Laura Patterson - Nov 5, 2011:

I am not using JAIL for Marie,

JAIL is currently a private framework that I built to speed AI and chatbot development. It is designed to be platform independent able to run in either a server or a client environment.

I looked at JAIL, and although robust in its API,
Thanks!

it is lacking in its ability to perform custom functions without a lot of additional scripting.
I am interested in how you reached that conclusion. Custom functions are just JavaScript functions. JAIL then allows you to call them directly from the responses. Something like:

This function would pop up an alert box and also return the text to the response stream.

to use the function in a response is trivial.
“Look who is here! [Laura()]”

I am from the school of building from the ground up keeping the code lean and flexible.
Me too!

I did attempt to request an Internet search using SKY-Net, but I guess that functionality is still in development and not available to your current release.

It can sometimes be confusing talking with Skynet-AI about how it was developed. It tries to retain the Skynet persona so to retain the illusion you might not get direct answers. This is intentional. Skynet does have full search integration. But, the user cannot tell it what search engine to use.

Skynet uses an approach that I call a cascading search algorithm. It parses the input and attempts to direct the search to the engine that has the best chance of responding. In some cases if one engine fails it moves on to a secondary engine. Some searches are performed through simple web page integration will others work with search engine apis.

The response I received was that it had full Google integration, which I will assume means calls to the Google Custom Search API.

Currently Skynet does not use the Google Custom Search API. All the Google searches are done via web page integration. As Google has become more restrictive of showing results in Iframes I moved some of the searches to other engines. I do have the Google Custom Search API running on my test platform, but haven’t felt the need to put it in Skynet-AI yet.

Skynet-Ai does use the Bing search API built in. It has pros and cons versus Google.

Will you also be incorporating ranking based on complete context or just keyword association?

As mentioned earlier, in the Skynet-AI implementation the AI attempts to understand the context of the query and direct it to an engine that will provide the most correct response. This eliminates the need for ranking the results (although I could add that if I found the need in other applications).

Thanks for the detailed response.

It would appear that the level of sophistication is much higher in SKY-NET than what I am developing for Marie. This may also because our project goals are somewhat different. You are building a development platform whereas I am building a commercial website where the AI is to be more of a component than the end product itself. A lot of effort in my part is going into the graphics and UI, where I have heard you state that your focus is in the core development of the platform, which it should be.

I also believe that it is counter productive to “reinvent the wheel” but at the time of starting my development , I was unable to find any client side platform to incorporate AI into my project, so I had to start more in less from scratch other than a little jump start with some scripting a friend of mine started but never really fully developed. 

I should not make any judgements as to JAIL’s abilities until I can evaluate a full beta release. My hat is off to you as to taking on such a monumental challenge, and I look forward to seeing more progress with JAIL.

Version .005 of Skynet-AI is now on-line.
I will be entering Skynet-AI in the Chatterbox Challenge, please help me by testing out this new version.

Key changes in this version:
- Initial implementation of temporal reasoning & date math.
I have added functionality to try to parse and memorize dates and to do date related math. Skynet now tries to understand things like “next Tuesday”.

- Streamlined high speed word math problem solvers.
I made additional structural changes in the word math problem solvers. Functionality that looked like it could be used by standard parsing was moved in to the “UBER-Parser”.

- Initial implementation of hyperlinked, cascading neural net.
The prior implementation of the Neural net used bundled clusters and basic cascading priority flow control to identify and generate responses. Now, flow can jump back and forth between clusters like a hyperlink.

- Initial implementation of direct wikipedia page access based on context.
Added a feature to try to resolve ambiguous wikipedia pages directly.

- Initial integration of Uber-Parser.
This is a sub-set of the JAIL Uber-Parser project. This project is designed to in-line bot grounding knowledge prior to response parsing.

- Extended dynamic memory/learning system.
Additional refinements have been made to the memory/response system.

- Extended spell check system.
Since most neurons are fuzzy, spelling errors have less impact than might be found on other systems. But, common errors which prevent understanding are auto-corrected before parsing.

I will check it out Merlin.
I hope to have Marie ready for public beta by March 1st so she can enter CBC too. Interesting since both our bots are JS based.

