[0JL]

Hi, I'm new here, so hello everyone!!

I'm currently making a small AI experiment, using RiveScript, I'd like to talk about. I've made a few posts about it on the RiveScript forum, but there's nobody there, hence I come here.

Here we go.

I believe that thinking is maybe just like talking to yourself. Since chatbots are good at talking, I'm working on a chatbot that I feed with its own replies, to see where it's going. Currently it can think, it can speak, and it lives in my cellphone. I'm using the Perl version of RiveScript.

The underlying (supposed) concept is that there's absolutely no passive knowledge in a mind. There's no upper ontology. Everything is activity.

Example:

In Opencyc, a penguin would be a kind of bird, which is a kind of animal, which is a kind of ...etc ...etc. In a human mind, I think it's just a thought "what's a penguin" that makes the next thought appear: "it's an animal". That's it.

What do you think?

( you can see my little toy here http://www.rivescript.com/yabb2/YaBB.pl?num=1405589167 )

[ivan.moony]

I think humans really work that way, but with some unification algorithm applied, although we can interrupt our train of thoughts in every moment to start thinking about something else. I think that there should be some kind of thought automation on association level, but also there should be some way of learning of new automation procedures. And those two should exchange the lead sequentially by some trigger, maybe another association?

[Art]

I fooled with RiveScript about a year or so ago and compiled a nice little bot for personal use / interaction / experimentation / not-ready-for-prime-time.
I heavily modified the AIML set (a,b,c,d...files), that had nonsense, garbage, swearing / foul mouthed remarks, foreign language and basically unwanted or unneeded responses.
This removed over 600,000 bytes from the files helping to streamline things a bit and to help make my bot sound less like a potty-mouth!

Mine runs in a Command-Prompt window from Windows 7.

I've written a lot of text files for it and it knows me by name with booted. I'd love to be about for it to learn from chat interaction but that's another head-banger!

If you wish to discuss / share any info don't hesitate to contact me. I'm certainly no expert at RiveScript but it is quite an under-rated language IMHO!

[0JL]

@ivan.moony
If I understand you correctly, I did exactly what you said! Thoughts are automatically associated with previous thoughts (that are part of what I call the "context"), and there's indeed a way to learn new automation procedures. What I want to do now is precisely a capability to interrupt the train of thought, starting on a fresh context, and once done go back to what we were thinking.

@Art
I agree, it is definitely under-rated. I saw your chat-log on the RiveScript forum, and I like it! I like the way you talk to bots, I find it fair, so to speak. "It" being the way you talk!

When it comes to AI, we all tend to look for a 1 global (magical?) solution. It's some kind of "holy-grail syndrome". It's not good. But if we start building a mind, one thought at a time, then we have a chance. It's not a head-banger I wouldn't say it's easy but look at this, it's a typical chat-log with an initially empty brain a la thinkbot, on RiveScript. Don't focus on the first line.

Code

YOU> newt xin when i say str please reply str prd tht 2newt 2xin star1 2prd 2tht 2xout star2 2thk 2xout ok i got it

thinking [ newt xin when i say str please reply str prd tht 1newt 1xin star1 1prd 1tht 1xout star2 1thk 1xout ok i got it ]
 
learning [
  + xin when i say * please reply *
  - 1newt 1xin <star1> 1prd 1tht 1xout <star2> 1thk 1xout ok i got it ]
 
YOU> when i say hello please reply hi there
 
thinking [ newt xin hello prd tht xout hi there ]
 
learning [
  + xin hello
  - xout hi there ]
 
thinking [ xout ok i got it ]
 
BOT> ok i got it
 
YOU> hello
 
thinking [ xout hi there ]
 
BOT> hi there
 
YOU> when i say i love str please reply i love star1 too sweetheart

thinking [ newt xin i love str prd tht xout i love star1 too sweetheart ]
 
learning [
  + xin i love *
  - xout i love <star1> too sweetheart ]

thinking [ xout ok i got it ]
 
BOT> ok i got it
 
YOU> i love you

thinking [ xout i love you too sweetheart ]

BOT> i love you too sweetheart

YOU>

It's all on my little webpage. I think it's worth investigating Any ideas welcome...

edit: I'd love discussing and sharing info... And also I'm looking for a good RiveScript bot base that I could "thinkify"... Can I borrow yours Art please?

[Art]

Sorry mate but I did mention that my bot is not ready for prime time. It currently contains lots of files that have been customized to my particular needs, likes and tastes. I can't send the entire batch of files but if you wish an individual file or two, please let me know. Otherwise, the group of files are simply too personal to loan or give to someone else. No offense meant.

I basically went through the AIML files and used two programs, Agent Ransack to locate each file and Crimson Editor to easily and quickly find words and phrases to either delete or modify to suit your needs.

Be sure to save the modified files. There are so many duplicates within some of these files as well as slangs, unwarranted cursing or obnoxious language that does not benefit any chatbot.
Once the foreign words, swearing, kid talk and lame Internet replies have been deleted, you should save the file(s).

The other changes I made were to the Admin, Begin.rs (where you'll provide your bot's age, gender, likes, etc.),  and make sure that any files you want your bot to use are in the same folder with your other .rs files.

Other than that, I'm using the standard RiveScript bot as I found on Noah's site.

Again, If you're looking for a particular file, let me know so I can review it first before I send it to you.

[0JL]

Sorry Art, I had misunderstood the "prime time" statement, I didn't want to be impolite. Thank you for your advice. It's ok



edit:

Instead of starting on an empty context when we want a temporary interruption of the train of thought, I've chosen to add a prefix to context items. It's easier and more flexible. Hence the new keyword "prfx" that means "prefix". Its purpose is to set the message that will be added at the beginning of every thought being saved in the context.

example

Code

YOU> thk foo

thinking [ foo ]

YOU> showctxt
 
current context =
  xin showctxt (10)
  xin thk foo (9)
  foo (10)
 
YOU> thk prfx super
 
switching to prefix [ super ]
 
YOU> thk bar
 
thinking [ bar ]
 
YOU> showctxt
 
current context =
  super xin showctxt (10)
  xin thk prfx super (8)
  super bar (10)
  xin showctxt (8)
  xin thk foo (7)
  super xin thk bar (9)
  prfx super (9)
  foo (8)
 
YOU>

Also, newt def and tget shortcuts has been removed. Now we have to use "thk something" to get it working.