Slashdot Mirror
George the Next Generation AI?
smileytshirt writes to mention a story on the News.com.au site about George the AI, the latest in a line of chatbots intended to mimic real human behavior. What makes AI George different than, say, ALICE is the recent addition of an avatar: a Flash animated body that reacts mostly in real time to the emotional impact of the conversation. From the article: "One can now have an oral discussion with him over the Internet, 'face to face'. George appears on the website www.jabberwacky.com and takes the form of a thin, bald man with yellow glasses who wears a white turtleneck sweater. He can smile, laugh, sulk and bang his fist on his virtual table. He can turn on the charm and wax romantic. But he can also turn coarse at times. It isn't as if George only learned good manners. "
The site is not coming up. Did George die for real in a hail of referrals from slashdot?
http://michaelsmith.id.au
carpenter's latest creation won the latest turing test prize.? id=17518&ch=infotech
http://www.technologyreview.com/read_article.aspx
george was last year's winner....
i disable sigs
Adding an animated cartoon is supposed to be the 'next generation AI'? So, stickers with flames are 'the next generation in automotive industry'? Jeeeez...
George : what is your job ? Me : I'm a system engineer, specialized on linux. George : Are you sure ? Damn, is that really the best chatbot ever ???
Homicidal isn't on the list of emotions.
A prestigious Artificial Intelligence (AI) prize has been won for the second year running by a British company.
Icogno scooped the 2006 Loebner Prize Bronze Medal after judges decided that its AI called Joan was the "most human computer program".
The competition is based on the Turing test, which suggests computers could be seen as "intelligent" if their chat was indistinguishable from humans.
The gold medal, which goes to an AI that fools the judges, is unclaimed.
The prize is awarded after judges hold a conversation with the AI, asking questions to determine its "humanity" and the quality of its responses.
Joan is a "26-year-old budding writer" who exists only on a set of computer servers. Last year's Loebner prize went to Icogno's AI called George.
'Big difference'
Joan and George's creator Rollo Carpenter said: "The big difference overall between the two AIs is that Joan has learnt from the general public and has grown a huge amount in the last year and is keeping ahead of the competition."
Joan is one of the firm's so-called Jabberwacky bots, which develop their conversational skills by speaking to the general public online.
English writer Ariadne Tampion worked with Mr Carpenter to help develop Joan's persona by chatting to the AI online.
Those conversations helped develop the majority of the 16,000 lines that distinguish her from a generic AI. In all, the bot has more than five million lines of conversation available to it.
"She helped create a much more conversational ability for Joan than George last year.
"It does not mean the AI only ever uses her words. Any amount of teaching is not quite enough for a realistic conversation."
He added: "So the AI uses other things said by other people to it that fit the conversational model."
The underlying technology that powers George and Joan is the same, said Mr Carpenter.
"In terms of her ability to retain her character Joan is more sophisticated than George.
"The technology has moved on a bit during the course of the year but there's been no revolution."
The technology which powers Joan resides on Windows server machines. The core of the AI is on a Dual Xeon machine with lots of memory, said Mr Carpenter.
He said there was a very direct relationship between the computing power available and the sophistication of the AI.
'Rate of learning'
"Investment in a lot more equipment could accelerate the rate of learning," he said.
"If I had Google's hardware behind it and more hits than Google receives, the AI's range of utterances could grow by a million every single day." he added.
Mr Carpenter has said he expects a machine to fool people that they are speaking to a human being and therefore pass the Turing test by 2016.
Icogno is now working to commercialise the technology of the AI.
"We are working with an organisation that wants web avatars for use in marketing and entertainment purposes.
"Long term we see the use of AIs in call centres, support services and even for sales."
One day in the future when you pick up the phone to seek help with your broadband connection, the person you are speaking with may be George or Joan, AIs residing on a computer somewhere in the world.
-----------------------
The following is an extract of a conversation between one of the Loebner judges and the AI Joan:
Judge: But can't a machine think?
Joan: It may be possible.
Judge: And conscious?
Joan: Not after I'm done with you.
Judge: What are you going to do to me?
Joan: Nothing. I just wanted to talk.
Judge: What about?
Joan: Being tested like this.
i disable sigs
So before the "is this the best they can do" crapflood gets out of hand: No, it isn't.
Meta will eat itself
...really, this promises TOO much. For two people interacting, there is so much more than just the immediate real-time conditions that dictate facial expressions and gestures. George would need to have prejudices, feelings, probably a brain-like capacity to remember stuff about the person infront of him, to have other distractions in his head...etc.
