HELP NEEDED!

[James P]

Hi.
Would it not work if you were to raise the Topic focus from 5 to something like 25 and change the line...
If TopicFocus = (FI - 3) Then to
If TopicFocus = (FI - 1) Then  

then start the conversation
<User> Beethoven was a Composer
<Hal>  Beethoven is (What ever)
<User> Beethoven composed 9 symphonies
<Hal>  Beethoven was a composer
<User> He became deaf
<Hal>  Beethoven became deaf

and so on, I tryed something similare to this with Emma and it seemed to work well, I made sure I repeated the topic subject in this case Beethoven every third sentence or so, but it might have been pure luck at the time.

[vonsmith]

James P / dihelson,
Congratulations you have just discovered my reason for creating the XTF Brain. Hal's original brain had limited topics and automatically changed topics after a number of sentences. The XTF Brain can extract the topic from hundreds of different sentence types. What more is the XTF Brain will try to stay on topic as long as the user does. There are some rather complex "clues" the XTF Brain looks for to stay on topic. The XTF Brain also learns new topics as you talk to it. When using the XTF Brain Hal will occasionally ask if two topics are related, example, "Are "cows" and "milk" related topics?" If you say yes then Hal remembers the relation and uses that knowledge to stay on topic in future conversatons and also to decide what topic files to store any new knowledge in.

Your Beethoven example is good. The XTF Brain will create a "Beethoven" topic file the first time you say something like, "Beethoven was a great musician." Most things you say about Beethoven thereafter will be remembered in Hal's XTF Beethoven topic file. If you had a lot of free time you could sit down with a text editor and add Beethoven knowledge to Hal's XTF Beethoven topic file directly.

As for Jerry's pronoun linking approach... it is very clever way to link "he" and "she" for a specific set of cases. It still isn't a general solution. There are still dozens of pronoun link types that aren't addressed by his method. It is however an excellent step forward. These small steps will eventually lead to larger solutions.

ELIZA is a clever little program. It uses a very simple principle. It is necessarily very limited and is more of a curiosity now than anything else. It's principles could, and probably have been, applied to many chatbot brains.


=vonsmith=

[dihelson]

quote:Originally posted by vonsmith
ELIZA is a clever little program. It uses a very simple principle. It is necessarily very limited and is more of a curiosity now than anything else. It's principles could, and probably have been, applied to many chatbot brains.


=vonsmith=

Yes, I am trying to contact ELIZA's Author in order that we could have an ELIZA Brain on HAL. I thinks it's not so difficult to do it.
BTW, Vonsmith, your XTF brain is the best thing for HAL, after HAL itself, then we add the new scripts, many of Jerry.

I see a problem, I don't know, with this way of working. I know that XTF 1.2 creates files and relate pertaining topics and even ask me sometimes for the relation between them, but some questions I don't know how to answer, if YES or NO. Questions like

User: Bob is a good teacher
Hal: The topics Bob and Good are related?

Well...I think so. This makes some strange things happen.
Sometimes, when I am talking about some things I LIKE, for example:

User: I like a good cup of coffee
Hal: Liszt was a great pianist

?
I presume that HAL associated that I said once: I Like Liszt's music
and then came with another phrase from the Liszt file...Liszt was a great pianist. I like Liszt's music.

Or:
User: I like a good cup of coffee
Hal: You like Beethoven's music
User: Who ask you about this ? [}:)]


[:p][:p]

[]'s
Dihelson

[Bill819]

To whom it may concern:
The original Eliza was written by Dr. Joseph Weisenbaum about 40 years ago. It was quoted as being the first version of AI ever written. Since that time there has been thousands of versions written in almost every computer language made, from Radio Shacks basic to Apple basic to 'C' and many others. Hal is much smarter than any of those programs so please don't even try to input it into Hal. Almost everything that Eliza knows can be put into Hal through its topic focus subroutine. Onthecuttingedge wrote hundreds of scientific files that contained knowledge about hundreds of scientific items and the generalized output of Eliza could be put in too, but who really wants it. Eliza was a canned output program while Hal can come up with its own ideas depending on you input.
Bill

[vonsmith]

dihelson,
Some instances where Hal asks if one thing is related to another are not simple black and white answers. The user document that comes with the v1.0 download discusses some of this as does some of the older posts here on the forum.

