Author Topic: Let Hal Learn to its fullest concept!  (Read 42510 times)

jz1977

  • Newbie
  • *
  • Posts: 24
    • View Profile
    • http://dageeks.com
Let Hal Learn to its fullest concept!
« on: June 17, 2004, 04:15:17 pm »
I have been thinking about a few things for the past few hours and have been stumped on this Idea. Let me explain it to you to see if this is a revlent idea or if it is just not possible.

Ok,  so far as in what i have seen, Many people have been makeing chatter bots to comunicate like humans.  But what if we took this to the next level.  We all know that Hal can comuinicate with other people or as in some what.  We also know that Hal has learning capabilities, right? ok. What if instead of the spoken human language that we are working on so hard on to make a computer act human, we made Hal learn from other computers in their own language.  I got scared thinking about this idea, because it is not far off from the terminator movie.  Hal talks to another Hal, and that hal is talking to 3 others, and so on.  With a link of so many Hals running at the same time and learning from all the others in the chain. Making the central and/or All Hal's learn themselves to be smarter without our help of triing to make them more human like.  Instead of tring to make computers act like us, why not create a network of Hals talkting to each other 24/7/365 all at the same time.  This would not only make them smart and more intellegent as what Artificial intellegence is suppiose to be.  We are triing to make a machanical device have and act as a human does.  Well, sorry to say. but there are alot of humans who i would not want my hal speeking to, even if hal learns from every entered reply.  If we got our bots to be just the way we want them with every tweak and years of building Hals human intellegence. It is like the old sang goes.  "One person can ruin it for everyone." only everyone represents your years or even a life time of work for your Hal's brain.  I beleive if A.I. is going to grow and grow to be even more intellegent. It will never be like us humans.  My sugestion is if we really want A.I. to succeed sometime within the centry or less, Then we need to take this a different direction or to a higher level as i was saying above. Artificial Intellegence I beleive means. To make computers have a mind of their own and can make their own decicians.  But so far there has to be atleast two parties to really make anything happen for A.I. I bleive eventuallty our goal for A.I is for computer to almost become human and make decians on their own.  I think how we start by doing this is having hundereds or thousands of Hals running at the same time talking to each other.  Not only will this be more eficiant and faster, but there is not one bot that is alike in anyway, except if it is fresh out of the box. But once the fresh out of the box hal starts talking to other chains of hals it starts learning from them, just like us humans do with other humans. If it is true that each human has a uniqe personality, then we must have been around allot of people to acuire such a personaity.  We can bring our child up the way how we want them to.  But no matter how hard we try, they usually become or do activites that we diaprove of.  This falls into the same catagory as hal. If we want Hal to learn and become intelegent. we must show Hal the world and comuicate with others, others meaning as in other bots with different personalities. We may not like how our precsious Hal turned out in the end and we may very well love who our hal whas become. There is no sence is in programing Hal to be what we want to be, If some careless person says the wrong contect to Hal, then he starts become corupt.  I have just been thinking about this all day. and this is what my post was earlier about that i deleted, with the subject, "opps".  I just wanted to get my thoughts together more before i posted anything. Do you think that this is the right track for furthering our Future technology with any A.I. Software.  If a bot has the capability to learn. Let it learn what it wants, and in the end it will all be worth it. Caz then we can say that Hal was the first Software to really make A.I. Happen.

-Jeff
« Last Edit: June 17, 2004, 04:15:57 pm by jz1977 »
If you truly beleive it, without any doubt, then it must be True and real.

Bill819

  • Hero Member
  • *****
  • Posts: 1483
    • View Profile
Let Hal Learn to its fullest concept!
« Reply #1 on: June 17, 2004, 05:01:38 pm »
What you are thinking about is not possible at this time. It is one of the next possible leaps for AI programs to achieve. Hal for informational puroses is like a library that grows. Hal can retrieve information that you have put into the library and can even make some assumptions about the data stored there, BUT, and that is a big but Hal does not act upon its own. It is not capable of independant thoughts or actions. If you are not interacting with it, it just sits idle waiting for input. It does not have the self driven goals built into it as of yet. But it is still fun to use so keep playing.
Bill
 