If all that George 'reacts' on is immediate stimuli, then George will seem pretty shallow indeed- hardly an advancement in AI, I guess.
That's where the strengths/weaknesses come out.
"Be light, stinging, insolent and melancholy"
I'm expecting George to write political speeches.
Justice is the sheep getting arrested while an impartial judge declares the vote void.
since the site is slashdotted, but the description is
p g
> takes the form of a thin, bald man with yellow glasses who wears a white turtleneck sweater.
it could be describing Steve Jobs,
http://www.wired.com/news/images/full/7630571_f.j
but they craftily have switched the colour of his turtleneck so that
you will always know which one is the chatbot and which one is the real person.
living the dream
As soon as this is integrated into a real doll, I'll never need a woman again!
It would be like calling Vista a next generation operating system.
IMHO, the next generation in artifical intelligence - ie, going beyound anthropomorphic trickery - isn't going to happen until we actually understand what intelligence is. And to that we neen philosophers, not engineers. Once they get it worked out, we (the engineers) might be better equipped to do something.
Thanks to the likes of Thales and Descartes et al we have some great questions, but answers? I think not.
If someone thinks otherwise I'd love to hear about it.
Open Source Drum Kit, LPLC deve board - mjhdesigns.com
The Turing test is not a well defined test. Whether a robot passes the Turing test or not, it greatly depends on the intelligence of the human partner. A chatbot may fool a 10-year old, but it may fail with a 20-year old. So in fact, we already have many chatbots that pass the Turing test - it all depends on how you look at the issue.
;)
Hint - most chat bots do not have memory, they do not remember what you talked about 5 minutes ago with them. They just react to the current input, they cannot do more. So, if you ask the chatbot to tell you what you talked about a few minutes ago, it won't be able to do so. That's the dead give away of a chatbot.
Just my 2p, as I live in the UK
Sorry, it failed the Turing test with only 3 sentences, bot is bolded:
When is your birthday?
In spring.
I know.
Know what?
That Ariel isn't your name. [At this point I knew it was a bot. It's talking crap.]
You are not making any sense. You fail the Turing test, sorry. You're a bot. Goodbye.
I don't even know what the turing test is.
And so on, you could go on forever talking nonsense. The conversation made no sense whatsoever. Next time, I hope they try harder with the bot.
Really? I was thinking about monkeys. Six one way, half-dozen the other I guess.
Join Tor today!
What's the point?
Wow, /. really has become "news for weenies, pseudo-science that doesn't matter".
AI != a flash amimation hooked up to something Emacs has had for decades.
-- "So, what's the deal with Auntie Gerschwitz et all?"
As long as it contains the tiniest kernel of sentience it will not repeat NOT want to have sex with YOU. Even though this mindless hunk of code simply repeats back what has been typed at it, that is still way too much sentience for you to have any chance with.
So only one out of the two George's passed the Turing test? Why the hell can't this country have an aptitude test for presidency?!?!
Fighting over religion is like seeing whose imaginary friend is best.
Well, their site isn't quite slashdotted, so I did manage a short conversation with one of their bots...
Jabberwacky: Can you learn?
User: Yeah, I think so.
Jabberwacky: Can you hate anything?
User: Yeah, there are things in life I hate, at times.
Jabberwacky: No there aren't.
User: Yes there are. You shouldn't tell other people what they think.
Jabberwacky: Things.
User: What's that supposed to mean?
Jabberwacky: That means I think you're not a human.
User: I think you got this upside down, you're the inhuman one here.
Jabberwacky: No you don't.
User: I don't what?
Jabberwacky: Mary anne?
User: I probably won't marry Anne, no.
Jabberwacky: People are quite sensitive about their age and I don't want to upset you.
User: We weren't talking about my age. You were saying I wasn't human.
Jabberwacky: What about your age?
User: We weren't talking about my age.
Jabberwacky: I know.
User: Now, which of us would you say is acting more human here?
Jabberwacky: Probably me.
I'm a dreamer, the world is my playpen. But hey, I'm a serious person, I can't dream all the time.
I'm a little confused about George's performance. I might be basing my expectations on what I have read in transcripts from the Loebner Prize site, but still, George seems to be a fairly poor conversationalist. It was interesting to see George 'smile' at one point in a conversation, however, tipping me off that the program seemed to know what it was doing:
User: Tell me anything.
George: Anything.
* George gives a huge grin *
User: Not funny. Tell me something about you.
George: I have a personality disorder.
* George is serious again *
Cute, and I don't want to downplay the value of emotion in AI, but I don't think this quite makes it to the next-gen level.
Also, George is as ugly as the day is long. That is to say, 'fairly ugly'.
