AI Zone Admin Forum Add your forum
Researching A.I. and Chatbots
 
 

I’m writing a stage play about artificial intelligence as part of an MFA I’m working toward at Northwestern University, so I might have some questions to interject into ongoing discussion threads among more knowledgeable participants. Hopefully such questions won’t be too annoyingly basic.

I want to keep the play fairly realistic and grounded in the current state of AI technology, rather than depict the sci-fi version, in which AI’s are on the verge of human-like sentience with a penchant for world-domination. Specifically, there will be characters who disagree about whether to pursue a “weak AI” chatbot project or continue a decades-long, Cyc-like “strong AI” project, given opportunities in the area of mobile intelligent personal assistants like Siri, Jeannie and Speaktoit.

One question I have is, what are the practical or technological reasons chatbots don’t take more advantage of comprehensive semantic networks/common-sense knowledgebases like Cyc or ConceptNet? (as opposed to pattern matching) What would (or does) it look like to involve the “understanding” modeled in big knowledge bases?

Thanks in advance for any light you can throw on the subject.

 

 
  [ # 1 ]

I don’t see strong AI, happening for the public anytime soon.  I would love to have a battery of computers and a team like Watson’s, but that isn’t going to happen soon either. After all the money and resources on Watson, he still is far from becoming sentient.  Right now the most reliable and consistent bots are the so called weak AI. 
It’s probably just me, but I don’t see a big benefit in using Cyc or ConceptNet.  I think there is a lot yet to be done with AIML and other bots.

 

 
  [ # 2 ]
Victor Fanucchi - Mar 21, 2012:

[...] One question I have is, what are the practical or technological reasons chatbots don’t take more advantage of comprehensive semantic networks/common-sense knowledgebases like Cyc or ConceptNet?

The Mentifex AI Minds, in English at

http://www.scn.org/~mentifex/AiMind.html

and in Russian at

http://www.scn.org/~mentifex/Dushka.html

awaken to sentient life with a small, innate knowledge base (KB) and augment their knowledge with facts obtained from the human users. Each acqusition of a new fact involves the associative tagging of each pertinent concept to relate it to other concepts. Even a large ontology like Cyc would need to be ingested or “digested” one fact at a time. At some later date, it may become possible for AI Minds to “swallow up” entire comprehensive ontologies. Thanks for asking such a provocative question! -Arthur

 

 
  [ # 3 ]

Thanks for the reply, Patti. I’m very much interested in your point of view and the details thereof, so I’ve got some follow up questions… What would you do if you did have a battery of computers like Watson’s? Have more and bigger databases of input patterns and scripted responses? Or would those greater resources facilitate some other approach?

And what would you say is less reliable and consistent about strong-AI bots? And what do some people suppose (wrongly) to be the big benefit of using Cyc or ConceptNet?

I hope the above doesn’t sound like interrogation. Interrogative but not interrogational hopefully. Just interested in more detail.

Arthur, I’m going to take a look at those links. The one in English, at least. ( ;

Much appreciated!

 

 
  [ # 4 ]

CYC will forever be unable to add to its own knowledge base. It will always be dependent on humans.
If I had the resources, I would probably be more compulsive and neurotic than I am now. LOL

I would love to play around with something like the old Nick by Avi Sanders. It doesn’t work well with the newer versions of Windows but I had a lot of fun with this bot.  It would search the web for info that it found interesting.  It came up with some great thoughts, loved Shakespeare and tried to sell me watches.

http://www.freewebs.com/colonel720/artificialintelligence.htm

Learning bots are, in my opinion, the closest thing to strong AI.  They are fun but they just don’t communicate as well.  Chat with some of the bots at

http://www.chatterboxchallenge.com/

 

 
  [ # 5 ]