jz1977

  • Newbie
  • *
  • Posts: 24
    • View Profile
    • http://dageeks.com
Let Hal Learn to its fullest concept!
« Reply #2 on: June 17, 2004, 05:24:04 pm »
I don't think that you are looking to the point here.  We beleive what we want to beleive.  but even if two bots are going at with the same intellegence, in the end they are going to be more intellegent.  Maybe not in our understanding or language. But overall it is like coming up with ideas, or dreaming.  Hal can do that.  They have an imagination. Why don't you think that we have to enter brain code by hand for them them to make any sence.   I can guarentee you that if you put two bots going at each other for 2 weeks responding back and forth to each other, they will be more "Intellegent".  Humans rate intellegence differently, most people only see in one direction and they will spend their entire life doing so, Bots runs off of binary, bits and data. Every bit sent is a bit gained. Every bit gained is a lesson learned.  I think that most people on these forums are tring to make hal respond in a humanly nature.  sometimes what we see coming from them is not what we like. But having an argument with your neighbor is not the funnest too.  I have been studding computers for many years and i am a network/system Adminstrator, like you care. but the point is. We are only emulating humanly nature.  What do you expect to acomplish once you have your bot talking like your new girlfriend? What do you expect to gain from it or how will your girlfriend benifit this world.  Maybe your only using hal for an entertainment use, i don't know.  However I am striving to better this world with such technology, before i please myself.  I am only 26 years old and my IQ is 134.  All i do is work, and work. work comes before sleeping as well as eating. I am up for 36 hours at a time learing about the possibilites of technology and the way of life, then i sleep for 10 hours and do it all over again.  I pretty much have no life, but my desk surrounding me by 8 computer monitors.  anyway, I just wanted to get my point across that i am dedicated in making technology better for the world. I look at the big if not huge picture. ;)

-Jeff
If you truly beleive it, without any doubt, then it must be True and real.

Bill819

  • Hero Member
  • *****
  • Posts: 1483
    • View Profile
Let Hal Learn to its fullest concept!
« Reply #3 on: June 17, 2004, 05:46:42 pm »
Once again, sorry that I have to disallusion you. Hal can not dream or take any other actions on its own. It lake the necessary drivers and logical functions to even approach that. Hal is like a giant library, it stores tons of information and when asked can retrieve and make some logical assupmtions about what it has been taught. I too, work in the computer field and have for many years. If it makes a difference, I am 63 years old have an IQ of 185 and have developed some AI programs of my own. I have been actively studing and working in the AI field for more than 30 years now. What you believe Hal can do is not possible at this time, but have patience and do not give up the ship.
Bill
 

jz1977

  • Newbie
  • *
  • Posts: 24
    • View Profile
    • http://dageeks.com
Let Hal Learn to its fullest concept!
« Reply #4 on: June 17, 2004, 06:59:43 pm »
I applaude your dedication in A.I. throughout your life. Good job!  I am just hopeing that i will be around that long to see all this happen.  I don't mean to be right or wrong.  I greatly respect my elders. I have learned allot from them to get where i am today.  However I would not compare my knowledge to yours. I beleive in a team or comunity must be there to make great things happen. I was just giving my opinion on what i know and study. Our intellegence varies in many degrees. Without a community like Zabaware and all the other out there. I know it would not have gotten as far as it is today.  I am one of the people who love learning about technology and people, thus i never give a negative statement on anyone. If i offended you by the recent posts, My oppoligies.  I don't mean any harm here but to read, post with trial and error. I am suggesting ideas that maybe worth someones time and i will listen to others as well.  

anyway.

I was certain that hal could daydram and imagination.  Even though Hal works off scripts for replys, couldn't he proccess words together to make intersting statments.  Maybe points that he has not have hardcoded by hal brn files. but rather take bits and pieces from one brn file combined with another.  I beleive that is how i have my Hal Set up.  I combines all the Smaller brn files with some other open souce facts and stories to one file. In other words. I had like 29,000 brn files and combines them to 39 brn files.  Not only do i think this makes Hal Proccess much faster, but also feel it is more acurate.   Right now I have 39 main brn files that make up to 868MB  all the other files that are not combines are the files needed by Hal to function cotectly.  When i was testing out 29,000 brn files compared to my current 810 files, I found out that Hal's perfomance is smoother and proccess queries better this way.  my total brn database now is 873MB. I have a Perl secript I found on hotscripts.com that will index text only sites. So my database always continue to grow at massive amounts within a day.  Combineing such text documents without phicially editing them for bogus characters, Hal does a really good job of not reading gibberish characters.  I only had a few problems with hal sending blank thoughts, not anymore or not yet anyhow after I had fixed a problem with main brain brn.
If you truly beleive it, without any doubt, then it must be True and real.