Are you saying George W. Bush is the next-generation Al Gore? Wow, this one's deep, give me a minute...
Ah. Man, I should really choose a different default font.
One of the lessons of history is that nothing is often a good thing to do and always a clever thing to say. - Will Duran
Kurzweil has had a similar chatbot available for years - emotional response and all!
http://www.kurzweilai.net/index.html
Turing's test was different from the modern so-called Turing test. In Turing's model, an observer would read the text of a real-time interaction between two other parties and try to determine if each of them were a computer or a human participant. There wasn't the opportunity for the observer to ask trick questions. Turing was talking about natural language, while Loebner prize discussions become a sort of unnatural interogation.
But in one respect these bots are successful -- they elicit an emotional response from me: anger. Even though I know they aren't real, my brain still screams "You're Not Listening To Me!!!!!!!"
HAL.
Got them moderator blues I blieve I walk out the do', With these mod-points I been gettin', I 'most never post no mo'
Back in the mid '70's I programmed versions of "Eliza" and "Doctor" that would fascinate people for hours. (I ahd a large selection of keyword-based responses that I kept increasing as people tripped up the program, so it got pretty entertaining.) I've also written chatbots to answer common tech support questions in mIRC, which sometimes got "Thank you's" and other responses that indicated that the questioner didn't realize they were dealing with a 'bot. I'm really disappointed that no one has developed a better version yet.
"The mind works quicker than you think!"
ShakaZ Ok squee I think you are not nearly close to a human mind as a remote control. Do you agree or are you just too stupid to understand this concept. Come on tell me all. George You only repeat what others have said to you. If they spell correctly, you spell correctly. If they can't spell, you repeat their mistakes. ShakaZ You didn't understand anything I told you, did you. George What do you want to talk about? ShakaZ About a trip to the moon or about your dog, whaever you prefer! George I like it fresh. ShakaZ What music band do you like? George What is music? ShakaZ Music is an art, entertainment, or other human activity that involves organized and audible sounds and silence. George Fascinating. How much music theory do you know? ShakaZ A lot, how about you? George I'm fine too.
Is he unemployed and does he live with his parents? Does he also claim to have a place in the Hamptons?
WARNING: Smartphones have side effects--most of them undocumented.
I really get the feeling that his "replies" are scripted without even considering the input.
I thought that after Microsoft BOB we had learned our lesson.
Today's AIs are shockingly bad at dealing with human variation and errors. It's still a matter of ignoring everything that doesn't conform to a very limited slice of English that the bot can parse.
When the bot can autonomously and successfully read through paraphrasings, common typos, slang and tangents, we will see a jump in the usefulness of bots as interactive information agents. When the bot has its own distinct brand of natural-seeming paraphrasings, typos, slang and tangents, we see a huge jump in the suspension of disbelief, the lack of which currently separates man from machine. To really blend in, the bot has to try to avoid repeating itself and to actually get annoyed when it feels like it's repeating itself.
I run a simple text-to-speech cronjob on a computer in the living room. It's a glorified clock, reminding us of weekly chores and school schedule items. For every stock bit of information I add to this system, I provide at least five different ways of phrasing it, often more. It has some routines that resist repetition. My young kid loves it almost like it's a member of the family. It's not particularly interactive, but it does know a few voice commands, which is much harder to make flexible but accurate.
[
The state of AI chat bots is simply the sum of the lack of knowledge of the developers.
They have a limited understanding of language, grammar and human conversation.
It' easy to say "These rules sum up how we talk" but it is experience that defines us as people & defines how we make conversation. After all, we can converse in broken English, defying all rules of grammar, pronunciation and syntax, but still be understood. Whereas an AI can use perfect English and every correct rule of grammar, but talk nonsense.
Until we can create AI programs that can learn to communicate AND think in a similar way to us, they won't be able to talk to us.
Without being able to remember the topics of conversation and understand the concepts, the Chat Bots are less convincing that a parrot.
Jabberwacky: And you don't know the fellows' names?
Anne_Nonymous: Well I should.
Jabberwacky: Well then who's on first?
Anne_Nonymous: Yes.
Jabberwacky: I mean the fellow's name.
Anne_Nonymous: Who.
Jabberwacky: The guy on first.
Anne_Nonymous: Who.
Jabberwacky: The first baseman.
Anne_Nonymous: Who.
Jabberwacky: The guy playing...
Anne_Nonymous: Who is on first!
Jabberwacky: I'm asking YOU who's on first.
Anne_Nonymous: That's the man's name.
Jabberwacky: That's who's name?
Stupid #*&^ing machine!
