Artificial Intelligence

Out of context: Reply #77

  • Started
  • Last post
  • 1,302 Responses
  • Krassy1

    Any QBNrs here who work in/with AI?

    Krassy 1Permalink
    UpvoteDownvote
    Flag
    • I coded a "deep belief net" in 2008-2009 from G Hintons paper. I wanted to predict horse racing results. To train the model I needed datadrgs
    • I tried to webscrape a horse racing website, but it turned out to be more complicated than the deep belief net, I gave up and forgot about itdrgs
    • Sounds cool!Krassy
    • But yeah, I was interested in hearing from someone with hands-on experience what the difference between straight up programming and AI actually practically is.Krassy
    • @drgs - Friend of mine did the same thing (he was a programmer working at the stock exchange). He had it going for a while but too much work to maintainGnash
    • it was making money for a whileGnash
    • These days there are multiple frameworks, so not much coding, but you need to understand the different types of models, where to apply which etcdrgs
    • And how to out them together. You need to account for what type of inputs you have, is it a vision problem, plain numbers, text etc, how deep/how man layers etcdrgs
    • Complex stuff, no doubt...... but just wondering at what point it becomes AI and not just a complex program of if/then/else logicKrassy
    • If you understand the decision tree and you can program it with if/then/else logic, then it's not really complex compared to problems where true AI is useddrgs
    • AI is a black box, which gives no explanation and no guarantee that the answer is correct. It just arrives at some solution by itself after you dumpdrgs
    • GBs of data on it. The "solution", or approximation what it really is, is spread across millions of parameters which are stored in those deep learning layersdrgs
    • still sounds like a complex program of rules and logic that yes, allow to arrive at various destinations.Krassy
    • So the millions of parameters thrown at it make it AI?Krassy
    • tnx for explaining, BTW.Krassy
    • Not challenging what you're saying; just still need more clarity (and no, I don't expect to get such clarity in the notes of a QBN post. Ha!)Krassy
    • So basically, the main advantage is that AI can find hidden patterns in data, or sort of approximate a pattern, which otherwise are impossible to programdrgs
    • But the culprit is that you need a lot of data, which records every possible side of the problem you are trying to solvedrgs
    • got it! tnx drgsKrassy
    Show [[ numHiddenNotes ]] more notesAdd Note
    SaveCancel

View thread