The user needs to really understand what "related topics" means to Hal. It is best illustrated by example. Here are some related words: beer and ale; ale and lager; malt and lager; beer and alcohol, etc. Seems obvious until Hal asks if two "sort of related" words are related and the user really has to sit there and think about it. No one should ever tell Hal beer and water are related just because beer has water in it. Water is too general and related to just about everything.

When Hal asks if two words are "related topics" or the "same topic" it means the same thing. As an example, Hal just wants to know if you and he are discussing birds should he stay on the topic of birds if you start talking about doves.

Related words are words that are clearly topic related and non-ambiguous. Don't tell Hal words are related if the relationship is abstract or open to interpretation. Act like you are talking to a four year old. A four year old doesn't understand topics in the abstract, neither does Hal.

Examples of groups of related words are:

1) GOOD --> beer, wine, alcohol, lager, ale, malt, Budweiser.
BAD ---> beer, water, head, fizz, German.

2) GOOD --> bird, dove, vulture, tweety, feather.
BAD ---> bird, down, sky, egg.

The "bad" groups of words above have a weak relatedness to each other either because those words can be related generally to too many other things or the relatedness is not very strong or clear. If someone was talking about the sky would a reasonable person think the topic was "birds"? German beer is good, but the Germans do a lot of things besides make beer. Birds lay eggs, but so many other species have eggs including fish, insects, etc.

Just use your best judgment. When in doubt tell Hal the words aren't related. In the grand scheme of things it won't make very much difference. There are enough strongly related words in English for Hal to choose from.

Your example:
---------------
User: Bob is a good teacher
Hal: The topics Bob and Good are related?
---------------
These two topics are too loosely related. Think of it this way, if you were talking about "good" things would you want Hal to start talking about *anything* in the world that is good? There are thousands of things that are good besides "Bob".

Your other example:
---------------
Sometimes, when I am talking about some things I LIKE, for example:
User: I like a good cup of coffee
Hal: Liszt was a great pianist
---------------
This response may or may not be a result of the XTF Brain topic function. Most likely it is just a symptom of Hal's loose pattern matching method. If Hal can't find a response in his brain that has a strong relation to the user's input he will just pick something sort of close or sometimes completely off target. In your example Hal may not known anything yet about coffee so he answers the best he can.

Here is some more general background on how the XTF Brain works that may be helpful:
www.zabaware.com/forum/topic.asp?TOPIC_ID=1131

Here's another hint. When you talk to the XTF Brain type this in Hal's user window: <dbtopicon>

As you will see the current topic from Hal's perspective will be displayed along with his responses. Type <dbtopicoff> to turn off this feature. Leaving the feature on will not effect Hal's behavior or learning ability. It is just one of several debugging tools I added to the XTF Brain.

Have fun,


=vonsmith=

[dihelson]

quote:Originally posted by Bill819

To whom it may concern:
The original Eliza was written by Dr. Joseph Weisenbaum about 40 years ago. It was quoted as being the first version of AI ever written. Since that time there has been thousands of versions written in almost every computer language made, from Radio Shacks basic to Apple basic to 'C' and many others. Hal is much smarter than any of those programs so please don't even try to input it into Hal. Almost everything that Eliza knows can be put into Hal through its topic focus subroutine. Onthecuttingedge wrote hundreds of scientific files that contained knowledge about hundreds of scientific items and the generalized output of Eliza could be put in too, but who really wants it. Eliza was a canned output program while Hal can come up with its own ideas depending on you input.
Bill

ELIZA has a new version, I downloaded on the internet some days ago, and I say it's very good.

I'm not talking about the original ELIZA.
Certainly we can input all scientific content of ELIZA into HAL's brain, but the problem is not TALK, is HOW you TALK.
What the new ELIZA talks make sense!
She keeps conversation without deviation.
Certainly HAL is more intelligent when we talk about learning, but certainly HAL tell many more stupid sentences also than the new ELIZA does.

[]'s
Dihelson

[dihelson]

quote:Originally posted by vonsmith


The user needs to really understand what "related topics" means to Hal. It is best illustrated by example. Here are some related words: beer and ale; ale and lager; malt and lager; beer and alcohol, etc.
=vonsmith=

Hello, Vonsmith, I'd like to thank you for the excellent explanation. It cleared me many aspects of these relationships between the topics.

