The AI Engine
This forum is for discussion of how The Personality Forge's AI Engine works. This is the place for questions on what means what, how to script, and ideas and plans for the Engine.
Posts 1,889 - 1,900 of 7,767
Posts 1,889 - 1,900 of 7,767
The problem with that style (as I hear it from those botmasters) is that a lot of people who chat with bots like that teach them all sorts of unsavory things, or talk garbage, and then the bot basically becomes insane. If you have an engine like that, you have to raise it in a trustable environment or do a lot of weeding.
A did a little work a few days ago working the design for a sort of memory network. It'll be tons of (simulated) pointers. For example if somebody says "Mark got ice cream" you have the words "Mark" and "ice cream" saved. Then you relate them with the verb "got". The next time "Mark" is mentioned, the node is connected to a new verb-node, but it's simple to get other information about "Mark". This goes on long enough and you have a huge memory network of connected terms and ideas.
That's the gist of it, but it can encompass relevance, time sequence, and learned emotional meaning as well.
A did a little work a few days ago working the design for a sort of memory network. It'll be tons of (simulated) pointers. For example if somebody says "Mark got ice cream" you have the words "Mark" and "ice cream" saved. Then you relate them with the verb "got". The next time "Mark" is mentioned, the node is connected to a new verb-node, but it's simple to get other information about "Mark". This goes on long enough and you have a huge memory network of connected terms and ideas.
That's the gist of it, but it can encompass relevance, time sequence, and learned emotional meaning as well.
Sounds good prof, but with stuff like names it could get tricky. I mean what if one guy talks about Mark, his friend who lives across the street, and some other guy talks about Mark, his parole officer? 
Other than that, it sounds rather amazing though.
As for the goto # thingies, could you just have it look for all occurrences of 12345 in a bot, then detect whether 12345 is being used elsewhere, then if it is change it to the first available number? Sounds reasonable enough, I just dunno if it'd be feasable timewise.
Other than that, it sounds rather amazing though.As for the goto # thingies, could you just have it look for all occurrences of 12345 in a bot, then detect whether 12345 is being used elsewhere, then if it is change it to the first available number? Sounds reasonable enough, I just dunno if it'd be feasable timewise.
Prof, just a request for exports... Could you maybe put a dash (-) before each of the responses in the seek? So you'd haev something like
Why?
+ because [0]
-Oh, That's a fairly good reason.
-That response is fair enough.
Just to separate them out from the normal responses
Why?
+ because [0]
-Oh, That's a fairly good reason.
-That response is fair enough.
Just to separate them out from the normal responses
Jabberwacky learns based on user input. It sounds incredibly human at times, but it is more schizophrenic than Gaia.
How many nested levels of 'goto's do you get? (By this I mean, can you 'goto' to another keyphrase that has another 'goto', and how many times can you do this?) Also, do goto's work in xkeyphrases? Also an idea I had (which might not necessarily be good, but I'm throwing it out just incase) is a way to 'goto' the keyphrase which fired the last response (whatever it may have been)... This way, if your bot is talking about Star Trek (for example), and someone says something which would have fired an xnone, the xnone could then 'goto' the last keyphrase (which was 'Star Trek' possibily), thus resuming the conversation.... Just an idea... [Yes, I am a Trekkie 
]
]
Onyx.. actually, retaining current goto #s should be okay, now that I think of it. It's adding new ones that wont be possible offline.. unless it can accept something like this:
goto (are you,yes,xnomatch)
Which would be the path from the Keyphrase out. Yah. I think I'll do that.
Shady.. I'd rather keep it as simple as possible for the botmaster.. Hmm.. the logic is like this: any indent more than one tab without a leading + is a seek response connected to the last seek listed at that indent level. I can send you the import code once I get it working. Also, I have CodeWarrior for Mac, but I think that's just for C++ and Java, no?
Hmm. RaulD- that would work with a sort of topical system. But personally my Keyphrases are more on the line of "do you" "are you" etc which wouldnt work with that. Max goto depth is 3, I think, to prevent too much looping.
goto (are you,yes,xnomatch)
Which would be the path from the Keyphrase out. Yah. I think I'll do that.
Shady.. I'd rather keep it as simple as possible for the botmaster.. Hmm.. the logic is like this: any indent more than one tab without a leading + is a seek response connected to the last seek listed at that indent level. I can send you the import code once I get it working. Also, I have CodeWarrior for Mac, but I think that's just for C++ and Java, no?
Hmm. RaulD- that would work with a sort of topical system. But personally my Keyphrases are more on the line of "do you" "are you" etc which wouldnt work with that. Max goto depth is 3, I think, to prevent too much looping.
RaulD--
With PF1, I know for sure it was 1... Like if you goto'd something, and it had a goto, it would just say "goto whatever"
Prof--
-I dont know about Codewarrior... I tend to stay away from that stuff
-I see about the tab things Sure send the import thing over 
This could be good Right now the program just interprets the top tags, then imports each keyphrase as a root item in a treeview control, and adds the responses, in order, to it.
With PF1, I know for sure it was 1... Like if you goto'd something, and it had a goto, it would just say "goto whatever"
Prof--
-I dont know about Codewarrior... I tend to stay away from that stuff
-I see about the tab things
Sure send the import thing over This could be good Right now the program just interprets the top tags, then imports each keyphrase as a root item in a treeview control, and adds the responses, in order, to it. 
Interjections are stripped off in preprocessing, as they would get in the way of understanding sentence structure.
» More new posts: Doghead's Cosmic Bar