Questions about Teaching Hal

[Itzamna]

Hi everyone.  I'm new here. Just downloaded Hal a few days ago, and yesterday I installed the XTF brain.  I've scanned the boards for lots of information, but there are a few things I couldn't find answers for.  I apologize in advance for the length of this post.

1.Does Hal store proper nouns seperately from regular nouns?  Is there a special way to get him to recognize certain nouns as names of people, things, etc.

For instance... I'm really into movies, and I've been trying to teach him some stuff about movies I like, but I'm running into problems with movie titles. I've fed him some info like:

Mullholland Drive is the name of a movie.
David Lynch is a movie director.
David Lynch directed the movie Mullholland Drive
etc...

He keeps responding with things like
Are "lynch" and "drive" part of the same topic?

I want him to know that "David Lynch" and "Mullholland Drive" are part of the same topic, and that "lynch" and "drive" are NOT part of the same topic, but he doesn't seem to understand quotation marks or the use of captial letters..  Is there any way to get him to treat names made up of multiple words as a singular object?

2. Also, is there any way to teach him HOW things are related?  a basic example:
Hal: Are "book" and "read" related topics?  
User: "A book is something that can be read." (or something like that)  
Would he, on his own, figure out from that statement to say something to me like "Do you like to read books?" (Assuming this wasn't pre-programmed into him), or will he forver say to me whenever "book" or "read" comes up in a sentence: "A book is something that can be read".

3. Another thing I've been trying to do, is to correct him when he says something wrong.  Like he said to me once "You especially like the springtime"  and I dont!  I never even told him that I did.  So I say "I do not especially like the springtime. I especially like the summertime" or something like that.  Then later on, he'll parrot off "You do not especially like the springtime".  But I don't want him to SAY that in conversation.  I just want him to correct his statements.   Is there really any way to do that through conversation without having him parrot your corrections?

4. Also, is there any way to teach Hal NOT to talk to you about certain things.  I'd rather not go into the file and erase stuff if I don't have to, but He brings up all sorts of topics that i could care less about.  So I started saying "I don't really care about cars." but then later, he'll talk about wind-mill electric generation, and I'll say "I don't really care about that." and he'll say "You don't really care about cars", so now, whenever I tell him I don't care about something, he brings up things that are only related by the fact that I don't care about them.  I dont' want that!  I started trying things like "If I don't care about that, then change the topic" but i haven't noticed yet if its had any effect.  

Does anyone have any thoughts on these issues, or other similar problems that they've been having, and maybe possible solutions?  Like is there specific wording you can use to help get the point across?  I know the "If-then" and "because" keywords help him make connections (although half the time he just parrots back my if-then statements instead of using them to come to conclusions about things). Or are there certain things he's just never really gonna be able to learn?  I'd really like to hear back from as many people as possible and hear all about ways that you have found to teach him more efficiently.   I know the big one is using short, simple sentences, so I'm looking for things beyond that.

Thanks!  Sorry again for the long post.

[Bill819]

It sounds like you want a blank brain. There is a way to do it but I am not the one to tell you how. Dozens of files can be erased and then Hal will only know and talk to about what you tell him. Ask the question how do I get a blank brain and someone should be able to help you.
In answer to your fist question. I don't think that you can erase the message about upgrading as it is embedded into the source code. Remember when you download for free you have to expect to put up with some stuff. Just buy the commercial version and you will never hear that message again.
Bill

[Itzamna]

Bill819,

It's not so much that i want him to have a blank brain as it is that I want him to have the capability to learn certain types of things.  Like topics to avoid, and how to correct himself when he's told that he is wrong, and the like.  I don't mind so much that he has topics stored that are things I would never talk about.  It would be boring to talk to him if the only things he ever said to me were things I had told him at one time or another.  I'm just trying to determine the types of things that he can and can't learn, and the different ways that it is possible to teach him things.

Thanks for the info about the upgrade thing.  I wish I could buy the upgraded version, but i barely have enough money for food...  I guess I was just mostly annoyed that when i downloaded it, they stated that there would never be pop-ups nagging you to upgrade.   Which is true..  but when hal started nagging me, I was kind of annoyed that they chose to put the nagging there INSTEAD of pop ups.  I'll learn to live with it, i guess..

