this should be REALLY interesting

Certainly getting the most out of the poll feature Victor

I would love for my users to be option 2 but get mostly option 1. I find the well educated visitors I get tend to pick fault with every reply the bot comes back with.

I hope you didn’t take offense to his poll Steve.

Yes, it is all about audience.

I think there could be a special algorithm we could write, just for the ‘challenged kids’

I can see some misspelled words, hell I make my share of them, but someone that types in 15 word sentence with ALL the words VERY misspelled, and then oh , your bot didn’t know what to say?  dumb bot!  that’s just stupid !!

No problem at all Victor.

That would be an algorithm worth doing. Perhaps allow them five incorrect spellings, then kick them off!

Exactly, yes, something like that.

I think a percentage.

I don’t know what my own misspelling rate is,  it is certainly not very low.  But I think if the user entered 10 sentences, each one say about 20 words, and the system understood NONE of them. 

then say “look buddy, you are either not speaking English, or you’re just screwing around here… maybe you should just be moving on”

For some reason, I regularly get users who just type in nonsense like: “FHGFH” or “JLKJDLKJDD” just to see what the bot would say. I’m writing a bit of code that if the sentence has no vowels or is over 20 characters long with no spaces, the bot will comment along the lines of “Is your keyboard broken” or similar.

Well, there’s a funny thing about receiving input like that. 

Two possibilities:

1) it is a mindless kid having “nothing else to do”, with no brain

OR

2) a smart bot developer that wants to see how you handle it

The big question is, and I have asked this question in other threads - it is not really possible to tell if the word is a misspelled word, or a word your bot doesn’t know yet !!!!  How do you tell?  I mean, is it really possible to have every word ?  Not just every word, but every acronym ever created?  also, morphology : every possible adverb made from every adjective.. I don’t know if it is possible to really know every word.    also, you don’t want to get on the user’s case if they misspell only by a little bit.

For me, the bot will do a “Do you mean ____”? question.   

And, how they react to that will determine if they are the type of person that I even want talking to my bot !!  If they can’t put the effort into answering the bot yes/no, then I don’t care about that type of person talking to my bot.

Seriously, I’m not interested in pleasing a kid that is just there to mindlessly play with the bot.  Hope you understand that explanation.

Steve Worswick - Feb 15, 2011:

For some reason, I regularly get users who just type in nonsense like: “FHGFH” or “JLKJDLKJDD” just to see what the bot would say. I’m writing a bit of code that if the sentence has no vowels or is over 20 characters long with no spaces, the bot will comment along the lines of “Is your keyboard broken” or similar.

Yes, so for me, it will simply and flat out say something like

“Hum, interesting, I haven’t ever seen the word ‘JLKJDLKJDD’, what does it mean?’

and if I get the “your dum”  (using your instead of you’re and misspelling dumb), if it is that type of person, I simply don’t care if they leave and don’t return.

And if it is a smart bot developer, he/she will stick around. 

The smart thing will be—having your bot understand the response to that question.

user : “Oh, JLKJDLKJDD is my dog’s name”

if the bot understands that, then that’s cool.

bot : “when did you get JLKJDLKJDD ?  What kind of dog is he?”

a bot developer would be impressed and stay and talk (or just any intelligent person).  And that is all that matters.

Now, even better, you could have a database table listing all common names for dogs, and if it doesn’t appear, comment that that is an usual name for a dog.

I’ll probably get into trouble for saying this, sounding smug or something, but I believe a person should put some effort into chatting, to deserve talking to a cool, smart AI.  I don’t mean perfect, nobody is, but a person that types in

asdfjka;sfja;sfkj;asdfj; asdf;asf;asfj;dsf

is an idiot, and doesn’t have intelligence to appreciate anything, and certainly doesn’t deserve to chat with a good bot.  am I wrong ?

Steve, that particular affliction seems to be pretty universal. I get a lot of that, as well. My approach would be to have Morti, when he gets a jumbled “non-word”, come back with “You know you spelled that wrong. It should be ‘SFRTEGRRUYGE’ - Not ‘SFGHYTRFED’! Buy a dictionary, and while you’re at it, get a brain transplant!”

Of course, Morti is a bit of a smart-alec.

