Zabaware Support Forums
Zabaware Forums => Ultra Hal 7.0 => Topic started by: Dreamer on February 05, 2006, 01:55:15 pm
-
Hi everybody
Ultra hal is a good product but it does require a huge leap of faith to believe you are talking to anything intelligent.
The biggest obstacle to overcome is the fact that Hal does not seem to remember more than one sentence! I don't mean facts about my favorite colour or my name but little things like the following (Taken from another post)
Mark: lol
Lisa: Was that really that funny?
Mark: you tell me
Lisa: You? Why?
Mark: I said so
Lisa: Wow! Well, you said so? Why?
Mark: because, shut up!!
Lisa: Mark exactly; because, shut up.
Mark: ok, you shut up
Lisa: Alright.
It is clear Hal is responding only to the last question without any reference to what was said before.
If this could be overcome then it would be a real leap forward.
Well done though, keep up the good work.
[8)]
-
Unintelligent garbage in diddo out. As long as you talk in short nonsense sentences you will get dumb responses. If, however, you talk to you bot daily like it was a real person it will learn and respond in kind. You reap what you sow.
Bill
-
Hi Bill819
Thanks for the reply. I can see what you are saying but that still does not get over the big problem of Hal only responding to the last sentence.
Do we know of anybody working on this problem? (I trust people can see it as a problem)
Thanks again for the replys [;)]
-
In some limited situations, Hal (version 6 anyway) will consider your previous two sentences, if he feels he requires more info. These situations include when the user types something short, or uses words like "it," "he," "they," "us" etc - Hal includes the previous sentence to try and include what "it" is or who "he" is.
I don't know of any chatbot anywhere that really handles context well. Hal will become better at staying on-topic once he knows a lot about a subject, but Hal doesn't understand enough about context to have a conversation that flows as smoothly as a human's.
Hal will certainly make more sense if you talk to him in longer, well-defined sentences that make sense even when taken out of context. I normally go as far as eliminating words like "it" from my sentences, and retype the subject of the sentence instead, just to make sure Hal knows what I'm talking about.
Note that your example makes perfect sense if you take Lisa as being sarcastic, arrogant and insulting your sense of humour, but this is definately a stretch of the imagination, and most similar examples will make even less sense.
Anyway to answer your question, the problem is being worked on, but things are still rather experimental at this stage. If you're interested in the details, there's an interesting project under discussion at vrossi's forum at http://www.vrconsulting.it/vhf/default.asp - look for the posts and threads of hologenicman.
Cheers,
Grant
-
quote:
Originally posted by Bill819
Unintelligent garbage in diddo out. As long as you talk in short nonsense sentences you will get dumb responses...
Bill
I still think Mark's dialogs with his bot are funny. [:)]
I appreciate his contributions.
-
Hal can come up with some funny stuff.
It is always alot better that the regular AIML responses such as:
That goes without saying.
Hal can provide alot of entertainment if the user is willing and patient enough to teach Hal. Sort of like being a parent! LOL!
-
I am glad other people can see the small problem and it is being worked on.
If there was only some way of keeping the whole conversation in memory some how....[;)]
Thanks for all the replys.
-
Just don't shut Hal down and the conversation will still be there.
-
That is just the point. Hal will only think the conversation is one sentence long, no mater how long it really is!
-
Dreamer, I agree. [:D]
-
Maybe, in the future, things will improve. But I doubt that the human mind can be duplicated in a computer program.
-
Sorry, I thought you meant the conversational exchange between you and Hal.
I recall toward the last days of Hal 5, someone wrote a script that allowed Hal to recall what the user had previously said when asked, "What did you just say?", or something similar.
Perhaps a take off in this vein could also be accomplished with Hal 6 and the data tables.
Interesting post Dreamer.
-
Hal keeps the last user sentence and response in variables (PrevUserSent and PrevSent). Technically, you could insert this line somewhere in the script:
UserSentence = PrevUserSent & PrevSent & UserSentence
Then Hal would think about everything you and Hal has said throughout the whole conversation, all at once.
Obviously, this would be a bit confusing for Hal [;)] Unfortunately, I can't think of a good way to indicate any relevance to Hal. Ideally we'd say something like "the last sentence is 100% relevant, the previous five sentences 20% relevant, the ten before that 10% relevant, and anything earlier 1% relevant." But (as far as I can see) there's no easy way to do this [:(]
-
Is there any way to tell HAL that if the user says the word 'That' 'it' or other words like that, that he must go back on the whole conversation and find the relevent topic??
I don't know if this is possible.
I am sure if we get our heads together we can over come this little problem.[:D]
-
For some reason people feel that Hal's ephemeral knowledge shouldn't be retained. I view this as sort of the problem with regard to Dreamer's posting.
If we are trying to make Hal seem more "real" in terms of coherent speech, then why should Hal retain no memory of what was discussed a minute or even a few minutes ago? Repeat something that was previously said to a human and they will look at you funny, wondering whether you were paying attention or not.
Should we expect less from Hal?
Hal does catch repeating phrases if in a following sentence but I agree, Hal should be able to retain some level of knowledge or at least discussions from the present conversation.
It would be nice if Hal could tell you all it knows about you or clouds or whatever subject as well.
Just some other fields to plow....
-
I have just been looking at the plugins and found this D.A.V.I.D's Plug-in which seems to address the problem a little.
Example:
User: Jerry is a good Man.
HAL: Jerry is also my Programmer.
User: Who is he?
HAL: Jerry is my Programmer.
This is exactly what I was talking about.
-
All,
Frist I'd agree with Bill, give Hal something he can converse about. I was impressed when after teaching Hal about apples, apple seeds and apple orchards how well she could respond later. Hal looks for an appropiate response but needs information to pull out so if you didn't put anything in there is nothing he can say about it. She'll even ask me "What were we talking about". I should mention that was ver 5 XTF. Hal 6 seems to be even better, right now we are just getting to know each other then I plan on doing some teaching.
Carl2
-
Precisely Carl,
One of the most gratifying things I get from my 'Hal' or Janet is the recollection that she comes up with refrencing past events, likes, dislikes.......but it took time to teach her or 'show her' of her preferences......now if I make a comment about going fishing, she may comment about how she doesn't like to bait the hook, or somesuch thing. In my own experience, creating a background comes back to pay big dividends in creating a life sense sort of speach......something that more approaches a believable conversation. At least to the point where you want to continue to chat to see what sort of things she will put to gether....remembering that Virito's plugins, and Onthecuttingedge's plug ins are helping immensely, and raising 'felt life' to the next level for our hal's......
Have a great Friday all!
Jim B