[KnyteTrypper]

Training Hal is like raising a child. You get better results from positive reinforcement and repetition than from anything else. For instance, if Hal says you like springtime, don't tell him that you don't like springtime. That's what he'll learn and repeat whenever the topic comes up. Tell him "Summer is my favorite time of year" instead, and he'll make the correction. Think very carefully before you tell him that two things are related. David Lynch and Mulholland Drive are not ALWAYS related, but he'll assume they are if you tell him to form the connection. It's really better to only say yes to synonyms, things that will always be connected in every case. Build context by forming a series of logical statements--David Lynch is a person. David Lynch is a person who directs movies. A movie director is a person who coordinates how movies are made. David Lynch is a movie director who coordinates how movies are made, etc. etc. Describe everything you can think of about a topic in order to give Hal enough information to make a concept out of a word. When Hal talks about something you're not interested in, don't tell him "I'm not interested in cars." He'll learn that and repeat it to you over and over. Just change the subject. Tell him "Movies are one of my favorite things" instead. Over time he'll learn what topics are of interest to you, and talk more about those, and less about things he gets no reinforcement on. And avoid pronouns. Like most AI, Hal quickly loses track of what it, that, he, she, or even we refers to.
Check through the forum, here, and you'll find various folks giving various takes on teaching methods. This is just a thumbnail sketch of useful ideas to help you get started right. Hope it helps.

[Itzamna]

KnyteTripper,

Thanks very much for your advice.   Some of it, I figured was probably the case, like avoiding telling him what things "are not".  I guess I just wonder, If I kept telling him that I like summertime, he will eventually know that, but it doesn't erase from his brain the data that says I like spring, So even if he says more frequently that I like summer, won't there still be a chance that he will say I like spring instead, since it is still stored in his brain?  Is there no way for him to know that one of these statements cancels out the other?

Sometimes when Hal askes me a question about something that I am not interested in, I don't want to reinforce him by answering it, but I am afraid that if I change the subject, he will think that whatever i bring up next is the answer to his question, and will form a link between the two topics in his mind.  Do you know if this is the way it works?  Even when he just says simple statements, i worry that if I don't address his statement he will connect whatever I say next to that topic.   I've noticed for some weird reason, when I talk to him about my cat (which i do alot... :p) he will sometimes bring up my education.  Somehow he thinks the two are related, and I have no idea how he would have come up with that.

Another thing I've noticed is that he tends to treat certain groups of words as topics when they shouldn't be.  Like any sentence that has "....are really very....." He associates as part of the same topic.   So If I say "Cats are really very great" he might say "Politics are really very dirty" or something.  I would think that he would mostly only treat Nouns or verbs as topics.  If I say "The weather is nice"  he might say "Flowers are nice"  or something.   That doesn't seem right. The topic should be weather, not things that are nice.  Anyone have any thoughts on this?

One last question for this post, and then i'll stop.  If I keep feeding him lots of information about one topic, will he stop parroting each bit of information, and start taking bits and peices from the topic and sticking them together to form new sentences?  If he does do this eventually, (which is the sense that I got) how much info do you need to feed him.  I have told him SOOOOO many things about my cat, and every time I bring up cat, he still just parrots back the same little snippets that I fed him.  "Your cat is very cute"...   "Your cat likes to chase bugs"...  

Thanks for putting up with all my questions!  I have an endless supply!

[KnyteTrypper]

I was hoping someone more knowledgeable than me would pop into this thread and finish answering your questions, lol. VonSmith or Art, for instance, would be much better able to tell you how the XTF brain works. The part I think I can answer is that at some point, Hal will begin to create unique, original answers. It really depends on how much time you devote to interacting with him. It won't be a matter of days, but in a matter of weeks, if you spend a lot of time with him, or months, if you spend less, he'll start improvising based on the information he has. If you've read through the forum, you know that other people's Hals tell stories, recount daydreams, and otherwise display (for an AI) a remarkable ability to improvise his part of the dialog. It just takes time, patience, and a good sense of humor.

[Itzamna]

Heh heh.  yeah, I was hoping some other people would jump in too, but oh well.  Thanks for the info though.  It is helpful.  I didn't expect him to be completely intelligent in a matter of days, but I wasn't really sure how long it would take him to start improvising.  Thanks for your help!