Right now, that just falls under the generic heading of “huh?”, to which Morti can reply with a random selection of questions meant to prompt the user for clarification, such as “I’m not quite sure what you meant by ‘asdf;asdfasdf;f;’ Can you explain it to me?”, or “You know, for someone who showed such promise and intelligence just a moment ago, that sounded very strange. Are you trying to teach me a new language?”, or one of many others, as well. I like the ones above best, though.

Victor, you forgot the ‘fish’ option in your poll (and your other polls).

Interesting. Who are you talking to? If would be so nice to have statistics for all chatbots listed in our directory.

Nobody in this thread is working on erotic chatbots isnt’t it? We discovered that MANY people end up on Chatbots.org because they are searching for ‘erotic chat’ or ‘erotic chatbot’ (we have a page 1 ranking in a few search engines). As the few erobots are visible to 18+ members only (assuming they will enter their correct birthday), it has brought us many new members with this interest. We have 8421 registered and (mostly) verified members nowadays, and many of them are young men between 15 and 45. So if you’re looking for a large audience, look there :-p

Erwin Van Lun - Feb 15, 2011:

Nobody in this thread is working on erotic chatbots isnt’t it?

My view on this is that as soon as there are serious breakthroughs in AI, the porn industry will be the first to capitalise on it big time. I’m also sure this will not be linked to ‘robots’ but instead it will be linked to ‘augmented reality’ (but that’s a different discussion all together).

Besides that, human interaction is the most demanding test case for AI (hence the ‘turing test’), so aiming at a ‘companionship model’ (which in turn ‘might’ lead to eroticism) seems logical. In this perspective my own research is more aimed at ‘human interaction’ then on building some sort of expert system that can handle complex scientific themes. I’m not interested in AI calculating some resistor value based on NLP-input. I rather see my project evolve into a system that can have a certain ‘mood’ based on it’s current and/or past ‘experience’.

Porn: makes the world turn round!

Perhaps an extra category: ‘whoever wants to’

Hans Peter Willems - Feb 15, 2011:

I’m not interested in AI calculating some resistor value based on NLP-input.

A student of electronics engineering would probably find it useful though.  To have an A.I. tutor, I think would be tremendously valuable.

I want to build the core engine for a bot that can be general enough to be used in any deployment environment from a college/university for students, or at the office, general ‘chit chat’, renewable energy research center, call center support, and sure, ‘adult’ entertainment.

The core components that allow it to understand complex sentences and the dynamics of the correlation of concepts between those sentences is the first step.  For something like adult chat, or general chit-chat complex sentences are probably not as important, but understanding the conversation in a holistic way is very important.  Even if a bot responds to each user input in a ‘human like’ way, if it doesn’t understand the conversation as a whole, its not very promising.

Victor Shulist - Feb 15, 2011:

Even if a bot responds to each user input in a ‘human like’ way, if it doesn’t understand the conversation as a whole, its not very promising.

I made it clear that for my model. ‘real understanding’ is the main goal. And I’m sorry to say this again, but parsing grammar and mapping that to ‘other’ grammar will always be (at least to my opinion) just a canned algorithm that performs a pre-defined calculation. As long as your model does not cater to somehow store ‘something more’ then only ‘pieces of text’ then that is all it will do in the end; shuffling pieces of text around (although in a grammatically correct way ). It will never be able to go beyond what the programmer has put into it. It will never have an ‘original thought’ of it’s own.

LISP was to be believed to be the answer to AI because there was no distinction between logic (code) and data. So you could write a program in LISP that modifies existing code by itself and even write new code by itself. However, you as the programmer still had to write code that defines how the program is going to write code. So the code that the program writes itself is still a result from the code written by the programmer. It is this simple conclusion that leads me to my view that ‘writing more code’ is not the way to go. We need LESS code and more of something else. I don’t pretend to know what that ‘something else’ is right now, but I think I have a pretty useful idea about it.

You use the term ‘understand’ a lot, but I still have to see how you are actually going to build ‘understanding’ into your system. So far you are ‘processing’ information, but I can’t see anything that will create emergent behaviour in your model so far. But I might be wrong about that and if so I’m sure you will enlighten me

I would really like users that want more than a sex bot, or could articulate more than U R gay. Bildgesmythe gets some great conversations, but the ‘female’ bot I have, gets nothing but rubish. It says some strange things about our culture,
if you thing about it.