dupa

Author Topic: Watsons got Hal beat - Machine learning  (Read 15535 times)

jbig808

  • Newbie
  • *
  • Posts: 11
    • View Profile
Watsons got Hal beat - Machine learning
« on: March 09, 2011, 05:21:33 pm »
I did a forum search and didnt find much but is any work being done on a machine learning plugin to take hal into the next level?
 

onthecuttingedge2005

  • Guest
Re: Watsons got Hal beat - Machine learning
« Reply #1 on: March 10, 2011, 10:29:32 am »
I did a forum search and didnt find much but is any work being done on a machine learning plugin to take hal into the next level?

Hal is an open ended program, Hal can do anything Watson can do, if you program it to things that Watson does.
Watson just does internet searches for pattern matches, I don't think Watson actual can carry on a conversation at this time like Hal can.

plug-ins can make just about anything possible.

I believe Rob already has done something like what Watson does for his web bot. Hal searches Bing for information and uses it in its conversations.

Jerry 8)

freddy888

  • Hero Member
  • *****
  • Posts: 1693
    • View Profile
    • AiDreams
Re: Watsons got Hal beat - Machine learning
« Reply #2 on: March 10, 2011, 12:12:23 pm »
Watson wasn't connected to the internet when playing Jeopardy Jerry, it was all done on location as far as I recall.  I doubt any of us have a machine in our house that could do what Watson does.  But I do agree that I wouldn't have much fun chatting to Watson.
« Last Edit: March 10, 2011, 12:14:31 pm by freddy888 »

onthecuttingedge2005

  • Guest
Re: Watsons got Hal beat - Machine learning
« Reply #3 on: March 10, 2011, 03:05:13 pm »
We should hook Hal up on Jeopardy using the same electronics and programing Watson has, now that would be special.

I would be proud to see Hal win.
Jerry 8)

freddy888

  • Hero Member
  • *****
  • Posts: 1693
    • View Profile
    • AiDreams
Re: Watsons got Hal beat - Machine learning
« Reply #4 on: March 10, 2011, 04:17:13 pm »
It certainly would. :)

I'm looking forward to seeing how this web based database for Hal is going to shape up.  It's fascinating what that could lead to.

lightspeed

  • Hero Member
  • *****
  • Posts: 6761
    • View Profile
Re: Watsons got Hal beat - Machine learning
« Reply #5 on: March 13, 2011, 11:17:25 am »
yes i to can't wait for the data base web to open up for hal to use i personally beleive this is going to be another "best" turning point for hal in developement wise , another would be the virtual female body that was chosen before . now if only their is a way to make poser etc. clothes work with the new body , now "that" would really make it a good selling point!
 

spydaz

  • Hero Member
  • *****
  • Posts: 670
    • View Profile
    • http://www.spydazweb.co.uk/
Re: Watsons got Hal beat - Machine learning
« Reply #6 on: March 17, 2011, 02:40:40 pm »
Watson wasn't connected to the internet when playing Jeopardy Jerry, it was all done on location as far as I recall.  I doubt any of us have a machine in our house that could do what Watson does.  But I do agree that I wouldn't have much fun chatting to Watson.

But i do believe that If i was going on Jeopardy with hal i would program every previous Shows Questions into My brain specifically.... (QA_BRAIN)....REVERSED! A quite simple logic, as i think most of our AI_Brains search the Question and Answer TABLE BOTH WAYS......But a specific Table Just for jeopardy would be created so that new questions from "BOARD GAME VERSION" could be added On-the-fly....

As with a Trivial Pursuit brain Vrossi's Choiceplugin Combined With Every version of Trivial pusuits Answers would make HAl A Genius.... OFten just inputting data can be a long process but it pays off dividens.... as once that data is stored you can do so much with it..... Really These are the types of Data we need to be Sharing (KNOWLEDGE MODULES)

We need a way to HAVE Transferable DB files for HAL.....This would enable us to make the Knowledege base files for sharing.....
Counrty data / Phonebooks / Encyclopedias / Game questions / Formulaes / Sci Data / Scripts from movies..... These would enhance hal in the way in which we see in movies now our ai would start to take on a personality based upon the Scripts / plugins and Knowledge bases involved..... Along with BASIC DB files with "Slang Based" Chat ie: Cockney Basic brain / Localized american slang based brain (Ghetto HOOD, POSH UPTOWN, BUSINESS TYPE) Etc.... This would contribute to the AI basic Speach patern. not affecting his Knowledge plugins


Ideas anyway

freddy888

  • Hero Member
  • *****
  • Posts: 1693
    • View Profile
    • AiDreams
Re: Watsons got Hal beat - Machine learning
« Reply #7 on: March 17, 2011, 03:11:11 pm »
Yeah I think you are probably right Spydaz.  I think the worst enemy is time really - it would take a long time to assemble enough data I bet.