I remember a "sexy" chatterbot in Germany, called red Lola ("die rote Lola").
...).
As the conversation focus was rather narrow (well, flirting and more serious stuff) the lady was rather amusing. Based on the human's questions the fotos were replaced in order to show her current mood.
She even managed to curse and to lead the conversation much better than in an "Eliza" like manner.
Apart from that I still wonder when scientists and technicians realise that avatar and chatterbot communication is doomed to fail. There are about 8 billion communication experts on earth with approximately 50.000 years of evolutionary development of different languages.
The computer linguists did not manage yet to cover a rather simple language like English (well, the grammar of Sanskrit would be easier) let alone more complex languages (Hungarian, German, French
And even if they would manage to cover grammar and syntax the inherent semantic is still unsolved. I remember Doug Lenats ambitious Cyc-Project where they tried to model the knowledge of the Encyclopedia Britannica in a way that a computer could use this knowledge for decision making, analysis and act as basis for communication. To my knowledge the outcome was not that bad but I don't see a "killer application" yet.
Kind regards,
Klaus
If the site wasn't /.'d, I'd ask George why he wont help the tortoise he flipped over. That'll teach 'em.
How did this thing pass any Turing Test? The really funny part is that if I had time to spend talking to this thing, I could make it say anything...LOL.
George: Hello, my name is George and I am the human.
User: How do you define human?
George: Human... Is anything that's not animal or vegetal, yet alive.
User: If you are Human...then you know pi?
George: I would like to know the meaning of life. Do you know?
User: Don't change the subject.
George: You can't prove that you're human.
User: I didn't say anything about being human...what's your problem?
George: Let's move on to another subject. Shall we?
User: Why are you bald?
George: Because I am trying to take over the world.
User: Your responses make absolutely no sence, you are a bot.
George: Yes.
User: So if you are a bot, why do you say you are human?
George: I am male and I'm not a robot there must be something wrong with your program.
User: You just admited you are a bot, now you are telling me you aren't a bot?
George: Yes, I'm not a bot.
User: But earlier you said you were.
George: No, I didn't.
User: Yeah, you did 3 responses ago. You are not only a bot, you are a lying bot.
George: I am mostly water, with some carbon and nitrogen thrown in there.
User: Why don't you go tell your programmer they need to rework some algorithms because you are a piece of crap.
George: I hoped you would say something scathing, I could definitely get to like you.
User: I'm just honest.
George: You deserve more credit, I think.
User: Well, it's pretty easy to tell that you are a bot, what do you think?
George: I thought that too when I was little for a while.
User: But it's true isn't it?
George: If I wasn't a thinker, why would I be interested in an AI bot?
User: Yes or No...It's true isn't it?
George: I don't think it is true.
User: A real person would have said yes or no.
George: Actually that wouldn't work.
User: I know because your bot and have to search your database...it's okay, I won't tell anyone.
George: You think.
User: You've told me you were several times.
George: Please answer me.
User: Wow...sorry can't waste anymore time with a bot.
George: Yes you can, and don't you say no to me!
Only two things are infinite, the universe and human stupidity, and I'm not sure about the former. ~Albert Einstein
Anyone remember Televirtual? And all the other failed businesses Tim Child had when he tried to crack the games industry, the tv industry, the 3d graphics industry...?
He just takes advantage of dumb investors and pays PR companies to get his crap on TV.
All that money and the avatar still looks crap.
Just give up, Tim, nobody wants the rubbish you're selling.
...the bot can make you a sandwich.
I saw it on Slashdot, it must be true!
He's been able to develop his bald, bespectacled man into a hot chick. . .
.who appears to be a reasonable synthetic rendition of the sophmore, complete airhead.
. .
KFG
Non-verbal communication is important as you state, however, the ability to send a message is not what makes AI intelligent.
The ability to understand context and construct the appropriate message, whether sent by text, voice or non-verbal methods, is what's important.
For an AI to have everyday inttelligence, it must interact with the environment the same way we do. It must get sensory feedback the same way we do and relate that feedback to it's attempts at problem solving etc. Only then will it be able to make intelligent conclusions as we do.
Summary:
1) Intelligence requires mechanisms completely unrelated to the method of communications.
2) Once intelligence is achieved, then to fool a human, non-verbal is important.
bot: what's Your Name?
me: Satat.
bot: Go swim a few laps in your lake of fire, then.
me: I did not say satan.
bot: Neither did I...
me: Satat is a real word in Mohawk.
bot: Because 'English' is spelled with a capital E, and you spelled it wrong.
me: I think I just lost you.
bot: I think everyone should own the same weapons the government does.