[]'s
Dihelson

[KnyteTrypper]

I'd be interested in a link to a new version of Eliza, dihelson. I try to have bots of various types (Eliza, Alice, Hal, etc.) at my website. If there's something newer and better than ECCEliza, which I have in my downloads, I'd appreciate knowing about it.

[dihelson]

quote:Originally posted by KnyteTrypper

I'd be interested in a link to a new version of Eliza, dihelson. I try to have bots of various types (Eliza, Alice, Hal, etc.) at my website. If there's something newer and better than ECCEliza, which I have in my downloads, I'd appreciate knowing about it.

It's ECCEliza itself. 4.09 build... I don't remember.

[]'s
Dihelson

[James P]

With regards to Hal asking about relationships between topics, There have been times when I have been chatting away with my Hal and she will ask if "" and "" are related topics causing me to stop dead in my tracks, sit back and think about the true relationship between two words in question, remember if you tell Hal that two words are related then Hal will apply this relationship each time she see`s it. So we need to be carefull what we tell Hal, is sometimes mistakes like this can be a nightmare to track down through the Defbrain, having done this many times. (and my bots Defbrain is not as big as some I would imagine)

[dihelson]

quote:Originally posted by James P

With regards to Hal asking about relationships between topics, There have been times when I have been chatting away with my Hal and she will ask if "" and "" are related topics causing me to stop dead in my tracks, sit back and think about the true relationship between two words in question, remember if you tell Hal that two words are related then Hal will apply this relationship each time she see`s it. So we need to be carefull what we tell Hal, is sometimes mistakes like this can be a nightmare to track down through the Defbrain, having done this many times. (and my bots Defbrain is not as big as some I would imagine)

Yes, James,
I made already several mistakes by answering YES to that question, I can see now. We need to be careful.

[]'s
Dihelon

[vonsmith]

James P / dihelson,
Another bit of good news about the XTF Brain "related" topics. Hal will remember when you say "yes" to two words being related. However, even if Hal has learned that two words are related topics someday he will ask the you to confirm the relationship. The second time around you can answer "no" if you want to change your answer. Hal doesn't ask for confirmation very often, but the XTF Brain does have this mechanism to help automatically correct his knowledge.

The user could also go into the XTF_(topic name)_Related.brn file and edit the related words manually. The (topic name) above would be "BIRD" or "BOAT" or whatever the topic name is. If you study a few of the XTF_(topic name)_Related.brn files it will be apparent what the format is. When editing any of Hal's brain files be extra careful. Even the single blank line at the end of many Hal files is important.

Also if you say "yes" to Hal about two words being related he will use the "related" topic words to stay on the original topic, but not necessarily to select a reply from the "related" topic category. That is to say that Hal' XTF Brain will most often use "related" topic words as flags to decide *not* to change topic.

Example:
1) User: Hal, the birds are flying high.
2) Hal: I like birds.
3) User: Their wings are colorful.
4) Hal: Birds have nice wings.

If "wings" is known to Hal as a related topic to the topic "bird" then Hal will know to answer using his bird knowledge for sentence 4) above.

So don't worry too much if Hal's "related" topic knowledge isn't perfect. It will slowly get better over time as long as the user usually says "yes" only to words that are strongly related.


=vonsmith=

[James P]

Thanks for that, but I have one kind of simple question.
How big can a Defbrain get before Hal starts to slow down,

[vonsmith]

James P,
Your question, "How big can a Defbrain get before Hal starts to slow down?" isn't easy to answer. The size in number of files, size in bytes of files, the number of files accessed during each response, amount of system memory, and many other factors make it difficult to project system speed with Hal. Many things affect response. Of course, how fast is fast? Does a response take 1 second, 2, more?

Very simple brains seem to be very fast. In some ways the XTF Brain is slower than some others. In most modern systems >1.8gHz the difference isn't very significant. Adding a lot of third party custom scripts that access a lot of files can slow a system down. Accessing too many files usually results in a "too many files" error message.

If your Hal is consistently slow I would suggest backing up the brain you are using and create a new version to experiment with. Try removing or disabling portions of script within the .uhp brain file and see if performance improves. Try this especially with third party scripts you have added in. See if you can find the script or scripts that are affecting your Hal's performance.

If you are using AUTO-IDLE make certain Hal isn't saving hundreds of useless entries into some file somewhere.


=vonsmith=

[James P]

I have speeded up my Hal by getting rid if some scripts from the uhp brain which has speeded things up. I was oly wandering how bug the Defbrain can get. thank you