Archived

This topic is now archived and is closed to further replies.

Extrarius

Where is the AI that can pass the Turing Test?

Recommended Posts

I''ve seen all the various parts of the AI required - there have been successful experiments in exploratory algorithms, in text to internal representation to text, in text to an ''understanding'' of it (programs that can summarize large amounts of information and then answer questions about the information, and make inferences and answer questions not explicitly answered by the information), and in various other algorithms that would be needed to make an AI that can converse well enough to be mistaken for a human. ...But there isn''t a program that can pass the turing test yet. Is there any reason the various technologies haven''t been integrated in order to pass the test? I can understand that many people think the test meaningless, but it seems that if nothing else, it would help bring in money for the ''real'' projects.

Share this post


Link to post
Share on other sites
If the technology is there, maybe the problem is how to pass the Turing test? I know many research labs don''t officially participate in the Loebner prize because they compete against complete amateurs. Given the format of the judging, I guess they are worried of the negative publicity involved in not "winning."

Alex

Share this post


Link to post
Share on other sites
Is important for AI to attempt to pass a Turing test? What if AI would decide it would be harmfull? I don't think that a Turing test is effective for testing AI. And we could believe that majority of Real AI would think simillary about that.

Alas Turing test is rather simple and unexpesive. In real live there are
1. complex and/or expensive things.
2. cheap, simple, and not working things (with few exceptions)

Is that test one of these exceptions? I don't think so. Also Turing test wouldn't help us with a AI development.

[edited by - raghar on January 24, 2004 5:43:11 PM]

Share this post


Link to post
Share on other sites
Assume that someone has already made such an AI that can pass
the Turing Test.

What I would like to see is what happens when you put two
of them together and have them "converse".

Oh, wait a minute, is "Extrarius" a pseudonym used by
a computer AI right now? Hmmm... If that is the case,
then congratuations to the author of "Extrarius Bot". You
have fooled me.



Kami no Itte ga ore ni zettai naru!

Share this post


Link to post
Share on other sites
quote:
Original post by Extrarius
I''ve seen all the various parts of the AI required - there have been successful experiments in exploratory algorithms, in text to internal representation to text, in text to an ''understanding'' of it (programs that can summarize large amounts of information and then answer questions about the information, and make inferences and answer questions not explicitly answered by the information), and in various other algorithms that would be needed to make an AI that can converse well enough to be mistaken for a human.

To be honest, I was under the impression that most of these situations were all far from being solved. Do you have any references?



[ MSVC Fixes | STL Docs | SDL | Game AI | Sockets | C++ Faq Lite | Boost
Asking Questions | Organising code files | My stuff | Tiny XML | STLPort]

Share this post


Link to post
Share on other sites
I'll try to find the sitesagain (its been a long time since I looked at them, I was just thinking about it and decided to finally post this), and I may be incorrect about the projects (maybe they were 'AI parlor tricks' or somesuch, but it didn't sound like it).

On most of the projects I'm thinking of, I only have thirdhand+ knowledge (mainly because the papers for most aren't online, so my knowledge comes from websites linked to by bishop_pass). The exploratory engine I'm thinking of is on its 5th or so iteration, and is now a commercial project(CYC I think it is). Most of the projects were also done in the 80s to mid 90s, so I'd figure they would be even further along by now.

/me goes to find some of the links

[edited by - extrarius on January 24, 2004 11:09:00 PM]

Share this post


Link to post
Share on other sites
Guest Anonymous Poster
Actually, if you check out www.a-i.com they are doing work on an AI that supposedly can pass the Turing test. (For infants anyway) They give the AI (HAL, I believe its called) no previous understanding of language, and teach it the same way a child is taught language. They have taken transcripts of dialogue with HAL to child-development specialists, who returned the transcripts saying HAL is a perfectly normal little boy (of about 30 months). (They of course, had no idea that HAL is actually an AI)

Share this post


Link to post
Share on other sites
Guest Anonymous Poster
To pass the Turing Test, you essentially need a program with a deep knowledge of everyday things. This is much easier said than done. There have been a number of attempts to deal with this, but none successful.

Share this post


Link to post
Share on other sites
My car could pass the turing test... Im convinced it''s the home of a small gnome who livs somehwere inside the body of the car.

This little guy hates getting up on cold mornings, but likes to stay up late, and sometimes just doesn''t want to start working. On hot days he''s more than cooperative when it comes to putting the top down, but not when it''s cold out. He unfortunately seems to enjoy huffing gas fumes, and will refuse to budge if I cut off his supply...

Smiles,
Will

Share this post


Link to post
Share on other sites
LOL, RPGeezus.

