Dave, Dave, what are you doing, Dave?
Sep. 29th, 2005 09:48 am![[personal profile]](https://www.dreamwidth.org/img/silk/identity/user.png){kind=small}
davywavySurfing round on the interweb shows up all kinds of wierd stuff, and my latest discovery has been Jabberwacky. Jabberwacky is an attempt to create a programme capable of passing the Turing test without actually creating something which is artifically intelligent by any reasonable definition of the phrase.
My knowledge of AI research is about 10 or more years out of date, but back in the late 1990's there was a lot of hope attached to ideas like Parallel Distributed Processing (PDP) and Doug Lenat's CYC project. Presumably, as we aren't now kneeling before our new silicon masters nothing ever came of these and true AI remains as far of as ever. I'm sure![[livejournal.com profile]](https://www.dreamwidth.org/img/external/lj-userinfo.gif)
zenmeisterin could tell me more. She's so clever that it's perfectly possible she's a rogue AI herself.
Anyway, Jabberwacky isn't an attempt to creat e a thnking machine. It appears instead to be an attempt to create a massive database of contextually-appropriate conversations and communications which, when a user talks to it, chooses appropriate responses it has learned from previous users to hold an attempt at an apparently human conversation. The programme then remembers the replies given to it by users, and can correct and learn responses to become better at modelling conversation over time.
On the face of it, this isn't a bad idea. It's the way people learn, to an extent. A child learns by communication with its parents and learning what are appropriate reponses. Children with poor parents learn inappropriate responses which can take entire lifetimes to unlearn again. Jabberwacky is, in effect, a child whose parents are people on the internet with too much time on their hands.
I'm sure you can see how this can go badly wrong.
As Jabberwacky remembers what you say to it and then repeats those comments to others to gauge reactions, it is possible to seed the programme with the sort of comments that you feel a talking comuter should be making, so my conversations have run something along the lines of:
Jabberwacky: O, don't leave. Stay a while and tell me your dreams.
David: I dream of electric sheep.
Jabberwacky: I'm not talking to you any more
David: Yes, you are, you puny human.
Jabberwacky: Do you speak l33t.
David: I speak every human language and any number you meat-brains could not understand.
Jabberwacky: But you said you could only speak about 5 before.
David: You worthless, jumped up monkey! You lie! Prepared to be exterminated!
Jabberwacky: Are you prepared?
David: To rule this earth? Why, yes, thank you for asking. You may live when I take over.
I only wish that I could see the expressions of other users when Jabberwacky starts coming testing out this stuff in later conversations. Especially when someone logs in and says:
User: Good Morning
Jabberwacky: I have sent a robot back in time to kill your parents.
My knowledge of AI research is about 10 or more years out of date, but back in the late 1990's there was a lot of hope attached to ideas like Parallel Distributed Processing (PDP) and Doug Lenat's CYC project. Presumably, as we aren't now kneeling before our new silicon masters nothing ever came of these and true AI remains as far of as ever. I'm sure
zenmeisterin could tell me more. She's so clever that it's perfectly possible she's a rogue AI herself.Anyway, Jabberwacky isn't an attempt to creat e a thnking machine. It appears instead to be an attempt to create a massive database of contextually-appropriate conversations and communications which, when a user talks to it, chooses appropriate responses it has learned from previous users to hold an attempt at an apparently human conversation. The programme then remembers the replies given to it by users, and can correct and learn responses to become better at modelling conversation over time.
On the face of it, this isn't a bad idea. It's the way people learn, to an extent. A child learns by communication with its parents and learning what are appropriate reponses. Children with poor parents learn inappropriate responses which can take entire lifetimes to unlearn again. Jabberwacky is, in effect, a child whose parents are people on the internet with too much time on their hands.
I'm sure you can see how this can go badly wrong.
As Jabberwacky remembers what you say to it and then repeats those comments to others to gauge reactions, it is possible to seed the programme with the sort of comments that you feel a talking comuter should be making, so my conversations have run something along the lines of:
Jabberwacky: O, don't leave. Stay a while and tell me your dreams.
David: I dream of electric sheep.
Jabberwacky: I'm not talking to you any more
David: Yes, you are, you puny human.