Thanks to everyone who has helped me test out the current version over the past few weeks. I discovered a number of bugs and added a few features based on your feedback. The Chatterbox Challenge starts next week. If you have an opportunity, have an another discussion with Skynet-AI and help me put the finishing touches on before the start of the contest.

Skynet-AI

Just had a 15 - 20 minute chat with Skynet-AI.  (7:53 pm Eastern Time).  Yes, it does seem to be improving.

Thanks Victor. There are still a couple of glitches in the memory subroutine that showed up in your and a couple of other conversations. Hopefully I’ll have a fix by tomorrow.

Great job Merlin.  I am very impressed with your memory recall functionality since you are mainly using JS and not depending on database queries. Memory functionality is something that I am just starting to develop with Marie. She does have the subject and topic history arrays, but these are not being used in such a way that Marie can recall the entire context of a converstation.

I am looking forward to see what comes next with Sky-Net.

But, if taken gramatically, isn’t “2 add 2” a bit out of the way, and not a sentence to be properly understood by an Artificial NLP recognizant?

(This was to a post of Ethan Whitt)

Merlin - Nov 5, 2011:

Laura Patterson - Nov 5, 2011:

I am not using JAIL for Marie,

JAIL is currently a private framework that I built to speed AI and chatbot development. It is designed to be platform independent able to run in either a server or a client environment.

I always wanted to know, how can one develop a language, or else, at least a Library?

Can I get any Idea about this?

Abhishek Bhattacharya - Mar 29, 2012:

But, if taken gramatically, isn’t “2 add 2” a bit out of the way, and not a sentence to be properly understood by an Artificial NLP recognizant?

(This was to a post of Ethan Whitt)

Yes, you have a choice when you are doing NLP on how fuzzy you want your recognition to be.

In the case of Skynet-AI it ignores “2 add 2” but will recognize “2 and 2” or “2 plus 2” as math queries.

Regarding frameworks: I decided to follow a philosophy I think was attributed to Abe Lincoln,
“If I had an hour to cut down a tree, I would spend the first 45 minutes sharpening the axe.”

I wanted to create tools that would allow me to rapidly build bots/AI. My history before Skynet-AI was to build about 1 bot per year. Even with Skynet-AI I do a lot of test projects as separate proofs of concept before it gets into the bot. Skynet-AI’s advanced math capabilities are one example.

Javascript frameworks are popular with jQuery probably being the most used. Here is a write-up of things to think about if you are interested in creating your own:
http://dailyjs.com/2010/02/25/djscript-part-1-structure/

Oh, thanks. I was just wanting that sort of an information regarding building my own!

I’d look forward to gain entire knowledge about it, and then repost here for problem redressal!

Skynet-AI placed second in Chatbot Battles, the most extensive competition of chatbots yet.
http://www.chatbotbattles.com/

It did especially well in the question answering sessions (which is what it was originally designed for).
It beat some very capable bots including:
Tutor- Ron Lee’s ESL AIML bot
http://www.rong-chang.com/mobile/tutor_mike.php
“Tutor” won one fourth place and one first place in 2011 Loebner Prize Contest

Elbot - a chatterbot created by Fred Roberts, using Artificial Solutions’ technology.
http://www.elbot.com/chatterbot-elbot/
Elbot won the Loebner Pirze in 2008, and was a multiple year finalist and most popular bot in the Chatterbox Challenge.

CallMom A.L.I.C.E. - Dr Richard Wallace’s AIML bots
CallMom Fake Kirk
http://callmom.pandorabots.com/static/callmom/index.html
CallMom is Pandorabot’s new Adroid phone application.

Eugene Goostman - Vladimir Veselov/Princeton AI
http://www.princetonai.com/bot/
Eugene won the largest turning contest ever held earlier this year.

Talk-Bot - Wendell Cowart
http://www.frontiernet.net/~wcowart/index.shtml
Wendell was the creator of the Chatterbox Challenge, a long running competition for the best chatterbots. Talk-Bot was a long time competitor and past winner of the contest.

I have decided to take Skynet-AI off-line for a time in preparation for the next version.

During the last 3 months Skynet-AI has been participating in a thesis study on English language learning.

In an exit vote the participants were asked to choose the most suitable bot for language learning and Skynet-AI was voted as the best.

When the author publishes his results I will try to provide a link.

Wow, congrats Merlin. What was the native language of the study participants?