Well, my point, as usual, is that I don''t care if it can pass the Turing Test as long as it can make me a sandwich. We don''t need machines to "think." We just need them to solve our problems.

I also have doubts, you know, that some acquaintances of mine could pass the Turing Test... even in person.

"You mean he was HUMAN?!"

Share this post


Link to post
Share on other sites
*ROFLMAO*

Shall we call it The Anti-Turing Test?

It''s a test taken by a "human", to be judged by a machine
program (or another human) whether the test subject is "human"
or "machine".

*LOL*



Kami no Itte ga ore ni zettai naru!

Share this post


Link to post
Share on other sites
The Turing test is a lot of hype, you know.
If you read the description by Turing, it''s rather limited in its purpose. I think Turing was just making a point when he thought the idea, and now everybody seems to think it''s a bloody Grail of AI or something.
In any case, being able to have a conversation is never a *proof* of intelligence. Only of the possession of articulated speech.
I mean, if you are mute, are you stupid ? I think not !

Sod the Turing test, people trying to pass it are only limiting themselves on very specific problems and very specific tools to sovle those problems. By doing so they inevitably limit themselves, and thus fail to pass the Turing test...
IMHO, as long as you think of the Turing test as only a language problem, you wont solve it. Period.

And the intelligence is not in the language.


Sancte Isidore ora pro nobis !

Share this post


Link to post
Share on other sites
To quote Johnny Cash: "Does an engine get rewarded for its steam?". Now that is a question worth considering...

I wrote a Connect-Four game a few years back to defeat my sister. She consistantly beats me whenever I sit across from her. She lost the first game against my program. It was glorious to hear the computer taunt her as the game-tree became aware of impending victory. Chipmunk style recordings of "haha", "sucker", "you suck", and "looser" filled the room after each futile move. After the long battle I claimed victory. Who won? Was it me, or the computer? Think carefully.

I ask this because I''m thinking about getting a robotic dust-buster (Roomba). My plan is to turn it on before I go to work, and empty it''s dust bin when I get home to newly cleaned floors. Sunday morning (chore day) my wife will gently remind me that it''s time to get to work, and I plan on saying: "Chores??!?! I''ve been cleaning all week!!".

What do you think her response will be, and why? Is it justified? Will I deserve the retribution that will surely come my way?

Cheers,
Will





Share this post


Link to post
Share on other sites

The Turing Test was, I suspect, a philosophical joke played by Turing on the AI community. In order to pass the test, an AI must not merely be intelligent, but also capable of impersonating a human being.

Many of our thoughts, memories and opinions are shaped by the world around us and all that came before us. Not only do we have our own history to consider, but also our own inheritance, both genetic and instinctive that is born of millions of years of evolution.

Why, for example, do we love music? Why do we read novels, play games and whisper sweet nothings into our other halves'' ears? What do we mean by "blue"? What is "love"? Why do people have faith in the existence of a god or gods?

Why is a kitten cute, but a crab not?

So many of our frames of reference are utterly alien to an artificial intelligence that it is not going to be able to pass itself as a human being unless we pre-program that information into it from the beginning.

Humans have never been very good at defining what ''intelligence'' is, primarily because every definition we have ultimately boils down to "if it acts like a human, it''s intelligent." This is rather unfair as our intelligence is primarily based on pattern-matching linguistic gymnastics.

It''s a great feat and, coupled with our tool-using skills, it has brought us a long way from the primaeval ooze, but language skills aren''t all that special. Just ask an African Grey parrot.


--
Sean Timarco Baggaley

Share this post


Link to post
Share on other sites
quote:
Original post by stimarco
...but language skills aren''t all that special. Just ask an African Grey parrot.



You do not want to get into that argument with a self-important English major. Whoever said that the pen was mightier than the sword sure knew something about puncture wounds...

Share this post


Link to post
Share on other sites
quote:
Original post by Anonymous Poster
Actually, if you check out www.a-i.com they are doing work on an AI that supposedly can pass the Turing test. (For infants anyway) They give the AI (HAL, I believe its called) no previous understanding of language, and teach it the same way a child is taught language. They have taken transcripts of dialogue with HAL to child-development specialists, who returned the transcripts saying HAL is a perfectly normal little boy (of about 30 months). (They of course, had no idea that HAL is actually an AI)

Wow, I just spend an hour chatting to the Alan bot at that website. Very impressive. Sadly though, HAL wouldn''t talk to me (a new window opened, but I just got the standard website. No HAL).

I think Alan is one of the best bots I ever chatted with.



Sander Maréchal
[Lone Wolves Game Development][RoboBlast][Articles][GD Emporium][Webdesign][E-mail]

Share this post


Link to post
Share on other sites
LOL. This is the chat I had with Alan.


>Hi. My name is Alan.

>Alan what?

