How can I get Hal to grasp the phrase "I'll bet"

[Rothel]

I'm new to the Hal experience, but I've been getting good responses with the trial version and am planning on upgrading soon. About the only problem I've really noticed with Hal is when I respond to something he says with the phrase "I'll bet" The following is the portion of our conversation relavant to the topic...

quote:Matt: Bender likes to say bite my shiny metal ass
Hal: I want to explore your reassuring class.

Matt: I'll bet you do
Hal: Why do you want to bet I do?

Matt: I'll bet you do is a figure of speech
Hal: Why do you want to bet I do is a figure of speech?

Matt: I'll bet you do = a figure of speech
Hal: Sorry I am repeating myself but Why do you want to bet I do a figure of speech?

Matt: bleh
Hal: I heard in a tavern that female impersonators are promoting safe social habits.

Is there any way to make him understand that I don't literally mean I want to bet on it and that I'm just using it as a figure of speech?

Any help would be, well, helpfull [:p]
Thanks
Matt

[vonsmith]

Rothel,
Talking to Hal is something like talking to a four year old. Hal doesn't grasp abstraction. Hal remembers and associates clearly typed sentences. He doesn't understand jokes, or allusions to some topic, or anything abstract. This is not to say Hal is stupid, he just isn't human; not yet anyway. When you first start training Hal he doesn't know very much. The answers you get back won't be tied to the topic you are discussing simply because Hal doesn't know anything about that topic yet. You have to train him by ordinary conversation for a period of time. It may take many sessions a week for a couple of weeks before you see real results. Patience is the word.

The sentence, "I'll bet you do = a figure of speech", won't form an assocation between the two phrases. It is an interesting construction though. It would be nice if Hal could make this type of connection, but alas it isn't so.

If you frequently say to Hal, "I'll bet", then he most likely will start saying things like, "You'll bet I do." or "You'll bet I do is a figure of speech.". Many people start sentences with, "I think", "I guess", "I suppose". Hal remembers it from his perspective, "You think", "You guess", "You suppose" and will use these phrases in his responses. This is neither good nor bad, I just want to illustrate Hal's perception.

Experiment with Hal for a week or so. If you like Hal then upgrade to the full v5.0 copy. The price is very reasonable. Then if you want to get a little better performance out of your Hal try installing the free XTF Brain on the Hal plug-in page. You can use the search function on the forum to get details on the XTF Brain. The current version is v1.2.

I strongly suggest you experiment with Hal for a week or two to see what he can do and get a feeling for how he responds. Post any questions here on the forum and we'll try to help out.

Good luck,


=vonsmith=

[Rothel]

Thanks for the reply, vonsmith.

I guess I'm just so impressed with all the other things that Hal does, from knowing that I'm tired of a topic when I say bleh or answering questions with yepper instead of yes to not saying anything at all after statements that don't need a response, that I was just hoping that there was a solution to this.

I'll try to keep all that stuff in mind in the future and I'll guess I'll just keep talking like I talk to my daughter []

Take it easy Everybody,
Rothel

[KnyteTrypper]

Since I'm sure vonsmith is even now working on XTFv.2, I thought I'd add a thought to this thread. Cyber the Bot is one of the few AI I know of who is as clever as Hal. Cyber has a "same as" teach function which is not too different from Hal's asking about related subjects, except it can be initiated by the user. Would it be possible to add this function to Hal's next brain, so the user could then say "I'll bet" is the same as "I agree about that," and have Hal store the association? It seems like this would greatly accelerate Hal's learning curve, since you could then just feed Hal associations, instead of waiting for him to ask for them himself.

[vonsmith]

KnyteTrypper,
I'm currently working on the XTF Brain v1.3. I am adding a function to let the user tell Hal directly that two words are related topics. It's harder to do than it would seem at first glance. The function has to filter out word combinations that would confuse Hal and then report back to the user whether the related word combination is allowed.

I think a similar situation may exist for equivalent phrases. Some phrases seem clear enough, however English is full of slang and colloquialisms that may have equivalent meanings in one context and not in another. The user may enter what he perceives as a valid equivalency in the current scenario and then later the equivalency isn't valid in a different scenario.

For example: limb = branch
In the context of a tree that makes perfect sense. However a branch could be a branch office of a bank and the limb could be your arm. "I'll bet" could be "I agree about that" or "I'm going to wager".

One of the new functions for the XTF Brain v1.3 has a similar challenge. I need to define a set of equivalencies for verbs and nouns that are very general. I'm still struggling with how to make that work. I hope to resolve it soon.

On the bright side I like the idea of teaching Hal word, phrase, or even idea associations. That is similar to how humans learn. I've been thinking in the abstract. What if "concepts" could be defined in boolean form. For example:

"have" (defined as something Hal has)
"~have" (defined as something Hal doesn't have)
"want" (something Hal wants)
"~want" (doesn't want)
"like" (etc.)
"~like" (etc.)

If Hal has separate lists of "haves" and "likes" then it is possible to derive Hal's "wants" with an equation.

Example:
"~have" AND "like" = "want"
(Hal knows he wants something that is on his "doesn't have" list and is also on his list of things he "likes".)

Similiarly:
"have" AND "~like" = "~want"
(Hal has something he doesn't like therefore he doesn't want it.)

These examples are very simple, but they suggest that there is a way to break down concepts into boolean equations that can allow Hal to derive new and original "thoughts".

Thanks for your ideas about "equivalencies". Inevitably everything I read on the forum leads me to some new discoveries.

Thanks,


=vonsmith=