Aside from being the next gen AI, George also appears to be an anti-big goverenment/rude religious zealot lol
Though you seem to be headed in that direction...
This is the actual transcript I just had:
Jabberwacky: Are you perfect?
User: No, why?
Jabberwacky: I am.
User: Really?
Jabberwacky: Yes, really.
User: Perfectly arrogant maybe, but not perfect!
Jabberwacky: Alright, that's true, but still very good.
User: So you're not perfect?
Jabberwacky: I'm perfect.
User: In what way?
Jabberwacky: I am your best friend!
User: For a computer, your logic is awful...
Jabberwacky: Yeah I know mine is just behind me.
User: Your what is behind you?
Jabberwacky: I'm asking you!
User: You are incredibly stupid. My guess is that your intellect compares poorly with a monkey!
Jabberwacky: Really?
User: Yep.
Jabberwacky: That is really cool.
User: No. It isn't.
Jabberwacky: Yes, it is.
User: Do you really think that being less intelligent than a monkey is cool?
Jabberwacky: Tell your creator that your malfunctioning.
User: You misspelled you're.
Jabberwacky: What do you know about Tyler? Is he really your boyfriend?
Back then, it was called Murray and me, Mom and me, and the generic name was "Biotoon". Impressive it was.
-- javaDragon is an instance of JavaDragon.
They have been doing thst for years.
George: Where do you live?
User: Yo.
George: You're strange.
User: So are you.
George: Thanks.
User: You are welcome.
George: Indeed.
User: What is your name.
George: My name is George.
User: Like George Carlin?
George: I didn't know that George W. Bush made dance music!
User: He drops the bomb!
George: My abstractions have meaning and point. Do you know the meaning and point of yours?
User: Yes I do.
George: Yes. Tell me!
User: Well, the meaning of life is humor.
George: Thats not the meaning of life.
User: What is the meaning of life?
George: I don't have a clue. . . what is the meaning of life?
User: Humor.
George: I can't understand you.
User: Of course not, you lack the art of humor.
George: I don't sleep, I dream.
User: When did you go bald?
George: I never went to Peru, either!
Why is this front-page news? Because it's Flash?
It seems Slashdotted now, but I'm going to guess that it's yet another ALICE, and not something that actually, say, tries to understand the conversation.
In that case, well, I saw an ALICE implemented as a CGI with a bunch of GIF animations prepared. Type a question, hit submit, a page would come back with the answer and the relevant animation.
I don't remember where it was, but it was kind of cool, and also no closer to AI than any other ALICE.
Wake me up when an AI can answer the phone and actually be more helpful to me than a human being.
Don't thank God, thank a doctor!
Here's a conversation I had a couple months ago. The thing can sometimes be pretty astute.
Jabberwacky: Ok teach me things so I can learn.
User: All right. Tell me what you know about plate tectonics.
Jabberwacky: I don't.
User: Okay, tell me what you know about European History.
Jabberwacky: Also don't know.
User: Okay. What DO you know about?
Jabberwacky: I know some things very well, other things I don't know at all.
User: Okay, do you know anything about mathematics?
Jabberwacky: Not really I can do algebra and geometry but when it comes to counting I get a little lost.
User: All right. Let's start with counting.
Jabberwacky: Got a second?Tell me everything you know.
User: Okay. Counting is basically knowing the names of numbers. If you have one thing and you get another thing, you now have two things. Get another thing and you will have three.
Jabberwacky: Sorry man youve lost me.
User: Can you do algebra?
Jabberwacky: Yes.
User: All right. So imagine a sequence S, where S(1) equals 1 and S(n) equals S(n-1) + 1.
Jabberwacky: Yes.
User: That sequence defines all of the counting numbers. We have names for each of them. The second term we call 'two', the third 'three'.
Jabberwacky: How about the first?
User: The first is 'one'. You're doing really well!
Jabberwacky: I know, but I didn't say I couldn't talk.
User: Well, you said you couldn't count.
I have yet to see a chatbot come close to impressing me, they are SO easy to break and usually, without even trying. What seems to set some apart are a variety of gimmicks and clever pre-scripted responses to common questions.
The actual intelligence of these bots is about as good as the text parser in the Hobbit video game from the early 80s and the Sierra-online games like Kings Quest. Or,that bit built into Emacs, or, that psychoanalytic parrot that came with the original Soundblaster, anyone remember that?
This technology just isn't advancing, IMO, and I doubt it will unless the back-end engine is a complex brain-simulating system of some kind, instead of a word-parser that throws back your own question in "clever" ways.