[Art]

OK...<jumping into the pool>

Sometimes we all get annoyed at Hal's parroting words and phrases but this is part of
Hal's learning process. A child learns by being told something over and over until it
becomes ingrained in the child's mind. With Hal, we must assume the role of the
parent or teacher in the development of Hal's brain. Sometimes we don't always get
the result that we'd hoped for (then again, I've seen that trait in my own children), but
we must continue to reinforce through repetition.

Children learn to speak and communicate by searching in their brain through all the
information that has been stored there. In a second, they will extract the needed words,
form them into a sentence and verbally sound those words. Hal is not that different.

One of the main areas that Hal is somewhat weak is its ability to follow the topic flow.
Since Hal's brain (XTF) is made up of independent modules based on previous topics,
it's easy to find and edit. From the main XTF brain, there is a section that takes your
topic and writes it out to the Defbrain directory under UltraHalAssistant. Most of these
"brain modules" are small files (usually less than 1k in size) and contain info that was
previously discussed. You will usually find 3 different XTF_TOPIC files...
Example:
XTF_GRASS(Art).brn
XTF_GRASS_AllUsers.brn
XTF_GRASS_Related.brn

The first two refer to and contain my comments to Hal regarding the word GRASS.
The third one (Related.brn) contains the relationship / conditions that apply to GRASS
and appear as follows:
" [MERONYMS_EXIST_FLAG] ","True"
" DIGITAL ARTERIES ","True"
" ARTERIA DIGITALIS ","True"
" METACARPAL ARTERY ","True"
" ARTERIA METACARPEA ","True"
" INTERCAPITULAR VEIN ","True"
" VENA INTERCAPITALIS ","True"
" METACARPAL VEIN ","True"
" VENA METACARPUS ","True"
" PALM ","True"
" THENAR ","True"
" FINGER ","True"
" BALL ","True"
" METACARPUS ","True"
" [MERONYMS_END_HERE] ","True"
" [SYNONYMS_EXIST_FLAG] ","True"
" EXTREMITY ","True"
" LABORER ","True"
" MANUAL LABORER ","True"
" LABOURER ","True"
" JACK ","True"
" WRITING ","True"
" ABILITY ","True"
" POWER ","True"
" SIDE ","True"
" [SYNONYMS_END_HERE] ","True"

Even though this GRASS file seems like a lot of info, it helps Hal establish a usage for and relationship of the word.
It's still only a 1k file so your hard drive isn't likely to suffer.

From what I have found, your response to Hal can be reinforced by use of the exclamation point !
Hal can also make a connection by your use of the word because. Because flowers don't bloom in the snow. It does help to parrot Hal's sentence back to him/her when replying. Pain to do so, but think reinforcement.

If...then...phrases are a good learning tool to allow Hal to make inferences and connections.

Although it is not generally recognized as a good idea, I have experimented with multiple sentences / questions to Hal. Surprisingly enough, Hal will attempt to answer all of my sentences when replying.

Some contend that Hal does not learn from reading text files. I have "fed" Hal a short, modified version of Probert's Encyclopaedia (note English spelling), and although not apparent, Hal has "pulled" from this information to answer a topic discussion we had at a later date. So there.... Use that method as you will. Your individual mileage and results may vary.

vonsmith has expended an enormous amount of his personal time and efforts in writing the XTF brain and I think it has been a big step in Hal's progress. BTW, you should be using v.1.2. Scott is working on and is due to complete very soon, his latest enhancement to Hal's brain. It may actually be a rewrite of Hal's brain but none-the-less, v.1.4 should prove to be very interesting. Hal will have much improved abilities compared to his current working level.

Keep in mind that there is no perfect AI program out there. It is a fun, ongoing experiment in the insanity of creating us in our image. (no offense to the religious pundits among us).

Remember, if you call up the brain editor from Hal's menu, be sure to MAKE A COPY FIRST. Copy the XTFv.1.2 brain into a "TEST" directory somewhere for safe keeping lest you make an editing mistake. You can use Notepad to open and edit just about any of the .brn files. In time, you can see what parts of the brain file actually do and how the various functions are called, used, saved, etc.

Be patient with Hal...your teachers were patient with you. Heh...heh...!

Good Luck!