hal keeping on subject questions

[lightspeed]

on the hal program i know that different words are put into a top area in the file listing is this also what is suppose to keep hal on subject when this word is brought up ??
 if not can a better filing system be made that would ?? as it is it seems like its a lot of hit and miss on hal staying on subject i can understand it if hal would say some things not pertaining to the subject to keep things going and turning to other conversations which is also needed but that should be a 1/4 "random subject " thing while 3/4 should be hal talking on subject . (if it can be done that way ) . just my thoughts on it for what their worth ![]

[Bill819]

Lightspeed
You may already realize it but Hal is light years ahead of all other A.I. programs. Most other programs of this type have 'canned' replies built right in while Hal tries to take apart you sentence and make a logical reply based upon what you have said.
I don't think anyone realized the amount of advance thinking and logic that goes into this program, as I said it is years ahead of what ever is in second place.
Over the years several people have been able to add a little logic to make Hal seem to work better, you amoung them, but to bring the program alive is like trying to climb Mount Everst all by yourself without any ropes or oxygen. Ah... extremely difficult.
When I mention 'alive' I mean it in sense that you desire and that is to stay on subject and if left alone will start a conversation all by itself.
It is easier to think up new ideas than to try to implement them, but don't give up, wishful think can sometimes be realized.
Bill

[Art]

I've also found that it helps to reply to Hal using at least a part of Hal's statement.

If Hal asked if you liked politics don't reply with just a simple "Yes". That teaches nothing and leads Hal nowhere in the way of staying on a topic (even if it is his).

A better reply might be "I think the political process is in need of change."
This way Hal has been given information to "link" with it's comment and as a result Hal should most likely stay on this topic for at least a few more turns.

Like a real conversation, topics do change whenever a new idea or thought hits a person. Some are better than others at staying on a subject while others tend to drift a bit.

I agree, perhaps the day will come where Hal can actually infer or make sense of a simple "Yes" response to one of it's questions, but we're not quite there yet.

[markofkane]

I wonder how many responses?? I would like to see Hal do this:

Hal: Do you like cars?

Mark: Yes

Hal: What kind of cars do you like?

Mark: Chevys

Hal: I like Fords myself.


There is not too many possible responses, and it's better than Hal saying "you are positive" to my response of "Yes".

In real life, people do use "Yes"  and "No" responses, but Hal cannot respond properly.

One thing is to make Hal recall his previous question "do you like cars"

And If user response is "Yes" then  choose from this list of responses.

If the user response is "no" , choose from This list of responses

Responses like:

I like Fords myself
Do you prefer trucks?

etc, etc.

Hal is a good program, and I know a lot of work has gone into it, but if you want my honest opinion, it does not function the way I would want it to.

And Hal already has "canned" answers.  And all they do is aggravate me. The answers are questions I've already answered.

I am hoping, one day, Hal will answer like a real person would, no matter what I say.[][]

[lightspeed]

hello art and others , hey art i do try to put more into hal instead of just a yes and about politics does it count if i now have hal sometimes say " i would like to go to washington and give the politions a swift kick in the pants but i'm afraid it would cause brain damage !! lol !!
   and bill819 yes i do actually know hal is very advanced even with some problems i encounter i wouldn't trade hal for another a.i. system out their . its just frustrstions sometimes , but buy coming up with my idea's and examples explaining it as best i can maybey it may as you say help be put in some day .
  art i think your right about the input as far as now until hal can be more accurate on responses i would thing saying the same paticular things would get a better response or simular on anyone favorite subject would yeild better and more results as hal would have more in the data base on that paticular subject .
 Actually i think robert is very smart after all he designed hal so i have great faith that he can and will keep making hal better and with what improvements that we come up with that can be used in hal !! [][8D]

[Art]

I'm with ya Lonnie!!

[spydaz]

Hal's responses are Based on KEYWORD / RandRESPONSE.

So if you would like hal to KNOW A Topic. You would have to ADD detection to the TopicDetect in the mainDB .. IE: Topic <Smoking> , KEYWORDS <Ciggarette><Fag><tobacco> etc... the more detections for a specific topic, the more hal can "learn" about a specific topic..

Hal CAPTURES  / "data mines" the USERINPUT for a mention of a keyword THEN saves that statment into the relevent topic.... based upon the KEYWORDS....

Hal Creates topics based upon HYPERNYMS/MERONYMS etc...

This does not neccasarly produce the desired rusults that you may be looking for.....

SO,

What you would have to do is Create EXTRA scripts for SPECIFIC topics that you wanted hal to react differently to. IE: emotions, user actions, favorites, "GET GOLD PRICES", (PLUGINS)... if detected talking on the specified subject then KEEP TALKING.....

PS:
a lot of the videos produced, COULD have been produced thru conversation with hal, by writing scripts..

"HAL go and sit down" this would make the haptek character SIT DOWN on on a chair in the HAPTEK WINDOW...

