Talk to my hand

[Adrian II]

The New Scientist (http://www.newscientist.com/channel/info-tech/mg18624961.700) has a decent article on artificial intelligence minus the hype, that is: more artifice than intelligence. Here's a conversation that the author struck up with 'Intellibuddy', a state of the art version of a chatbot called ALICE (Artificial Linguistic Internet Computer Entity) that was invented in 1995 by a Richard Wallace of San Francisco.


Intellibuddy: Hi, I'm Intellibuddy! Nice to meet you.

JM: I'm writing an article about artificial intelligence.

Intellibuddy: Will you include me in your essay?

JM: I will if you can persuade me you really are intelligent.

Intellibuddy: Would you recommend this action to a friend?

JM: Errr...are you intelligent?

Intellibuddy: Ummm I am highly intelligent. I have many intellectual functions.

JM: What are your intellectual functions?

Intellibuddy: Thought, reason, intelligence, logic and deduction.

JM: What is your IQ?

Intellibuddy: My IQ is about 250 in human terms.

JM: Wow, you're cleverer than my editor.

Intellibuddy: I'm glad your [sic] impressed, User.

You think I am cleverer than my editor.

JM: No, cleverer than my editor.

Intellibuddy: I'm glad your impressed, User.

You think I am cleverer than my editor.

JM: No, cleverer than my editor.

Intellibuddy: I'm glad your impressed, User.

You think I am cleverer than my editor...

The article then highlights the attempt by a developer named Doug Lenat to create a new generation A.I. system, 'Cyc' (pronouced 'psych') that will be aired in a couple months, and is supposed to pass the famous Turing test: a human who 'speaks' to it will not be able to discern that he is talking to a computer.

What you guys think? Would you recommend Doug's confidence to a friend?

[Dooz]

Must be one lonely person to resort to artificial intelligence for conversations...... ~:handball:

[Craterus]

Must be one lonely person to resort to artificial intelligence for conversations...... ~:handball: Lol, that's one way to put it..

[English assassin]

I can't see any reason why artificial intelligence should be impossible, but it seems to me most of the claims to date have involved self-publicity rather then real science (Yes Kevin Warwick this means you).

IMHO we will need to understand a lot more about how our own (rather questionable) intelligence arises before we stand much chance of designing a system which mimics it. (And I use mimics advisedly, Turning was no fool in limiting his test to the external appearance of the system rather than asking if it is "really" intelligent, though it would be more accurate to say that its a test for communication rather than intelligence. The communication makes some mental activity visible, but its not the same thing, as anyone familiar with the internet knows only too well...)

Roger Penrose has a theory that there is something in the workings of the brain that makes artificial intelligence impossible. I freely admit I cannot follow his reasoning there at all (though he is very clever so I am sure its tenable) but if you guess that intelligence, like self consciousness, emerges from a sufficiently complex computing set up I can't immediately see why the set up has to be biological rather than silicon. But its possible, you'd need to be a materials scientist-neurobiologist-computer scientist to have a really well informed view. (And according to Penrose you'd need to understand quantum gravity too which is a bit of a poser right now)

You can see why its a difficult field can't you?

(BTW NS might be a bit hard on intellibuddy. My two year old also has difficulty with first/second person pronouns, often announcing "I get out of bed now" when what he means is "You get out of bed now" (the little rat). But he is, of course, the most intelligent two year old in the whole world ever.)

[R'as al Ghul]

It's been a while that I've read Turing's paper,
but iirc his idea on intelligence was that
it is a matter of perception. He staged a test
where a woman would answer questions as if she was
a man and a man would answer as woman, both of them as
best as they could. The person posing the questions and getting the answers
was not able to distinguish whether the persons were male or female.
Yet they remained male and female, of course.
Now, if you substitute gender with intelligence and construct a machine that
gives answers indistinguishable from humans, then the person getting those answers
wouldn't be able to tell if it was a machine or a person answering.
Thus, it doesn't matter if the machine is "really" intelligent because it
is perceived as intelligent.

Apart from the loop in the initial example by AdrianII, the experiment is very interesting
and I've seen better dialogues/ results with it. I think that trying to simulate
intelligence or brain-functions at all is challenging and can tell us a lot about
our brain and how it works. Nevertheless it will take us a long time before we
finally can understand our brain and the rather abstract term "intelligence".
I would for example argue that most animals have a special intelligence which enables
them to survive. But, the intelligence of a bird is completely different from a dog's.
It's impossible to compare these forms of intelligence because they obviously serve completely
different purposes.

R'as

[English assassin]

I would for example argue that most animals have a special intelligence which enables
them to survive. But, the intelligence of a bird is completely different from a dog's.
It's impossible to compare these forms of intelligence because they obviously serve completely
different purposes

A good point, which generalises the communication point a bit. It seems to me quite likely that there can be intelligences we would not recognise, or not immediately recognise. As a social species a lot of our intelligence is directed to social interaction between individuals. Basically, its designed by sexual selection to be good at outwitting/impressing/co-operating with other human intelligences. We would be good at recognising that, and of course that is exactly what the Turing test focuses on. Intelligence in a solitary animal, or in a hive living species, might have very different features. Dialogue between the two would be much more difficult than just establishing a common language.

Intelligence in an artificial system might well be still more different. But I guess the question should be understood to mean "can we create a human-like artificial intelligence"

[A.Saturnus]

Roger Penrose has a theory that there is something in the workings of the brain that makes artificial intelligence impossible.

As someone pointed out, Penrose's theory explains as much as pixie dust. And the problem with his arguments against strong AI (the idea that computers can be intelligent) is that he doesn't have any. He just says "it's impossible" in more complex words.
Some people have argued well against strong AI (as Searle for one) but in the end all such arguments fall short. Unless you propose some really bizarr non-naturalistic theory of the mind, the computationability of everything can't be discussed away.

Said that, any claims that the Turing test will be passed soon are nonsense. Please note that the interview above is an unfairly light probing of intelligence.