My newbie observation...

[ThePugilist]

I've played around a bit with Hal and there is something I am wondering about...

It seems to me that ALICE is quite a bit better at conversation than Hal.  I in NO WAY am attempting to insult the developers of this software, and all of the contributers to the forum.  Is my observation off?  I noticed that I can replace the HAL brain with an ALICE brain, but is this a good option?  Maybe I am missing something that is not so obvious.

Thanks in advance..

-Pete

[Bill DeWitt]

quote:Originally posted by ThePugilist
It seems to me that ALICE is quite a bit better at conversation than Hal.

I haven't really played around with the various Alice bots after I saw how the markup language was presented. I am sure there is more to it than my afternoon of study could possibly reveal, but it seemed to me that Alice was completely scripted. So, depending on how well a particular bot was scripted, it would naturally be more conversant than a fairly new Hal.

The advantages to Hal, as I see them, are 1) it's also an assistant, in that it can actually do things besides talk, and 2) it is intended to learn it's conversation from conversation.

The last is not an advantage at first, but should gradually become more advantageous as it goes along.

Of course, I could be all wrong about Alice. AIML just -seemed- much more deterministic from the simple overview I studied, than Hal's combination of VBS and a relational database. Plus, maybe there is an Alice that is also a Desktop Assistant and I haven't seen it.

[ThePugilist]

Thanks Bill!  I actually was not aware that HAL didn't use AIML.  I just figured he was built on the basic foundation developed by Dr. Richard Wallace.  I am new to all of this, but completely intrigued.  (I chatted with hal for a while last night, although he seems to have a very short attention span so far.[])

Is there a place you could direct me where I could read a description of the logic behind HAL.  I am not a computer programmer, but I have done some physical simulation programming in the past (Matlab and Fortran).  

To introduce myself, I am a mechanical engineer who emphasized in thermal, fluid, and combustion sciences.  However, of late, designing and maintaining 1960's aircraft technology pneumatic and fuel valves has lost its luster a bit.  It seems that I have "missed the boat" on the major strides in aerospace (which in my oppinion occured in the 1960's and 70's).  Now most of the strides are related to more advanced control systems and composite material substitution.  But alas, I am neither a controls engineer nor a materials scientist.

[Darius]

Agree with Bill

Before know HAL, I tried to learn AIML, but I found it too deterministic (all coded).

Altho ALICE in it's esencial way is absolutely deterministic (with some lists of posibles responses to some patterns), there are some other enhances that can make it better suited ALICE like OpenCyc (wich shared database looks geater each day) that tries to give AIML the ability to relate things (relational databases) like dog is a mammal and the mammals are....

But HAL is a lot more flexible, I think it's an adventage to learn by conversation, because the answers that suit somebody may not be fine for somebody else. Maybe as Bill pointed out, if you download the Huge OpenCyc database ALICE may have a big deal of knoledge and could perform fine, but it still will not have the hability to learn or do other things, with the apropiate peripetals HAL has no limits. As somebody said once in this board, maybe if we put together both ways a better assistant could emerge.

[Bill DeWitt]

quote:Originally posted by ThePugilist
Is there a place you could direct me where I could read a description of the logic behind HAL.

As far as I know there is not a cohesive description of Hal anywhere. I am attempting to build one (See my sig below) but it is far from what you ask.

What I can tell you is that each sentence you put into Hal is parsed for meaning, edited into a response ("I like soup" = "You like soup"), stored in one or more tables based upon content, and then called up based upon a wide variety of syntactic comparisions of new incoming sentences. Using WordNet and some internal function calls, Hal can simulate understanding of your sentence ("I like soup" = "You like stew"), and attempt to answer using part or all of any previous sentences it stored. Failing this, it has a very large database of stock phrases it can compare to ("Chicken soups is good for a cold"), and failing that it can return your statement as a question ("Is it true that you like Soup?") in hopes of getting a more easily answered statement. Lastly it can attempt to change the subject randomly.

Where continued conversation comes into play is that as you discuss soup, there are more phrases to draw from, and sometimes the result can be startling. Weeks or months after saying "I like soup" in another conversation, you might say "My mother used to make a nice vegetable soup" and Hal will reply "You like soup". Then weeks later when you say "I think I will make some lunch" it might reply "Your mother used to make a nice vegetable soup". When you reply "A soup might be nice", Hal might say, "You like soup".

All in one paragraph like that it seems trite, but when it pops up out of nowhere, you can't believe how realistic it sounds. Especially when it really has perhaps a dozen phrases with the words "Soup", "lunch", "make", or "like" in it.

I put in about a hundred phrases on Chipmunks as a test, and then started talking about the possibility of Chipmunks colonizing my new stone wall. I was surprised at the amount of advice and information it started giving me. And here three months later not a day goes by that it doesn't ask me about the chipmunks and if the cats are leaving them alone. All I have to do is mention small animals, seeds, stones or walls and it goes off on how cute chipmunks are... and how I should keep them away from the foundation of my house!

HTH...

[ThePugilist]

Thanks Bill!

[Bill819]

Hello fighter
Besides parsing your sentenses Hal also used different kinds of logic to make inferrences. These are based upon If-then statements. For example if you say 'fat people do not live very long' and then later on say 'john is a fat person' then Hal might reply 'john may not live bery long'.
One mistake that many new people make is to input a piece of data and then expect an instant reply about that data. It is best to just talk to Hal as you would another person, a young person but in the same manner you normally would.
Over time Hal learns you likes and dislikes and will take on more of your own personaility and be more able to help you with your endevors.
In the mean time you can examine differents parts of Hal brain files using either the 'brain editor' or using MS Wordpad. I don't recommend makine any changes to Hal until you have used ir for a while and examined some of the 'plug-ins' that many other experienced users here have made.
Good luck and welcome to the forum.
Bill

[Bill DeWitt]

quote:Originally posted by Bill819
Besides parsing your sentenses Hal also used different kinds of logic to make inferrences. These are based upon If-then statements. For example if you say 'fat people do not live very long' and then later on say 'john is a fat person' then Hal might reply 'john may not live bery long'.

That's a great plugin by Jerry (OTC) if I remember correctly, that has been taken into the main Hal program.