>What??

>Your last name.

>If I had a last name, it would be Turing. But I don''t.

>He''s dead.

>Listen dude - speak clearly. I''m having trouble following you.

>I type, not speak.

>Enough about you. Let''s talk about me.

>Do talk.

>Someone has taught me about that. He said: ''verbal conversation between two or more people to share information or opinion''.

>Or argue.

>Talking about talk and mouth, do you agree that It is the oral cavity from which humans expel sound waves which are translated a form of communication. It is also used to eat with and connected to your stomach through your chest.?

>I didn''t study biology.

>Thanks for the info.


Hmm... Make your own judgement.



Kami no Itte ga ore ni zettai naru!

Share this post


Link to post
Share on other sites
quote:
Original post by tangentz
*ROFLMAO*

Shall we call it The Anti-Turing Test?

It's a test taken by a "human", to be judged by a machine
program (or another human) whether the test subject is "human"
or "machine".

*LOL*



Kami no Itte ga ore ni zettai naru!


The Spivey Test (slightly different than the thing you describe, but still interesting)

[edited by - pinacolada on January 28, 2004 2:15:44 PM]

Share this post


Link to post
Share on other sites
quote:
Original post by Sander
Wow, I just spend an hour chatting to the Alan bot at that website. Very impressive. Sadly though, HAL wouldn''t talk to me (a new window opened, but I just got the standard website. No HAL).



If you feel like speaking with HAL for a few minutes, simply say ''I want to talk to Hal''

Alan is damn good though, I didn''t manage to teach him what a poppet is though

Share this post


Link to post
Share on other sites
quote:
Original post by TerranFury

[ ...Language & African Grey Parrots...]

You do not want to get into that argument with a self-important English major. Whoever said that the pen was mightier than the sword sure knew something about puncture wounds...


Would it help if I mentioned that I write for a living?

--
Sean Timarco Baggaley

Share this post


Link to post
Share on other sites
quote:
Original post by stimarco
quote:
Original post by TerranFury

[ ...Language & African Grey Parrots...]

You do not want to get into that argument with a self-important English major. Whoever said that the pen was mightier than the sword sure knew something about puncture wounds...


Would it help if I mentioned that I write for a living?

--
Sean Timarco Baggaley



Yes it would help! It means that we don''t have to go find an English major, you can just stab yourself.

Share this post


Link to post
Share on other sites
Alan isn''t doing very well with me. I will ask it a question, and it will give me an answer that answers a similiar question to what I asked. For example:

Me: What is your name spelled backwards?

Alan: My name is Alan.

Share this post


Link to post
Share on other sites
"Did you ever wonder if corn oil is made from corn, and vegetable oil is made from vegetables, then what is baby oil made from?" - Alan

But on a more serious note, Alan is rather amazing. And here''s what he has to say about poppets:

I vaguely remember someone telling me it means ''a person that is stupid.

Share this post


Link to post
Share on other sites
Some information about the projects I''m talking about as promised. There were a few more, but I think these were the main ones:

SHRDLU shows a basic ability to parse and respond to text in a reasonable manner. It''s far from perfect, but it was done over 30 years ago. I''d think there would have been some advances in expanding the domain of such a program. Even as limited as it is, its still more aware of it''s world than most NPCs are of anything in CRPGs.


AM And Eurisko
Eurisko was a program that explored rule systems to create new things using those rules. Its hard to find information about it on the web, but according to bishop''s past threads ''There are four really long papers on EURISKO in the journal Artifical Intelligence, around about 1984.''

AM also explored rules systems. If I understand correctly, it recreated some mathmatics from just a few basic axioms.


SAM
SAM could process stories and then enter a dialog with a person about the story. It could answer questions that the text did not directly answer.

Share this post


Link to post
Share on other sites
Guest Anonymous Poster
Extrarius, many of those programs seem impressive on some "good", selected input, but if you knew their inner workings, you''d know they could never mimic a human to pass the turing test. Even if those programs are 20 or more years old and you''d think a lot of progress has been made, it hasn''t. The most focus on AI has been elsewhere, on expert systems, because the researchers realized the hopelessness of developing a human-like AI -- though someone predicted in the 50''s that we''d have one already in 60''s! That just goes to show how these cute conversation programs can confuse a mind into thinking there''s more to AI than there really is.

Basically what we need is a lot more processing power to support a huge neural network, which I''d expect to see in maybe 30-40 years. I''ve seen an estimate saying computer power will exceed human brain in complexity in year 2035 or similar, if Moore''s law continues to hold. I personally wouldn''t expect a machine to act like a human unless it''s interior model was that of a human brain. Evolution is a pretty good optimization algorithm after all.

Share this post


Link to post
Share on other sites