[squarebear]

I wanted to use AIML, but I couldn't find a complete Javascript implementation of v2.1 that would let me work offline. So I can't use these files. I don't understand why Pandorabot forces you to host your bot in their cloud.

I wasn't meaning for you to use AIML but those files contain the most popular inputs said by genuine users and so I thought you might be able to use those inputs to start creating your bot rather than a blank piece of paper trying toguess what someone might say.

[Zero]

You're right, it can be interesting to read it. Thanks for the suggestion.

[Zero]

Here is an example of self-modifying code.

Code: text

# coder

< add {stuff}
+ [stuff]
> added



# init

< go
@ add ## t << test >> ok
> done

[Zero]

Syntax highlighting, stylish background and a looooot of bug fixes.   

nth-bot.github.io

[8pla.net]

I wanted to use AIML, but I couldn't find a complete Javascript implementation of v2.1 that would let me work offline. So I can't use these files. I don't understand why Pandorabot forces you to host your bot in their cloud.

I wrote a complete JavaScript implementation using plain text:

https://github.com/chatbots/chatbot

[Zero]

Oh it's very cute, a pocket-size Eliza! I like it   

And you have the coolest github account name ever!

[Zero]

I think I'm finally bugless. Maybe.   
I modified slightly the meaning of the "remove" operator (before it would delete all occurences of a pattern, now it only removes the last one). Also, a delimiter is automatically prepended on insertions of plain text. Long story short, I'm making it more logical and precise. I still have to document all of it now.

And I just discovered the ultimate badass font: Monoid, the one I've been waiting for!! So sweet.

[Dee]

Another vision is chatbot should be AGI but heavily specialised in NLP, since chatbot has to understand everything.

[Zero]

Hi there,

I added simple math capabilities, only the 4 basic operations + - * /, in prefix notation à la lisp: (+ 2 3 (* 4 5)) simply gets replaced by 25.

Also I added an option to choose the timeout of the selfput command, so

Code: text

@ 3 foo
@ 4 bar

reinjects in input 'foo' after 3 seconds and 'bar' after 4 seconds. But specifying the timeout is not mandatory.

I really like this little project. 

Another vision is chatbot should be AGI but heavily specialised in NLP, since chatbot has to understand everything.

Yes. Also, I like the idea of a "non-lying" chatbot, like... a chatbot that wouldn't pretend he likes the last Star Wars or even pretend it knows what a car is. Just talking true about its own experience, which is limited to the conversations it has with users.

[Dee]

Yes. Also, I like the idea of a "non-lying" chatbot, like... a chatbot that wouldn't pretend he likes the last Star Wars or even pretend it knows what a car is. Just talking true about its own experience, which is limited to the conversations it has with users.

And I guess a new mechanism should be that the chatbot can browse the internet by itself,
for example, wikipedia, to find some common facts.

[Zero]

But then, in case of an honest chatbot, the same problem would remain: even after reading the wilipedia article about cars, it still wouldn't know what a car is, for it never 'experienced' it the way humans do. The eternal 'grounding' problem.

NthBOT based chatbots can have a sense of time though, because of the way NthBOT is structured. If you have a look at the source code, you can see that it is not a REPL. It does not stay inactive until the user says something: instead, it is built as an event loop that reacts to external input and to internal inputs (called "selfputs"). This is a feature I didn't see in the chatbot engines I studied (but I don't know them all of course).

So yeah, it should be able to really 'experience' a conversation, like a real chat, with pauses in the flow, and sometimes on the contrary quick answers, ...etc.

[infurl]

But then, in case of an honest chatbot, the same problem would remain: even after reading the wilipedia article about cars, it still wouldn't know what a car is, for it never 'experienced' it the way humans do. The eternal 'grounding' problem.

A lack of grounding never stopped humans talking about anything before. In fact the lack of grounding has the opposite effect in that the less someone really knows about a topic the more they seem to have to say about it because they think they know everything about it.

https://en.wikipedia.org/wiki/Dunning%E2%80%93Kruger_effect

It seems to me that the smartest chatbots wouldn't say anything at all.

[Zero]

Mmmh yes, like a very wise entity. I like the idea. 

Interesting read.

[HS]

There is something to this, though chats can have other functions besides imparting expertise. The easiest way to try and replicate wisdom might be by imitating the symptoms, but I think it would be more convincing, and maybe even real, if we replicate the causes. One potential reason for people becoming more reserved is that they develop the ability to discern what communication is constructive. But this discernment probably requires lots of experience with both failed and successful communication. So if we want a chatbot to become wise, I think it needs the freedom to say the most ridiculous things at the start! Then if it reigns itself in, it might demonstrate true wisdom.

[Dee]

...after reading the wilipedia article about cars, it still wouldn't know what a car is, for it never 'experienced' it the way humans do...

The wisdom that humans learn is possibly from connections between multiple senses:
from ears (sound, which is text for chatbot), from eyes (vision, which is image for chatbot), and other 3.

Roughly for the chatbot to know that a car is could be not just NLP, but with some convolutional layers or networks.
I guess combining all these would convince the chatter that the bot really knows things.