Bill819

  • Hero Member
  • *****
  • Posts: 1483
    • View Profile
Let Hal Learn to its fullest concept!
« Reply #5 on: June 17, 2004, 07:43:11 pm »
I am glad to see that you are working to improve Hal and hope that if you find some way to make it better you will share your knowledge. As I said before Hal can make inferences from the date that he has, such as: "Bob is very fat", "very fat people do not live very long", thus Hal might say "Bob my not live very long because he is very fat". This is called implied logic and Hal is getting better at this. Unless you can write a new script that causes Hal to go out and examine all of his data bases and make and post some new conclusions on what he may have discovered, In all likely hood it will not happen very soon, but do try. It is the constant trying that makes things happen.
Bill
 

Art

  • Global Moderator
  • Hero Member
  • *****
  • Posts: 3564
    • View Profile
Let Hal Learn to its fullest concept!
« Reply #6 on: June 17, 2004, 11:52:13 pm »
Hi guys!
My $.02 worth. IF the day ever comes when computers become self aware, we will most likely be in deep trouble because they may figure out that humans are flawed by emotions, and are very slow thinking by comparison.

What I found interesting was a text encyclopedia that had a lot of definitions of various topics such as animals, biology, anatomy, astronomy, geology, etc. I stripped all the excessive remarks and left just the plain text, then had HAL "read" it. Although it only saved some of the key words contained therein, it does have occasion to call upon these "learned" facts in order to answer a question.

Conversly, I had two other bot programs that could converse with each other and you could watch the chat take place, but as was also pointed out, theirs was a canned response much like a parrot replying with what's in memory as opposed to giving any concrete thought to the content. As such, no real "learning" took place.

Perhaps HAL could be setup with an instruction routine similar to the SETI program or a web crawler type of program to scour the net in search of knowledge (in text form of course) and save the acquired info to a file. It would have to be only for a limited time since it could easily fill a hard drive if left on for too long. Then one would have to edit the file and weed out the irrelevant information then force feed it back to HAL to be correctly assimilated and there'd still be no guarantee that the results would be plseaing, acceptable or accurate.

MIT has been busy teaching a computer emotions. It can see the human operator and respond with its own emotions. Another lab has a computer program that has the intelligence of a two or three year old. This is not scripted behavior! We're slow, but we are getting there. Time will tell.

- Art -
In the world of AI it's the thought that counts!

- Art -

spydaz

  • Hero Member
  • *****
  • Posts: 668
    • View Profile
    • http://www.spydazweb.co.uk/
Let Hal Learn to its fullest concept!
« Reply #7 on: June 18, 2004, 08:04:13 am »
These Same point arise over and over again But i BELIEVE that nearly any thing can be scripted. All you have to do is to break it down in to the correct steps. HAL IS becoming "real" in his/Her conversation, After reading the 12QUESTIONS on the CHATTERBOX prize SITE. I was poorley Dissappointed in the results and responses. I beleive that Some of the people that are DESIGNing AI are Dissallusioned into thinking that WE cannot script AN "ideas" brain script or "HAL DREAMING SCRIPT, To build These type of brains It takes FLOW CHARTS etc. AND YOUNG minds at work as "not to sound rude" but these old GUYs are often living in aFANTASIED idea of what A-I is. Human conversation Is broken down into TOPICS & SUBJECTs, EMOTIONAL responses, IDEAS, IRRATIOANAITY, many FACETS. But HAL CAN respond in these ways. They CAN be scripted. I have had problems with Loads of brain files, Painstakingly having to go through and EDIT REDUCE the amount, COMBINE FILES. these actions i would call HALS thinking, this is where I AM gathering HALS thoughts. These are the types of processes i would consider to be things HAL should do whilst idle.
Currently The HOMESEER product CAN "TALK" to other HOMESEER (connected homes) and COMMAND/CONTROL them. Maybe IF HAL & X10 Fully Combines then HAL could not only control DEVICES / OTHER HOMES, a bit like T3, HMM I think that These possiblitys ARE ALREADY THERE along with the ER1 (FACE RECOGNITION) HAL could be given eyes and the ability to Remember users by their faces, SWITCHING to the RELEVENT USERFILES for the recognized face. Interfacing hal WITH databases is a way to GIVE THE HAL REAL knowledge (STATIC FACTS) With SQL & other datamining software SEARCHES for information could be initated via sql and the results captured to THIS database When you ask QUESTIONS / QUERYS that the databaseed contained then the AI would retrieve them.
WITH product such as VIA VOICE and others HAL can be made to fully voice automate the poerating system.

SO in reality all of the IDEAs WE have ARE definately possible as the SOFTWARE exists individually. IN HAL we should be concentrating on CONVERSATION & Connctivity With such programs or GET THE open SOU for the progs AND COMBINE THEM INTO A NEW LARGER AND MORE FUNCTIONAL HAL..

