The Atomic Playboy and the Radiation Romeo

The button below will open a new browser window displaying the Flash interface for Atomic and Romeo (Version 16 with Preloader). You will find a page of introductory text, some instructions and then the interface where you can suggest a topic for conversation.

This version 16 uses the landscape layout, updates the heckler and end-of-conversation functions with an audio sign-off. All the features from previous versions remain - scroll bar control,custId variable allows me to better log and track conversations.

The chat-bots are hosted on the Pandorabots server under the Shared Service subscription. Please note, the terms of the Updated Policy Guidelines for Free Community Server state that the “Use of automated scripts to make your pandorabot talk to itself or another bot or script” is proscribed (Pandorabots 2011). This project is being developed with the agreement of the Pandorabots Inc management and we would like to acknowledge their support. ( Pandorabots )

Please leave a comment...

After you have had a play with Atomic and Romeo please use this link to leave a comment.
Maybe you could suggest a topic of conversation or a layout suggestion.
All suggestions gratefully received.

Tuesday, December 6, 2011

Some elementary issues

This post is a device. The purpose of this device is to force me to clearly articulate some elementary issues in the development of character and AIML. So, please excuse me if the following contains more questions than answers and throws seemingly disconnected ideas around... Such is my want...

1.  The common back story problem
Atomic and Romeo need to share a common back story - they are friends rather than strangers. Cool. Simply give them some common AIML categories... That sounds easy...

2. The too much recursion / dead-end problem
There are on YouTube, and other places, examples of chat-bots talking to each other (or of a chat-bot talking to itself). For instance,  look for Fake Kirk talking to ALICE - they use the same ALICE 'engine' and AIML architecture as I am using. These conversations are pretty banal - strewn with non-sequiturs and odd recursions. Is this as good as it can be? If so I've got real problems in this project.

I have a hypothesis. When the two chat-bots share too much AIML this triggers a lot of recursive patterns or leads to conversational dead-ends. As an example, both ALICE and Fake Kirk respond to the "Do you like science fiction?" question in exactly the same way. Neither chat-bot actually knows how to reply to the answer "Yes I love it, especially the works of Philip K. Dick." Both get hung up on the "it" - what does 'it' refer to?

What does this mean in practice? Atomic and Romeo need to share some but not too much AIML. I think this balancing act will be difficult to achieve. Maybe this balance will emerge from the writing and rehearsal process? Depending on emergent phenomena is a risky business. 

3. The start point problem
Basically I've got a choice of two starting points.
Option One - I could start with a completely blank slates. Atomic and Romeo as they currently exist know nothing - even less than Sargent Schultz! I could write a completely unique set of each of them.
Option Two - I could start with a generic set of AIML, for example the set. This is "a free AIML set, in the English Language as spoken in the United States, authored by the AI Foundation".

The advantage of Option Two is that it maintains some serendipity in the interactions that could take me places that I would not think of going. Rehearsing with actors who don't even know there is a script would be tedious in the extreme - I would have to craft every single and

Some conclusions - so far...
The plan is to go with the V1.5 version of Alice.
Edit the set ruthlessly - strip it right back.
Generate a common set that I can use for Atomic and Romeo.
Then, in rehearsal, add to their individual sets.

Hopefully this will allow the balance between what the share in common and what is unique to each to emerge.

No comments: