calculus at phenomenal speeds and with such accuracy that it can accurately predict the trajectory of a thrown object, then proceed to catch it. The mathematics involved in this are astoundingly complex, yet the average 8-year old can catch a ball. Can they 8-year old do differential calculus? Probably not. The question we never think to apply to AI is "Why would it want to know how to do math?"

A book by David Gerrold called 'When HARLIE Was One' had a pretty nice take on the Turing test, and although I read the book as a kid some fifteen years ago I can still remember many of the details because they were so vivid. I highly recommend it.

As long as we're proceeding into general semantics here, I have long thought it interesting that we often automatically refer to the notion of machines that 'think' as 'artificial' intelligences. It sort of begs the question of what we mean by artifice. We have yet to properly agree on contextual definitions of what we mean when we talk about ourselves as being intelligent, sentient, or whatever.

Hang on a minute.....

In the first section of his report, Turing stated that the question 'Can machines think?' was "absurd" (in the first paragraph in fact).

Did he mean absurd as in "That's the stupidest thing I've ever heard, what are you daft or something" or absurd as in "Of course they could if they were built right, and it's completely self-evident you ignorant peon"? The reason I ask is that while the first would seem to invalidate the Turning test as being simply his method of proving that any intelligent-seeming machine you may happen to encounter is not sentient and therefore not worth your time, given that it can't be confused with a human, but the second shows a genuine interest in the question.

He means it's a dumb question to ask - the wrong one.

To quote:
"Instead of attempting such a definition I shall replace the question by another, which is closely related to it and is expressed in relatively unambiguous words... We now ask the question, 'What will happen when a machine takes the part of A in this game (The Imitation Game)?'"

If you can find his article, it's a good read and (amazingly enough) it explains fully what he hoped to achieve with the Turing Test. I'm pretty sure it mentions something about testing the behaviour of the computer, because he argues that if the computer behaves as though it understands then for all intents and purposes it *does* understand. And it's on this claim that John Seale really digs his teeth in. (That is if I recall the article correctly and didn't fail the exercise I had to write on these.)

Alan Turing, 'Computing Machinery and Intelligence', from "Mind", 1950

John Searle, 'Minds Brains, and Programs', from "The Behavioural and Brain Sciences", 1980

Much to my irritation, Yahoo, Excite, Altavista, and even the unedifying Jeeves have failed to find the articles. One of them did, however, return a a rather interesting site along with the usual unrelated nonsense search engines are so infamous for. http://www.mrmind.com is a reversal of the Turing test where you try to convince a bot that you're human. I think it demonstrates how difficult the Turing test could be if the person talking to the bot wanted it to fail.

An interesting link. Thanks.



Amy the Ant

Also, try www.turing.org.uk

Don't know if it's any good, but the lecturer gave it to us and sang its praises so it's probably worth a look.

Thanks.

I didn't know he had anything to do with Enigma. Very interesting parallels can be made there.

If any1 has watch Star Trek: TNG for a while, they will probably know that the andrioid Data is considered as a sentient being. They used a set of 7 criteria that he had to fulfill to be considered sentient by the Federation. I'll list as many as I can remember, but if any1 can correct or add the others, then please do:

The being must:
1. use some form of respiration.
2. be able to reporduce.
3. have emotions.
4. be mobile.
5. be able to communicate.
6. be able to die. (even 'droids eventually cease to function)
7. ?

Someone fill me in on rule 7 please.

Hope this helps
Athon Solo

Of course, there is the classictest of:
I think, therefore I am.

Although some would say this applies only to existance, rather than sentience. Although, technically, (I think this was in Red Dwarf) a robot is only thinking it can think so:
I think I am thinking, therefore I am!

Ifthat makes any sense, which it probably doesn't.

Athon Solo

wtf has reproduction to do with sentience, exactly? Do people cease to be sentient when they become infertile!? Would an immortal God be considered non-sentient by these people? What about something which gains energy from sources other than respiration?