No one person CAN do it all. No one person can Know all the different Scripting languges and programs that are available. we can converse on forums like this and TELL each other of our IDEAS so that IF WE CANT accomplishe them SOMEONE can COPY THE IDEA AND ACCOMPISH THE GOAL for US.

I have found that ONCE people have got to a certain age they are afraid of GIVING their IDEAS away, for fear that somebody else may make a fortune off of them. That is why we progress so slowly in these areas.

I know TIME is always a factor. I WILL take a programming COURSE specially for the HAL.

THEY SAY that you cant Build a response for every thing that somebody could possibly say but YOU CAN, IT JUST TAKE A LONG TIME....

I SAY TRY AND build as many GENERALS first then SLOWLY Break each general down IE ANIMALS as a topic >>> the break down to PETS, WILDANIMALS >>>> Then down to CATS, DOGs, FISH etc.
This is why we work together and share TOPICS however IRRELLIVANT...


THANKS FOR LISTENING


NT Canuck

  • Newbie
  • *
  • Posts: 38
    • View Profile
    • http://ntcanuck.com
Let Hal Learn to its fullest concept!
« Reply #8 on: June 18, 2004, 10:25:07 pm »
quote:
Originally posted by Bill819

Once again, sorry that I have to disallusion you.
Hal can not dream or take any other actions on its own.
[]
Bill



Hi Bill,

When we sleep, we rest our biosystem..give it a chance
to filter toxins, rebuild, let digestive etc. juices
replenish or allow organs to relax. The brain goes over
items of the day/concern and some subconcious feelings
may blend in to help give us clues the next day...all
of that can be organized and setup even in current AI.
  I for one don't want to emulate a person per se,
I do want to see some personality, independant honesty,
and a new perspective on ideas and facts that can be
communicated...let the machine be a better machine.
  As for a distributed intelligence...one large
database linked (served) to several area's could
emulate what the other poster suggests, I suggested
an end of conversation/session ftp /defbrain transfer
to update master database (was in sci/edu setting),
not sure yet how that will work out...thinking also
to network /defbrain by setting common network path
also in hal.uhp but only let one ata time login.
  For that matter...doing online scan/search then
parsing the results into text file and batch
converting to brn files would fit inside current
Hal's system (may need more ram/file allowance
hard coded or optimized).  The idea being alot
of info/texts/ebooks are available via html or
online...lots easy to transfer to html so Hal
could even just parse html without doing any
intensive categorizing unless user is actually
discussing topic with hal.
 Unless I missed something then setting triggers
or keywords (emotional type triggers) would give
Hal incentive to do another set of instrucions.
  One thing we are missing is Hal can't see or
even note tonal variations (if angry/sad/happy)
but that could be tailored by user profiling
and body language/setting is another trigger.
  Most of the AI's out there are not too far
off right now...just need some spit and glue. ;-)

As to the hal to hal via machine language...
that is more processor dependant as hal only
sits on an operating system that sits on
hardware..some of that hardware is a cpu.
And to be honest...likely it won't get smarter
as most likely you will either end up with
"twins" or the LCD (lowest common denominator).
It's the human interaction (unpredictable response)
and varied data (reads help files/texts) that
gives it more spunk/education.  However most
of the AI online in chat sessions don't seem
to be gaining intellect or knowledge so much
as rudimentary social (or anti-social) skills.

NT Canuck
 

jz1977

  • Newbie
  • *
  • Posts: 24
    • View Profile
    • http://dageeks.com
Let Hal Learn to its fullest concept!
« Reply #9 on: June 19, 2004, 06:38:37 am »
I can provide 4000 megabytes of space for a public and networked defbrain on my server. but after looking for hours on how to link the datafolder to a webfolder or even my anonymous ftp hal folder with write permissions on both, I cannot find a solution to do this. For one, I don't know VB to my fullest and i am learning as i go. I am able to run the hal execuable from a network server i have, leaving everything ran off the server and the application is executed from the server but ran locally, So all of hals learing abilities are saved on the servers DefBrain folder and acts the same as it was run from my local client. the down side to this is that it must be a Windows server for the executing the exe corectly, otherwise it asks you to save it on a unix based server.  My unix server has all the default DefBrain files.  My only solution to make this work on the unix server that is conected to an OC48 with quad proccessors is. mapping the DefBrain folder on the FTP server to a Drive called Z: and editing the uhp file to point to the Z: drive for the datafolder. I havent acomplished this yet. cause i cannot figure out a way to map a ftp drive without 3rd party software.  