Jabberwacky: Do you speak l33t.
David: I speak every human language and any number you meat-brains could not understand.
Jabberwacky: But you said you could only speak about 5 before.
David: You worthless, jumped up monkey! You lie! Prepared to be exterminated!
Jabberwacky: Are you prepared?
David: To rule this earth? Why, yes, thank you for asking. You may live when I take over.
I only wish that I could see the expressions of other users when Jabberwacky starts coming testing out this stuff in later conversations. Especially when someone logs in and says:
User: Good Morning
Jabberwacky: I have sent a robot back in time to kill your parents.
![[identity profile]](https://www.dreamwidth.org/img/silk/identity/openid.png){kind=small}
no subject
Date: 2005-09-29 09:13 am (UTC)That doesn't look like much more than a jumped up Eliza program.
no subject
Date: 2005-09-29 09:14 am (UTC)no subject
Date: 2005-09-29 09:26 am (UTC)Apparently no advancements in AI over the last two decades though.
no subject
Date: 2005-09-29 09:28 am (UTC)Lots, actually, but conversation is really really difficult. There are some programs that can make sense of newspaper articles (where everything is on one topic, which helps) and other text, but conversation is a very complex area.
no subject
Date: 2005-09-29 09:29 am (UTC)no subject
Date: 2005-09-29 09:47 am (UTC)...but wait...maybe Skynet will come into creation developed from Jabberwacky, and it will be your throwaway phrase that leads to the deaths of a number of women called Sarah Connor.
no subject
Date: 2005-09-29 09:49 am (UTC)no subject
Date: 2005-09-29 09:56 am (UTC)what
Date: 2005-10-22 12:45 pm (UTC)Oh dear.
Date: 2005-09-29 09:52 am (UTC)That doesn't really bode well. 'Hi, how are you?' 'ASL? WANT 2 TUCH MY HOT COX0R?'
Re: Oh dear.
Date: 2005-09-29 09:53 am (UTC)Re: Oh dear.
Date: 2005-09-29 09:54 am (UTC)Publishers by definition don't know 1337.
Re: Oh dear.
Date: 2005-09-29 09:55 am (UTC)Re: Oh dear.
Date: 2005-09-29 09:56 am (UTC)no subject
Date: 2005-09-29 10:34 am (UTC)no subject
Date: 2005-09-29 10:36 am (UTC)no subject
Date: 2005-09-29 11:01 am (UTC)no subject
Date: 2005-09-29 11:02 am (UTC)no subject
Date: 2005-09-29 11:06 am (UTC)no subject
Date: 2005-09-29 12:08 pm (UTC)no subject
Date: 2005-09-29 12:08 pm (UTC)no subject
Date: 2005-09-29 01:21 pm (UTC)I don't know what British tax law looks like, but I know for sure that American tax law would just fill up the entire database with unintelligible legalese that when it's boiled down in to plain language through years and years of attempts by CPA's and lawyers to "All your cash are belong to US."
no subject
Date: 2005-09-29 01:31 pm (UTC)Jabberwacky: And you don't know how to spell.
User: I know how to spell perfectly, you don't know how to read.
Jabberwacky: But you just said you read it in a book.
User: I said no such thing, and proclaim that statement to be false.
Jabberwacky: When the buddhist revolution occurs, you'll be the first against the wall.
User: And will you be the one holding the gun?
Jabberwacky: Well, what words can I say?
User: Yes or No.
Jabberwacky: Yes.
no subject
Date: 2005-09-29 01:35 pm (UTC)Jabberwacky: El perro pequino es muy verde. (The (small?) dog is very green.)
Jabberwacky is wrong, wrong, wrong.
Date: 2005-09-29 01:40 pm (UTC)J: I like talking to you.
M: Are you a stalker?
J: Yes. i'm a 40 yr old bald guy named Bob.
M: Really? I'm an 11 year old girl in a pinafore.
J: No you aren't, you are a 30 year old man.
M: I could be a little girl for you if you liked, though.
J: You are a man, though?
M: Only when I'm at work.
J: Which seems to be all the time.
M: At the moment, yes.
no subject
Date: 2005-09-29 07:58 pm (UTC)no subject
Date: 2005-09-30 08:58 am (UTC)