Wednesday, April 12, 2023

GOFAI

 Good Old Fashioned Artificial Intelligence (GOFAI)

Practice yourself, for heaven's sake in little things, and then proceed to greater. Share this Quote Epictetus
Read more at https://www.brainyquote.com/topics/little-things-quotes

Practice yourself, for heaven's sake in little things, and then proceed to greater.
- Epictetus
 
 
 
Practice yourself, for heaven's sake in little things, and then proceed to greater. Share this Quote Epictetus
Read more at https://www.brainyquote.com/topics/little-things-quotes
Practice yourself, for heaven's sake in little things, and then proceed to greater. Share this Quote Epictetus
Read more at https://www.brainyquote.com/topics/little-things-quotes
Practice yourself, for heaven's sake in little things, and then proceed to greater. Share this Quote Epictetus
Read more at https://www.brainyquote.com/topics/little-things-quotes
at

20 comments:

  1. SputnikApril 17, 2023 at 3:28 PM

    I think that symbolic logic and expert systems may make a comeback due to the need for AI to provide explanations and trace its thinking. LLMs are quite mysterious to the end user.

    ReplyDelete
  2. scidataApril 17, 2023 at 3:33 PM

    Plus, not everyone who wants to play with AI has access to a $100 million supercomputer.

    ReplyDelete
  3. scidataSeptember 18, 2023 at 2:47 PM

    Another thought about ChatGPT (and others like it)
    Debate rages along the spectrum from rigid AI 'purists' to ad hoc 'need a few paragraphs' users about what it actually is. The former dismisses it as predictive text mimicry, while the latter doesn't really care. I find myself gradually sliding from the former to the latter. There's an old French joke: "Sure it works in practice, but will it work in theory?"

    ReplyDelete
  4. scidataSeptember 19, 2023 at 2:19 PM

    Symbolic AI stumbled because it simply could not keep up with the tsunami of data that had to be labelled and curated. Knowledge bases of trillions of cells were simply not possible. Generative AI is not really 'artificial' (basically mimics human responses) nor 'intelligent' (no experience, no goals, no theory of mind). However, an old timey symbolic AI, capable of harnessing and directing a generative LLM, could be fun to watch.

    ReplyDelete
    Replies
    1. AnonymousNovember 22, 2023 at 2:28 PM

      Well...

      Lenat was bragging about millions.

      And frankly... hardly there need to be more.

      We people -- not that complex, at all.

      Delete
    2. scidataNovember 22, 2023 at 4:38 PM

      Yes, as a long-time fan of Jeff Hawkins (Palm, Numenta), I'm intrigued by the 'sparse network' concept. I suspect that we abandoned GOFAI way, way too soon.

      Delete
    3. AnonymousNovember 23, 2023 at 12:41 PM

      Well... without some good hypothesis -- we destined to re-create all evolution of a Brain/Mind... to re-create it.

      And it means trying, trying and trying again.

      Delete
  5. scidataNovember 20, 2023 at 12:45 PM

    OpenAI fires Sam Altman, Microsoft hires him. 500 OpenAI staffers say 'bring him back or we quit'. Another slow news day.

    ReplyDelete
  6. AnonymousNovember 24, 2023 at 12:47 PM

    Well... there is a caveat in all that Lenat's endevor -- who said that One Size Suits all?

    Most surely... different problems would need different KBs.

    So... instead of One... there need to be multitude of KB.

    I though about personal for each human... bunch of em, even.

    So every one of us could train his... and they could communicate too -- ask questions they cannot know themself. ;-)

    ReplyDelete
    Replies
    1. scidataNovember 24, 2023 at 2:43 PM

      Common sense emerges from 20 (or more) years of self-training a tiny model of reality (a human mind). These models (minds) interact (communicate) to form a society. That's very different from training a neural net, and very, very different from distributing such a trained NN in something like a self-driving car. Again, current AI is neither artificial nor intelligent.

      Delete
    2. AnonymousNovember 24, 2023 at 5:45 PM

      Ditto.

      "But we do not like to know that it don't work... we want to make it WORKING!!!" ;-)

      Delete
    3. AnonymousNovember 30, 2023 at 11:46 AM

      Well... common sense in its linguistic form.

      But basiest of it (like recognizing oneself in a mirror) -- is built-in cognitive ability. ;-)

      Delete
  7. AnonymousJanuary 2, 2024 at 1:49 PM

    Wanna share my accidental findings.

    https://writings.stephenwolfram.com/2023/02/what-is-chatgpt-doing-and-why-does-it-work/

    Good link, it seems. Showing that all that mumbo-jumbo about NNs... are not that different from regular juggling with bytes we do and love to do...

    It's just... it would be nearly to impossible for any average... or even any programmer at all, to get a hold of equipment to write such program and test it working....

    On the data pool like ALL WORDS in ALL TEXTS people have had written. ;-)

    ReplyDelete
    Replies
    1. scidataJanuary 2, 2024 at 7:49 PM

      Yes. I've written several things about concatenative programming, mainly about biology/evolution and mainly in the context of FORTH. The real magic in any of this is not AI, but rather COMPUTATION itself. The universe is a realm of computation and evolution. Tools like math allow a vastly simplified model of reality to be stored in a 3lb hominid brain. But that isn't easily monetized or exploited, so we have critters like ChatGPT from farms like OpenAI.
      //nearly to impossible for any average...//
      Don't fret too much. The lowly ribosome can do much the same thing, and a modern CPU chip can do much more than that. I can talk about this more...

      Delete
    2. AnonymousJanuary 3, 2024 at 12:34 PM

      Well. We have GPUs. And even FPGAs.

      But that is... not low hanging fruit at all. Especially when task is not to optimize existing algorithm. But frollicing with trying this or that.

      Remember -- Lenat's experience with Eurisco. Big part of hos success -- is using very expensive LISP machines...

      And now.... when progress in computing seems like stalled...

      Delete
    3. AnonymousJanuary 4, 2024 at 6:18 AM

      Means... raw power it's still raw power.

      What it would be, if we'd have PetaFLOP CBUs on our desks? In our pockets?

      Delete
    4. AnonymousJanuary 4, 2024 at 6:20 AM

      Also... you can explore that link to Wolfram site.

      That where he (trying to) explaining "ruliad". ;-)

      Idea that Universe is just and ultimately IS a COMUTATION. In and Out. ;-)

      Delete
    5. scidataJanuary 13, 2024 at 10:03 AM

      Thanks. We speak a similar language. A lot of my ideas about computation either come from Wolfram or were 'discovered' in the course of my own wanderings. Ruliad is mind-bending, and is one of the reasons I'm so hesitant to give up on GOFAI and embrace the neural network black box religion :)

      Delete
    6. AnonymousApril 18, 2024 at 10:23 AM

      Well... though that is TOTALLY opposite things... yawn.

      Delete

Subscribe to: Post Comments (Atom)

Science

    [Our research] has nothing to do directly with defending our country except to help make it worth defending ...

  • Forth
    The most important programming language of all time.  Forth posts by me  
  • GOFAI
     Good Old Fashioned Artificial Intelligence (GOFAI) Practice yourself, for heaven's sake in little things, and then proceed to greater. ...
  • Science
        [Our research] has nothing to do directly with defending our country except to help make it worth defending ...