Chatterbox Challenge Contest Underway

Chris Cowart writes "Chatbots from around the world are taking part in the fourth annual Chatterbox Challenge. Chatbots are computer programs designed to imitate human conversation, with the eventual aim of creating true virtual personalities and artificial intelligences. The Chatterbox Challenge runs from April 1 to April 30 and Internet users can talk to the competing chatbots through the competition web site." According to the organizer: "Chatbot names range from Aida to Zoe, and personalities vary from a fortune teller and a serial killer to a dragon and a horse!"

Comments

This all seems a bit pointless

[zyridium]

To me the idea of chatting is to talk about things that are happening...

For this to work it needs to happen within the context of some event or thing or understanding from outside the confines of a chatroom (eg talking about some football match, etc)...

Who cares if a bot can a/s/l it up and come on to you...

Re:This all seems a bit pointless

[zyridium]

What would happen if I came to mention, say, the Madrid bombing to the sales bot?

It is essential that a bot have broad experience (that is also dynamic) for you to think they are another person. If you want a bot such as the sales bot that is fine, but you can't expect someone to think they are another person.... They might do their job well and have a person-friendly interface through speech... but that is all

I fail to see

[plnrtrvlr]

I fail to see how fooling humans into thinking that they are having a conversation with another human, when it is really a chatbot, will do anything to produce artificial intelligence. It's an illusion, using technology, nothing more. Truly, our illusions are becoming more and more sophisticated as our technology grows, but artificial intelligence will require a deeper understanding than simple information processing and deduction from that information. Human intelligence, and the advancements that we have made with that intelligence, has been largely dependent on intuitive leaps: people who processed the information at hand (and quite often available to everyone) in a new and unique way. Learning to emulate the more standard thought processes of a day so that a conversation can be emulated is merely an exercise in sharpre usage of processing power and data storage, not a method of understanding the uniqueness of human thought.

Re:I fail to see

[zyridium]

Obviously if a bot had AI, it would be pretty convincing...

The major failing I can see in this method is that conversation could (and probably would be) purely pre-programmed, with no ability to learn new expressions or grammatical constructions...

Re:I fail to see

[MsGeek]

While I was studying natural language processing I read an interesting book in which Horst Hendriks-Jansen describes how, during a child's development, intelligent behaviour is built on a "scaffolding" of instinctive behaviour. For example, adults treat babies as intelligent, purposeful beings who are aware of their surroundings - we've all seen new parents interpreting baby's every burp and grimace as an attempt at conversation. In reality, most of a baby's actions are instinctive, and often unrelated to the people it's "interacting" with, but adults nevertheless feel a strong urge to respond and comment, keeping the false interaction going.

Hendriks-Jansen argues that this misunderstanding allows the child to "bootstrap" itself into genuine interactions, by learning from the intelligent responses to its semi-random behaviour.

Actually, the person who came up with this theory was actually Lev Vygotsky, an educator in 1930s Soviet Russia. (No "In Soviet Russia..." jokes, please.) Vygotsky was building on the research of Swiss educator Jean Piaget.

I have seen bots "evolve" in very interesting ways when resident on IRC channels. Of course, inevitably someone with an ecchi sense of humor comes along and gives the bot a filthy new vocabulary. ^_^

Will a carefully tended bot become sentient or even sapient? Doubtful. But they're fun to play with nonetheless.

Re:Nice

[JoshWurzel]

Consulting isn't enough. What really seems to be the problem is that the resource pages keep changing the way they format their data, so it becomes impossible for a chatbot to parse without monthly updates. This week I can ask my chatbot for the score in Celebrity Jeapordy (Sean Connery wins with a wager of SUCK IT TREBEK!) and it'll return "Sean Connery won with $uckittrebek".

Next week, when I ask the same question, it'll return "href a=blahblahblah won with a score of $%d3b" because the site it references has changed its format. I seem to notice this problem with weather programs too.

True AI - Fundamental Problem

[RazorX90]

Even if someone makes a highly advanced and believable bot, isn't the idea of trying to create AI with current programming methods fundament flawed. Although I'm not a programmer and don't know the technical terms, how can something ever truly emulate human behavior if it is limited by (insert highly complicated explanation of programming basics here). I just want to point out the program will always keep the AI contained and, by design, prevent learning beyond the programs initial design. If we want to reach AI, won't we have to come up with an entirely innovative and hybrid meathead instead of trying to get close to it with current programming techniques.

Re:True AI - Fundamental Problem

[ookabooka]

I believe so, though their programming technique may not be in c++, but rather, a neural net. Basically this emulates the human brain, but does it in 1's and 0's. The more often a certain "circuit" is used, the stronger the connections become, and the more likely it is to fire. Conversely, the weak ones are pruned away. By exposing it to a certain situation, and "rewarding" it when it makes a correct decesion, you are teaching the computer. For instance if the you have a program to play backgammon, you save the net when it wins, and revert to the old one when it looses. Kinda forcing it to evolve into a good backgammon engine. I think that the only thing limiting us from using neural nets to create AI is neural density. A neural-net backgammon game will beat its brute-force counterpart anyday. However it takes the entire neural net to do that, there are only a few thousand nodes. The human brain has billions of nodes, and trillions of connections. So i guess u could say we got AI already, its just 0.00000001% of what the human brain is capable of. Frankly im surprised the thing can play backgammon at all.