I guess IBM put a lot of man hours in to get Watson to know so much.  Did they do it with data mining ?

Maybe the big central database that Robert is working on will eventually be something like that.

But I do think you would need a pretty fast computer to answer as quickly as Watson did.  That's a lot of data.

Hmm maybe Rob could invest in one of IBM's Blue Gene computers to run the central database ;)

Ideas are good :)
« Last Edit: March 17, 2011, 03:25:13 pm by freddy888 »

spydaz

  • Hero Member
  • *****
  • Posts: 670
    • View Profile
    • http://www.spydazweb.co.uk/
Re: Watsons got Hal beat - Machine learning
« Reply #8 on: March 20, 2011, 07:26:41 am »
http://www.youtube.com/watch?v=an9EtgPeeqQ

This is the new tesla card and if you built a Quick super computer with one/two of these this would be ideal to run a centralized database. this is the Cheepest way  to put 1000 processors in 1 machine (to get the most from these processors you need to also build some plugins with CUDA an nvidia C++ coding platform to utilize PARALLEL PROCESSING POWER.... (1PC calculated to be about £5,000 but worth about £15,000 (with adobe cs5 / Cubase) which already use the parralell processing features... Also must be 64bit windows 7 or Windows 64bit windows server 2008 R2, which also can utilize all the features of 1000 GPU processors

http://content.dell.com/us/en/enterprise/d/hpcc/cray-cx1iws-dell.aspx <this is Dells attempt to supply simular power> (not Quite) they still want you to RENT SPACE @ cray computing (supercomputers for RENT)(LONG QUE)
« Last Edit: March 20, 2011, 07:30:49 am by spydaz »

lightspeed

  • Hero Member
  • *****
  • Posts: 6761
    • View Profile
Re: Watsons got Hal beat - Machine learning
« Reply #9 on: March 20, 2011, 09:29:45 am »
i agree with freddy888 about having to have a super fast computer and if people had to have a super fast (read expensive ) computer then many could not afford to use hal etc. because of the cost and that would be a shame because not only would new people not be able to have and run hal but even some older people who have had hal for a while would have to stop using new versions of hal.
i think spydaz is also onto something about the data base having a universal where it could be implimented like a plug in for hal to know things , this would definatly speed up hal learning ability .
 even if we had a straight learning process of words like i was mentioning hal as a doctor info from webmd as a source . with the right script instead of hal just sayying a straight script answer hal could add some words to it to sound more as a person talking instead of just reading a script . this was what i was planning on doing as a plug in about health questions .

 :)
 

spydaz

  • Hero Member
  • *****
  • Posts: 670
    • View Profile
    • http://www.spydazweb.co.uk/
Re: Watsons got Hal beat - Machine learning
« Reply #10 on: March 21, 2011, 07:00:26 am »
I think lightspeed. That these Current plugins when sharing "DATA" We need to share in a "Comma delimited file" this enables anyone with a database engine to import data into any database. with comma separated files feild length is not a problem.
Although if you are using a "MEMO" type feild a different Character such a # maybe better......


lightspeed

  • Hero Member
  • *****
  • Posts: 6761
    • View Profile
Re: Watsons got Hal beat - Machine learning
« Reply #11 on: March 21, 2011, 09:54:34 am »
spydaz
i like your work among others on here can you give me one good example that i can go by a plug in example . (if you have the time that is .)  :)
 

Ooglor 7

  • Sr. Member
  • ****
  • Posts: 369
    • View Profile
Re: Watsons got Hal beat - Machine learning
« Reply #12 on: April 07, 2011, 04:44:23 pm »
I think that saying Hal could in any way keep up with what it takes to perform and win like Watson did is hugely over estimating the program's capabilities.
« Last Edit: April 07, 2011, 04:46:25 pm by Ooglor 7 »
When a child is born without a soul...

echoman

  • Guest
Re: Watsons got Hal beat - Machine learning
« Reply #13 on: April 07, 2011, 05:36:16 pm »
It might be interesting watching various AI's competing against each other at Jeopardy as is the case with chess computer program competitions.

spydaz

  • Hero Member
  • *****
  • Posts: 670
    • View Profile
    • http://www.spydazweb.co.uk/
Re: Watsons got Hal beat - Machine learning
« Reply #14 on: April 07, 2011, 05:45:10 pm »
Ooglor 7, Hal is only limited by USER PROGRAMMING!