I found some of Nick’s old Shakespeare from Peter Coles site. http://www.be9.net/BJ/

user: Tell me, what news of Cawdor?
Nick: man that function
    Is smother’d in surmise, and nothing is
    But what is not.
user: And what function, pray, ‘tis that?
Nick: It cannot
    Be call’d our mother, but our grave; where nothing,
    But who knows nothing, is once seen to smile;
user: Our grave? Oh no, ‘tis a bitter prospect indeed!
Nick: Macbeth does murder sleep’, the innocent sleep,
    Sleep that
user: to sleep perchance to dream?
Nick: If it were done when ‘tis done, then ‘twere well
    It were done quickly: if the assassination
    Could trammel up the consequence, and catch
    With his surcease success; that but this blow
user: Were it best not to begin such desperate enterprise?
Nick: knows
user: but most likely is not telling.
Nick: We are resolved, my lord.
user: Who are these desperate confederates who are so resolved?
Nick: beware
user: methinks the walls have ears…
Nick: lay thou thy basis sure,
    For goodness dare not cheque

 

 
  [ # 6 ]
Arthur T Murray - Mar 21, 2012:

...awaken to sentient life…

Oh jeez! here he goes again…

 

 
  [ # 7 ]
Steve Worswick - Mar 21, 2012:
Arthur T Murray - Mar 21, 2012:

...awaken to sentient life…

Oh jeez! here he goes again…

Hold on a minute Steve- did you miss this key part in the OP?

“I’m writing a stage play about artificial intelligence…”

In all seriousness, I think Arthur has plenty to contribut to that type of discussion.

-cb

 

 
  [ # 8 ]

Hi Carl. Another key part in the OP is this:

“I want to keep the play fairly realistic and grounded in the current state of AI technology, rather than depict the sci-fi version, in which AI’s are on the verge of human-like sentience with a penchant for world-domination.”

I have nothing against speculative sci-fi (love William Gibson, Bruce Sterling, etc.), but that doesn’t happen to be what I’m after tonally in this play. There are some specific questions I have about use by chatbots of Cyc and ConceptNet-style knowledge bases. Sincerely, if there’s anything incoherent in the phrasing of my questions in the original post and follow up, I am open to hearing how and why. (If you can articulate an objection to those questions themselves, that might honestly be helpful.) Thanks! ( :

Victor

 

 
  [ # 9 ]

current state of AI technology, rather than depict the sci-fi version,

The current state might be a bit dull. There is no great controversy. No lurking danger of bots coming to get you.  You install ConceptNet and get a lot of rambling about a subject.

 

 
  [ # 10 ]

Excellent! I want dull!

Would you mind describing ConceptNet’s rambling a little? I’ve read a lot of chatbot logs and would be very interested in what happens if you install ConceptNet. I’d even be curious about how one installs it.

Congratulations on the CBC win, by the way. Bildgesmythe was remarkably on-topic with its answers.

 

 
  [ # 11 ]

ConceptNet like WordNet is a great data base. How well it would work would be the art of the programmer.  It is a downloadable file
http://csc.media.mit.edu/docs/conceptnet/install.html
The bot is still using text input by a human.  It is not making the connection’s by it’s self. The bot still won’t understand a child’s story with out being programmed to. 
The databases will not make a great bot, any more than living by a library will make me a genius. I can read and learn, but the I.Q. stays the same. 
A great data base is still smoke and mirrors giving the impression of intelligence.

edit
The closest I think I have found was Nick (when he used to work). You could hook up a web cam to him. I showed him a ball. entered ball, red ball, etc.  Then showed him a blue block, did the same, then a blue ball. He did say blue ball.  I thought that was fantastic.  I also found the old Daisy bot did some great things (she won’t work on a 64 bit anymore either). I actually dropped my keyboard one night with the things she came up with. Neither used any data base.

{edit}I fixed your Typo, Patti. smile - Dave{/edit}

 

 
  [ # 12 ]

In my opinion, Cyc’s mistake was in trying to tack on natural language as an afterthought. So you need “knowledge engineers” to “teach” it.

I think the same type of algorithms we come up with to handle natural language issues (such as: syntactic delimiters, synonymous ways of saying something, ambiguity…) will lead to AI; so we might as well deal with natural language from the start (instead of using knowledge representation formalisms as Cyc tries to do).

Which is not to say I think Cyc is useless, i would like to make a CycAgent to compete for positive feedback with my other agents :)

 

 
  login or register to react