Re:True AI - Fundamental Problem

[Anonymous Coward]

There's not actually much of a consensus that neural network simulations tell us much about how the human mind works. Many psychologists suggest that the brain is just the "hardware level", and its structure has no more to do with how we think than a pentium processor has to do with the workings of a spreadsheet. There are /a lot/ of problems with explaining human intelligence in terms of trained neural networks. For exammple, how can we entertain two distinct thoughts at once, if all we have are patterns of association? It's also not the case that the kind of models of neural networks that are used bear much resemblance to real neural networks -- they're greatly simplified. And finally, neural networks are turing complete, so there's no reason why the brain couln't be used to implement a symbolic processing system at a higher level.

Ultimately neural networks just make statistical associations between inputs and outputs. There is no evidence that they really have the kind of magic ability to learn that their proponents claim they have.

A good overview of the shortcomings of neural network modelling can be found here: http://www.ucs.louisiana.edu/~isb9112/dept/phil341 /myths/myths.html

Chat between bots?

[Planky]

I'd like to see a discussion between to of these bots. Could be interesting to say the least: 1, Bob: Whats your name 2, Eva: My name is Eva, whats yours 3, Bob: Bob. ... goto 1

But what standard?

[autocracy]

Hold a conversation with me, or with some 1337 luser? I think I can actually write the second in 5 lines of obfuscated perl...

I think the limits of faking conversation are most defined by the limits of who you're talking with. Who is this supposed to impress anyway? At the least, I'd like to see something that fails miserably, but attempts to "learn." That'd be better than a smoke-and-mirrors anticipation of what somebody might try to say, or by constantly guiding the conversation to a pre-determined point.

Re:Nice

[ingenuus]

That's an interesting point:

- What is the storage restrictions on a chatbot in the competition?

- Is it allowed to google for a reasonable human response to your statements? ... i.e. basically using other past human responses as its own response.

- This is particularly interesting because, in this way (with a large enough db like the web), a chatbot could appear to be human, but we probably wouldn't consider this AI.

- If a chatbot reiterates something it downloads from the web, is that copyright infringement? In the case of humans, if we read something, I think we are generally allowed to quote it from memory.

Rod Speed, the ultimate Usenet chatbot

[oingoboingo]

Does anyone here remember from the early days of Fidonet on dial-up BBSs, and more recently Usenet, a particularly offensive person named Rod Speed? This guy used to (and still does) post at an incredibly prolific rate, with some of the most anti-social, deliberately offensive tripe I've ever read. The posts were always so similar that it was suggested for a long time that Rod Speed was actually a bot. In fact, some people created a Rod Speed chatbot, and I swear you can't tell the difference between its responses and those of the 'real' Rod Speed.

This guy even has his own FAQ..just go to Deja and search for "Rod Speed". He really blurs the line between chatbot and human. Rod....Rod...are you on Slashdot?!?!?

old skool

[man_ls]

Anyone remember this? It was the best AI program in the world.

My name is Doctor Sbaitso. I am here to help you.

SAY WHATEVER IS IN YOUR MIND FREELY. OUR CONVERSATIONS WILL BE KEPT IN STRICT CONFIDANCE. MEMORY CONTENTS WILL BE WIPED AFTER YOU LEAVE.

SO. TELL ME ABOUT YOUR PROBLEMS.

Unfortunately, Slashcode Lameness Filter seems dead-set on not allowing me to post the exerpt from it that was always said...

Any good chatbots for IRC?

[antdude]

I am currently using SeeBorg and I name him Homer (J. Simpsons). Most of the times he say stupid nonsense stuff, and he does learn to read IRC conversations. He records every IRC lines into a file, LINES.TXT, where he will randomly use phrases to say something. He will blurt out something one out of ten chances after someone else says something. If you call his name, then he will mostly reply back.

I used to use Alice, but the IRC script was very buggy and tended to hog CPU so I dropped her. :)

Few Thoughts

[Phoe6]

Where is Alanin the list? He is most Human like bot,I have ever talked with. The most interesting aspect is that he learns from the Conversation and does not need only a bot master to program. This is somewhat recursive right? You chat with Bot and Bot becomes intelligent with each conversation. Thats how it should go and thats how we may find a bot which actually knows many detailed facts because many people are taking with it and many persons are providing their Intelligence. In one of the previous posts, I saw someone mentioning Chatter Bots are illusions and nothing more that Data Processing.
Wait! Have a look around the Robotic Word itself. Are they all based on intutions and natural ways? NO! They are built using the detailed logic which every human follows and does not bother to look around with the Logic of his actions. Finer and logical ways of reasoning paves the way innards to AI.
Google is a good AI Software and not just data processing tool, because you get the result of what you are looking for ( in your mind) as the very first result. Chatbots,information processing will certainly help us improve the fineness data relationships and knowledge deductions.
Alice Vs ALAN here! And my own Phoe6 Here.
Njoy Chatting.

Re:Lame entries really...

[The Grassy Knoll]

The "fella" is Jim Wightman. Given what his "bot" does, the name of his company is rather "unfortunate" (says The Guardian) - it's Neverland Systems!

Anyway, no one is allowed to see his paedo-catcher bot working and he recently reneged on an agreed interview with The Guardian's Bad Science column (all this info is online at www.guardian.co.uk)

>Now there must be some sophistication behind that

Or trickery... That's a simpler explanation!

.