Vanilla 1.1.9 is a product of Lussumo. More Information: Documentation, Community Support.

    •  
      CommentAuthorAngus
    • CommentTimeJul 7th 2017
     
    Posted By: Andrew PalfreymanI place transparency as a very high priority


    So do we all. Jargon creeps into everything and has to be resisted. OTOH a vocabulary of art is to some extent an advance of the language to encompass concepts not previously covered. Sometimes you need specialists to deal with such things.

    There has been a commendable movement recently to write contracts and suchlike in the plainest way possible. When I read acts of parliament they are generally not particularly obscurantist. The legal gobbledegook you abhor is probably just bad legal practice.
    • CommentAuthorAsterix
    • CommentTimeJul 7th 2017
     
    Posted By: AngusPresumably judicial review boards and appellate courts. The question will be whether these are human institutions or AI institutions. It's all a process leading to us working for the Machine rather than for the Man.

    In the end, what difference does it really make, provided there is beer and skittles?


    How can one (AI or human) review a decision if the mechanism for the decision is concealed from inspection.

    The US has been here many times before, with bogus "labs" and "experts"; e.g.,

    https://www.theatlantic.com/politics/archive/2015/04/csi-is-a-lie/390897/
    http://www.cbsnews.com/news/annie-dookhan-chemist-at-mass-crime-lab-arrested-for-allegedly-mishandling-over-60000-samples/
    http://www.dydflaw.com/?t=11&la=209

    etc., ad nauseum.

    And a fair amount of evidence simply "disappears" from police premises, usually taken by those who work there.

    So will an AI whose inner workings is hidden by "trade secrets" be any better?
    •  
      CommentAuthorAngus
    • CommentTimeJul 7th 2017
     
    I'm not sure if we are actually disagreeing. The mechanism of human thought is already concealed so I can't see any difference from a concealed robot mechanism. The legal mechanics of human decision making are already public and I xon't see any reason why the legal mechanics of robot decieion macing couldn't also be public.
    •  
      CommentAuthorpcstru
    • CommentTimeJul 7th 2017
     
    Posted By: AngusThat's interesting. Are the inner workings of a human judge's mind as he decides a sentence any more accessible than the secrets of the AI algorithm?

    Well, you can ask the Judge and usually their deliberations and decisions are a matter of written record so they generally "show their working".


    Generally a judge would be asked to provide a rationalisation for what he does, and that might or might not actually clarify his thinking. It would probably be relatively simple to ask the AI to do the same thing.

    Hah! I think that is 100% wrong. The answer for the (Narrow) AI (which is what we have today) is that it made that decision because that is what it was trained to do. Any other 'answer' would just be an answer another part of the AI was trained to give. The real problem is perhaps the other way round, how do we know the Judges workings are any different?
    •  
      CommentAuthorAngus
    • CommentTimeJul 7th 2017
     
    That is essentially what I was trying to say. The AI I have in mind is a learning network. A set of rules type AI can be examined, bar silly issues like proprietary software.
    •  
      CommentAuthorpcstru
    • CommentTimeJul 7th 2017
     
    Posted By: AngusThat is essentially what I was trying to say. The AI I have in mind is a learning network.

    Right, so all that can be said is that the networks works like X (which can be represented as a mathematical formulae) and that they operate in two modes, a learning mode where data is used to weight the nodes and and an operational mode where the trained network process an input and produces an output. Unless we introduce some randomisation deliberately, all the processing is strictly deterministic.

    In some ways that is a better explanation that retains consistency in the face of, for example Lunch. i.e. when we have statistically significant evidence that Judges decisions are being skewed by concerns other than rational, legal decision making, how much store can we really put in those rationalisations anyway? At least the AI network will, once trained, be consistent.
    •  
      CommentAuthorAngus
    • CommentTimeJul 7th 2017
     
    Yes. And as I said, it could be trained also to give rationalisations for its judgements but you've no more reason to believe them than the rationalisations of real judges.
  1.  
    •  
      CommentAuthorAngus
    • CommentTimeJul 10th 2017
     
    Posted By: pcstruthe networks works like X (which can be represented as a mathematical formulae) and that they operate in two modes, a learning mode where data is used to weight the nodes and and an operational mode where the trained network process an input and produces an output.


    Is it known that X can be represented as a mathematical formula in practice? I suppose at root an AI is a deterministic machine, but after some level does it not enter the chaotic realm where you really can't predict the outcome because it is so sensitive to initial conditions? Like humans?
  2.  
    Whom are you calling chaotic?
    •  
      CommentAuthorAngus
    • CommentTimeJul 10th 2017
     
    It's a term of art. Like chiaroscuro.
  3.  
    Yeah well, just watch that casting of shade ok?
    I have a meat factory to run here
    •  
      CommentAuthorpcstru
    • CommentTimeJul 10th 2017
     
    Posted By: Angus
    Posted By: pcstruthe networks works like X (which can be represented as a mathematical formulae) and that they operate in two modes, a learning mode where data is used to weight the nodes and and an operational mode where the trained network process an input and produces an output.

    Is it known that X can be represented as a mathematical formula in practice?

    As in the algorithm, yes.

    I suppose at root an AI is a deterministic machine, but after some level does it not enter the chaotic realm where you really can't predict the outcome because it is so sensitive to initial conditions? Like humans?

    They are chaotic in that for small differences in the input, the outputs are 'unpredictable' (other than by running the algorithm). However, there is a convergence which should be robust in the face of small changes; which is to say that two slightly different networks trained to recognise tennis balls in a picture (or trained on slightly different data) should converge on the 'solution' of tennis ball recognition - but only one of them gives false positives when my friends Highland Terrier is in the frame.
    •  
      CommentAuthorpcstru
    • CommentTimeJul 13th 2017
     
  4.  
    •  
      CommentAuthorDuracell
    • CommentTimeJul 28th 2017
     
    •  
      CommentAuthorAngus
    • CommentTimeJul 28th 2017
     
    Posted By: Andrew Palfreyman7 Days of Artificial Intelligence


    "worshiped"??
  5.  
  6.  
    •  
      CommentAuthorpcstru
    • CommentTime2 days ago
     
    Musk still worried and thinks you should be too. I'm starting to tip into agreement with Musk.