I simply said it's because it's too complex for a human to keep in the brain all at once, but that doesn't mean it can't be stepped through. you're not really denying that, are you.
That isn't what you "simply" said:
"programmer, this is step debugger. step debugger, this is programmer.
I'll leave you two alone now, it seems you have a lot to catch up on..."
that's not what a step debugger is. a step debugger lets you run the programm step by step, while inspecting its state. you can watch it grab input, and how *exactly* it reacts to it and why.
The term "step debugger"and the idea of running a complex ANN "step by step" have nothing to do with ANN's, nor is it possible to use a step debugger or to go through the program step by step and anyone with even a BASIC familiarity with neural network programming would know that. Instead of bothering to check up on what I was talking about, you posted your flippant response and then made it worse by exapanding on it, continuing to apply concepts from programming in general which don't apply here.
ANNs do not use explicit algorithms which allow one to run the program step by step. Every time a complex ANN is run, the input neurons communicate with perhaps multiple hidden layers of neurons until the information reaches the output neurons and a final response. Each run results in an adjusting of weights according to the intial complex nonlinear algorithms. However, the code never specificies how the weights are adjusted after or during each run, nor is it possible to "stop" the program and "see" what connections led to this or that weight change or this or that final output.
Before you make anymore claims such as
and yes, I do actually know what I'm talking about
make sure you do. Because you continue to apply traditional programming logic to a field designed specifically to avoid that "step-by-step" process, and our inability to know the system trajectory in some cases, or to know how the result was derived, has nothing to do with programmers being unable to "step through" the code. And while your debugger approach is standard just about everywhere else in programming, if you knew what you were talking about, you'd know it doesn't work here, and you wouldn't have made reference to step debebugging.
turing machines usually work in a deterministic fashion unless some non-deterministic element is introduced.
Again, this only demonstrates that you are completely unfamiliar with artificial neural network programming, but rather than retract the rude comments you made you'd prefer to just dig yourself a deeper grave.
First, an actual "turing machine" requires and infinite length of tape, and one of the points (or results) was to strike another blow against Hilberts dream (which Turing did using Cantor's diagonal infinite proof method). But more importantly, "Turing machines" outside of the theoretical concept and even within Turing's paper use formal and linear (which does not exclude loops) logic. ANN's are fundamentally different and deliberately so. They began within cognitive science and AI research after classical computing methods failed to create programs which could learn, adapt, and evolve in the way hoped.
Unlike other programs, even extremely complex ones, ANNs are designed to "write" their own code in a sense. They adapt to input in highly complex ways making it impossible for the programmer to always know how or why certain changes in weights resulted or why the output was what it was, and also to run through these changes "step-by-step" to find the answer.
This isn't saying that they are indeterministic (although again, that has been suggested), but it does mean that your mocking comments about how it's just a matter of using a step-by-step debugging approach means you don't understand how ANNs work.
"we haven't found evidence for it to be deterministic" doesn't equal "we have found even just the slightest indication that it isn't", either, so I find all of that kinda misleading. hence my sarcastic remark...
Only that isn't what I said:
in a paper "Investigation of the determinism of complex dynamical systems using simple back propagation neural networks" (International Journal of Computer Mathematic, 2006), the authors attempt to provide solutions for state determination for a particular type of ANN which, while under some circumstances is deterministic, under others "no such statement can be directly asserted."
If they say "under these circumstances it is deterministic, but under these we can't assert that it is" (which is exactly what they say) than yes, that does equal "we have found even just the slightest indication that it isn't." The whole reason to bring up determinism of this type of ANN was to note how under certain conditions it is deterministic, and under others that can't be said. If it can't be said, then they can't say that for a reason.
so you don't really think they're non-deterministic, you just enjoy the mystique of hinting at that they might be... lol?
I think the fact that we can now create programs which are so complex we have difficulty (and are sometimes unable) to understand how our own code evolved and adapted as it did, when these programs do not even get close to the complexity of the "mind" (and lack self-awareness, something we aren't even close to being able to exaplain), should give us cause to think that perhaps the view of the universe which has dominated human thought until it began to change in the 20th century is not accurate. Perhaps, given our tools to understand nonlinear system, and the fact that our focus is on solutions (rather than understanding why we can't explain what we can't explain), a causally deterministic view of the mind is inadequate.
no, not much beyond the basics, but unless they run on some kind of freaky new CPU nobody told me about, that's hardly relevant.
Given your mocking "step-by-step" solution to the problem faced by experts in mathematics, computer science, cognitive science, etc., you'd think that either they'd have figured out all they had to do was use debugging techniques everyone else has for the past several decades, or it is relevant and you don't know what you are talking about.