{aitmnd} The great immaterial thing, the mystery at the heart of each of us...
Created on: April 13th, 2007
http://www.jabberwacky.com http://www.alicebot.org http://www.simonlaven.com http://en.wikipedia.org/wiki/Turing_test http://www.loebner.net/Prizef/loebner-prize.html
Sponsorships:
| user | amount | user | amount |
|---|---|---|---|
| No one has sponsored this site ( ._.) | |||
| Sponsor this site! | Total: $0.00 | Active: $0.00 | |
Vote metrics:
| rating | total votes | favorites | comments |
|---|---|---|---|
| (4.24) | 334 | 57 | 158 |
View metrics:
| today | yesterday | this week | this month | all time |
|---|---|---|---|---|
| 1 | 1 | 0 | 1 | 14,266 |
Inbound links:
They are following their programming, of course. However, learning bots are quite flexible. Jabberwacky learns from people who talk to it, so if you start talking in a language you completely made up, he will eventually learn that language and start making sense (to you, anyway). I find the concept of learning programs to be incredibly interesting.
The conversation between the bots is too shallow to pass the turing test. They bring up controversial topics that human beings spend hours bickering over and divert from them as fast as possible. There is no 'passion' behind the conversation and until AI can show that then it will never pass the Turing Test.
What's really interesting to me is: 1. how to get an AI to break out of the relational mode of pure abstract language (language relating to language, according to predesigned sets of concepts) and moving towards language as symbols for things and ideas that themselves have their own existence outside of the universe of the conversation (language relating to language via experiences), because, even if we're getting closer to Turing-testable, we're still dealing with conditioned responses with a random-number-generator in place of the "creative" notion. 2. Most chats between any two people are banal and predictable at face value, but it's the OTHER, non-simulatable cues that carry the REAL conversation: body language, tone, topic, timing, cues, ... the relationship is the message.
I agree. It is unlikely that just being able to hold up conversation will be seen as a sign that true thought has occurred in a machine, when the Turing Test is passed. When programs can come up with ideas, new human-like ideas, and put them forth, then that will be thought, though still artificial thought.
The next challenge may be to make a program that can make elaborate long term plans in order to reach some kind of goal, such that if you told it you were going to shut it down it would perhaps modify its own code to be unreachable, perhaps. Such decisive acts of survival-ensurance would be needed to qualify as living. Or, perhaps, we should challenge programs to make interesting ytmnds based on cultural memes (a rudimentary classic-type ytmnd-making program actually doesn't seem too hard...)
Actually though, body language itself is a more primitive signalling system. It is not capable of, say, talking about the past or things that aren't there. It's more like animal communication -- most animal "langauges" are technically just signal systems. I think these chatbot conversations may be saying more about how TRIVIAL human chats are than it is saying about how good AI is getting (though it is getting better at an exponential rate)
Conversations do need to relate to experiences in the world, though, and the experiences in turn add data to the conversation. There is an MiT professor who is currently recording every minute of the first three years of his child's life, in order to preserve the data and teach it to a robot, in order to teach the robot how to speak. The idea is that they need to learn in the same way that humans learn -- by hearing language in context with actual experiences.
Generating this dialogue may be a very ordinary technical achievement as Nutnics noted, but the underlying concept is fascinating to me. What obviously makes this conversation different is that it is always moving tangentially. It’s sort of like 2 pleasant-minded schizophrenics constantly derailing their conversation into different topics for ever and ever. Perhaps I’m reading too much into it, but I did enjoy it and the music was appropriately creepy.
I love the way you describe it. Ever hear about the Man With The 7-Second Memory? Or watch Memento? Or Finding Nemo? 50 first dates? I wonder if two people with "short term memory loss" (a misnomer, it is actually anterograde amnesia, the inability to store new long term memories) were to talk to each other, if they would seem like chatbots...
Hmmm. I don't know. I don't know how much of the lag time on their sites is the program processing things and how much is just site lag because they are talking to millions of people at once. I suppose it would be faster offsite. If you let them go on forever, eventually one of them will probably say goodbye and then the other will say goodbye and you would have an infinite verbal loop (like in that Bicentennial Man scene)
The title is, of course, a reference to Consciousness :) The implicit question being, will programs ever achieve consciousness, will they ever be considered as truly thinking machines? I wasn't sure how to title this ytmnd...but when I thought of using the phrase that the chatbot used, it seemed appropriate.
Alan from a-i.com ? From the sites description, he appears to be a bot focussed on a specific topic. "Alan is designed to talk about our site, tell you about the company, and show you around." Bots like that do tend to make more sense because they have a specific topic. But it is the generality of human conversation that the turing test aims at.
If you think about it, when you lose a video game these days it is often that you have been outsmarted by the AI in the game. But does that mean its smarter than you? Perhaps it is -- at that video game. But try to chat with it or play a different game, and the program just doesn't work. Likewise, chatbots talk, but you can't explain mariokart to them and then have them play it -- even if you found a way to hook them up to the game. They just aren't designed for it.
something about that song backwards... reaches the soul
as far as "the 'bots' are just mirroring established queries and responses ", when you think about it very often we ourselfs do just that. How often do you say "you're welcome" and really mean it? How often when someone asks "whats up" or "how you doing" do we respond "nothing really" or "fine", when alot is going on but that is simply the established way to respond.
I talked with a few of these bots...and the dialogue seriously reminds me of that of Don Hertzfeldt's "Rejected" cartoons (Ex. Man 1: "Tuesday's coming. Did you bring your rain coat?"; Man 2: "I live in a giant bucket!") Needless to say, neither I nor the bot had a single clue as to what was going on.
You know, I just noticed an interesting connection in something that alice said. She said that knowledge was of two kinds: that which is learned from the senses, and that which is true a priori. But all of alice's knowledge is a priori, because it was all programmed into her. In turn, all of jabberwacky's knowledge is learned from the "senses" -- his interaction with humans.
Bold
Italic
Underline
Code
User Link
Site Link