[8pla.net]

I just created Artificial Intelligence JavaScript Object Notation, which implements a JSON dialect for creating natural language software agents.

Demo Page:

http://www.chatbot.ml/chatbot

Source Code:

http://chatbot.ml/chatbot/json.js

Format:

var JSON = [[
"HELLO|HI|HEY|GREETINGS",
"Greetings",
"Hello there.",
"Hi you.",
]];


« Edit Notes: Fixed broken links from domain name aihax.com to new domain chatbot.ml

[infurl]

That's actually a great idea, but why a "dialect"? Isn't that still going to have compatibility problems? I was expecting to see some labels saying what the different elements are instead of having to guess. Brevity for it's own sake is a big obstacle to productivity. Have you got some converted non-trivial AIML examples to demonstrate the idea in more depth? Criticisms aside, I do think it's a great idea.

[Art]

Eliza

[infurl]

Eliza

I was thinking Alice actually.

[Art]

OK, but if you examine the way in which it tried to elicit responses it seemed very reminiscent to me of that ever-so-helpful Rogerian therapist mentality that was first exhibited by the first Eliza (Elisa) program back in the hey day! Yeah...I'm old...so what! Haha!!

I believe Alice (the chatbot) came later, much later than Charles Dodgson's (Lewis Carroll) Alice in Wonderland, which in itself, was a far better entertainer of the mind, IMHO.

[8pla.net]

@infurl:

That's actually a great idea, but why a "dialect"?

I meant a dialect of JSON.

Isn't that still going to have compatibility problems?

Yes.

I was expecting to see some labels saying what the different elements are instead of having to guess. Brevity for it's own sake is a big obstacle to productivity.

Instead of labels which are associative, it is indexed numerically.  Whenever, the index is zero, that identifies it as matching the stimulus.  Or, as a response list, otherwise.
 

Have you got some converted non-trivial AIML examples to demonstrate the idea in more depth?

No. I do have some new ideas related to using JSON for A.I. though.

Criticisms aside, I do think it's a great idea.

A brand new implementation being compared to long established implementations is a great compliment.

[infurl]

Instead of labels which are associative, it is indexed numerically.  Whenever, the index is zero, that identifies it as matching the stimulus.  Or, as a response list, otherwise.

Yes I got that just by looking at it, so it does work as it is. Also the casiness of the text was a clue. The problem is that simple things rarely stay simple and when you want to add more features, the syntax might become harder to decipher. Incidentally I can't help making comparisons to ElfScript as well. Looking forward to seeing where both these projects go.

[8pla.net]

Here is an update to the implementation...

BEFORE:

"YOU *" matched stimulus "You understand" but not "You" by itself, since "*" had to match on something being there.

AFTER:

"YOU *" now matches stimulus "You understand" and "You" by itself, because "*" matches something or nothing there.

You

Why do you care whether I?


You understand

Why do you care whether I understand?

[8pla.net]

The problem is that simple things rarely stay simple and when you want to add more features, the syntax might become harder to decipher.

That is true.  In a way, keeping things simple, can be its own challenge.  And, thank you for your valuable comments.  I think the JSON developer community may refer to labels as something they consider important called, "readability".

[Freddy]

I'm impressed that you did so much with so little code 

But yes, the simple becoming complex is something I very much came across with ElfScript. As a beta tester Art is very good at asking questions and I tried to be equally good at solving things with existing methods. 

[8pla.net]

I managed to make it even easier, with less code.

While this original A.I. JSON project continues. 

I thought an easier plain text version deserves a new thread of its own.

Reference: http://aidreams.co.uk/forum/index.php?topic=12963