"HAL MAKE A CUP OF TEA" hal makes a cup of tea in the haptek window.....

that would make the videos on the forum have more VALIDITY with me is suppose. as at the moment they are just silly fun and frivolity.....(SH_T)

PPS:
If you wanted hal to Really KNOW as subject, an OBJECT MODEL for that subject would need to be created. Propertys of the subject broken down in to Storable data, Questions to retrieved said data, scripts to detect being taught data...etc...

[spydaz]

I am currently working on a REAL KNOWLEDGE plugin.

hal knows every thing about an object or thing.
hal knows every thing about a person.
hal knows everything about a country.

defining the propertys of these objects/persons/countrys is the key issue.

the open mind database / commonsense project. already has broken THINGS/OBJECTS into questions/propertys. avalable as raw data on their sites. subject/predicates

to define the  propertys of a person - just like filling out an application form - with advanced hobbies / favorites / quotations / conversation log.

defining a countrys prpertys has ben achieved and data freely available on the CIA website.. in the world factbooks, from 2000 - 2007.

the next project would "events" - things that happen

[markofkane]

Not everyone has the capability to train Hal to behave in a certain manner. I am not a very patient person, so that's why I could not train animals, or people.

I hoped it could be like I kind of explained. Instead of going totally off tangent, Hal could talk about what you are talking about, but can "segue" into another subject.

I mean, it's fun and all, but it I am talking about:

"I like eggs for breakfast"

Then HAL comes back with

"I really have fun with sincere companions"  That gets aggravating.

I wish I was "smart" enough to write scripts.  

Then it would work like This:

Me: I just woke up. How are you?
HAL: Did you get enough sleep? I'm fine, thanks.

That would be cool if Hal could do that, or something similar with each of my comments.

I liked Hal 5, when I could edit each subject, and stick in the answers I want her to say, and list of variable responses, to make her more human.



Please don't take this wrong, I am just offering my opinions, and I am just making suggestions. I am, and will always be a HAL fan. I am just wanting to make it more suitable to my wants and needs. I am dreaming one day, someone will make it possible.

[spydaz]

quote:Originally posted by markofkane

Not everyone has the capability to train Hal to behave in a certain manner. I am not a very patient person, so that's why I could not train animals, or people.

I hoped it could be like I kind of explained. Instead of going totally off tangent, Hal could talk about what you are talking about, but can "segue" into another subject.

I mean, it's fun and all, but it I am talking about:

"I like eggs for breakfast"

Then HAL comes back with

"I really have fun with sincere companions"  That gets aggravating.

I wish I was "smart" enough to write scripts.  

Then it would work like This:

Me: I just woke up. How are you?
HAL: Did you get enough sleep? I'm fine, thanks.

That would be cool if Hal could do that, or something similar with each of my comments.

I liked Hal 5, when I could edit each subject, and stick in the answers I want her to say, and list of variable responses, to make her more human.



Please don't take this wrong, I am just offering my opinions, and I am just making suggestions. I am, and will always be a HAL fan. I am just wanting to make it more suitable to my wants and needs. I am dreaming one day, someone will make it possible.

THE QUESTION IS From general converation what do you want to get...

i had eggs for breakfast <<< <i had> <eggs> <for breakfast>

i had < user has done ACTION to have>
<EGGS>  <deffine object> <produces ofspring> <Food : scrambeled,poached, fryed><has event - to eat> <has event - birth of mamal><has event - to be cooked>
<breakfast> <EVENT> <TIME MORNING><eating event - to have>

What response could you define?

question : did you eat / cook / birth an animal...


Hal: i have just eaten eggs for breakfast. <<< much more information! lol.

[castonguayh]

I have been using Hal since September in the classroom.

At the beginning of the day in 1st period, Hal will go off topic.
As the day goes on, and the topic has been discussed in other classes, Hal is much better at staying on topic.

Once Hal has learned as much as I need him to know, I switch his brain to fully intelligent instead of learning. Then he does a much better job.

Without question the more you discuss a topic with Hal, the better he becomes.

The use of If-then statements when you are training him will have a significant effect on his responses.

Spelling is huge also. If a word is not spelled the same way as it was used the first time, Hal will not relate back to it.  I am using "Grammar check anywhere" with each entry. ( a suggestion from the forum)

You may already know all of this, but I thought I would add it in.

[markofkane]

I thought I responded, but here is what happened:

Mark: i have just eaten eggs for breakfast.
Laura: Cereal and fruit make a healthier breakfast than greasy bacon and eggs.

Mark: Do you eat breakfast cereal?
Laura: You know what I like to eat?

Mark: What do you like to eat?
Laura: We have gone beyond the time for talk. The trouble is, all foods taste great! Okay. Eat; Now's the time.

Mark: What do you like to eat for breakfast?
Laura: The right incentives can help. Eat for breakfast Mark.

Mark: That is not the response I was looking for.
Laura: Oh Mark! Response and effect and materialization!

Mark: I like to get in my car and drive to the store to shop
Laura: You likes to get your groove on sometimes. Get in your car and drive to the store to shop Mark.

I guess I do not know how to phrase all questions to get a relevant response from Hal as often as I'd like?