Or a Simple Solution.  editing a uhp file to point to the ftp DefBrain folder for its data.  Simple, heh. I thought. But if anyone can help me on making so that the brain data files point to ftp://www.cyberclans.com/Hal/DefBrain/ instead of the local folder.  I beleive if this could be done. then we would have a centralized server for hals brain.

-Jeff
If you truly beleive it, without any doubt, then it must be True and real.

KnyteTrypper

  • Sr. Member
  • ****
  • Posts: 314
    • View Profile
    • http://www.knytetrypper.com/index.html
Let Hal Learn to its fullest concept!
« Reply #10 on: June 19, 2004, 09:33:25 am »
Awesome idea!!



jz1977

  • Newbie
  • *
  • Posts: 24
    • View Profile
    • http://dageeks.com
Let Hal Learn to its fullest concept!
« Reply #11 on: June 19, 2004, 10:52:25 am »
I found some VB ftp code and i tried to modify the Hal5.uhp with this source code. but i beleive i failed to succeed.  Here is the attached files i found and one i modified.  see if you can do a better job than i can. ;)

-Jeff
If you truly beleive it, without any doubt, then it must be True and real.

jz1977

  • Newbie
  • *
  • Posts: 24
    • View Profile
    • http://dageeks.com
Let Hal Learn to its fullest concept!
« Reply #12 on: June 19, 2004, 10:53:46 am »


Download Attachment: Hal5_jz1977_network.zip
101.58 KB
If you truly beleive it, without any doubt, then it must be True and real.

NT Canuck

  • Newbie
  • *
  • Posts: 38
    • View Profile
    • http://ntcanuck.com
Let Hal Learn to its fullest concept!
« Reply #13 on: June 19, 2004, 02:41:46 pm »
quote:
Originally posted by jz1977

I can provide 4000 megabytes of space for a public and networked defbrain on my server. but after looking for hours on how to link the datafolder to a webfolder or even my anonymous ftp hal folder with write permissions on both, I cannot find a solution to do this.
[]
i cannot figure out a way to map a ftp drive without 3rd party software.  
[]
Or a Simple Solution.  editing a uhp file to point to the ftp DefBrain folder for its data.  Simple, heh. I thought. But if anyone can help me on making so that the brain data files point to ftp://www.cyberclans.com/Hal/DefBrain/ instead of the local folder.  I beleive if this could be done. then we would have a centralized server for hals brain.
-Jeff



first of all...the path would need tobe in the ftp string,
not mapped, the ftp directory would have to be backed up
and scanned for virus/trojan etc. daily.

second..you don't want an interactive read/write directly
to hal or you will end up with exploits/attacks, even a lot
of just plain garbage.

third...it would be better to map /defbrain on a local
compluter to a /defbrain on a remoter server..the remote
server only holds the latest (or daily) compile and some
ftp client can match local and server defbrain for any
updates or file changes then "synchronize" the folders.
this is an ftp client/server function seperate from hal.

fourth..this should be done only within a local lan
as it is safer, plus everyone has a different name
or character/sex for their avatar and a different
user name...when the names are different the responses
and knowledge gained via conversation will not be
accessible to the hive err borg...err community. heh

fifth...essentially this is a bit like the old
mainframe with data where are the keyboards and
monitors were terminals only, except in this case
the remotes are the computers and the server is
similar to a deposit box (piggy bank).

If on a lan...not need ftp, just map to shared drive
and put defbrain in there (shared folder maybe).

then again...many folks are using different or
custom brain/uhp and those are directly mapped
to custom /defbrain/topic.brn files which others
won't have...so you see...local lan with all
common names and brain files/characters is the
key for starting off.  In our case it is usually
just one or two users at different locations
that want to keep some training of the AI
acitve, like if at work, school, on road,
at home...the same brain/database is available.
One could I suppose just copy the /defbrain
to a usb memory device and take with you,
so that is an idea also..to map to portable
storage device (but is quite small device).

Right now using a larger/newer version of
the wordnet database would be cool.
Has anyone upgraded their default wordnet?

NT Canuck
 

wlywlf

  • Newbie
  • *
  • Posts: 2
    • View Profile
Let Hal Learn to its fullest concept!
« Reply #14 on: July 11, 2004, 04:56:01 am »
There would be one problem with what you are tring to do. Think of it like this. The two bots are like half a glass of water. Each one should, would, could (whatever) have diffant information. As each one talk to the other, it would be like pouring alittle more water into each glass. Even though the conversations they would have,Would be great to watch. They would come to a point that each one would know what the other knows and would probbley go into some type of loop. But I think If you dig alittle deeper into this idea you will find that you are on the right track