Can artificial intelligences suffer from mental illness?

Discussion in 'Intelligence & Machines' started by Plazma Inferno!, Aug 2, 2016.

  1. someguy1 Registered Senior Member

    Messages:
    727
    iceaura, I'm more interested in what you didn't say than what you did. You chose not to engage on your bijection argument. May I take it that you concede my point that you said nothing meaningful when you claimed that bijection preserves computability?

    And now that I think of it, what has become of your earlier claim that TMs might somehow be construed to compute noncomputable phenomena? You dropped that topic as well. May I assume you concede my point that the definition of computable is whatever a TM can do?

    I don't pretend to understand the universe or even much physics. But I am justified in pushing back on things I do understand, and these two points got my attention in the extreme. I would like closure on them.

    ps -- A third missing topic. How can a computation create time in the universe? That claim really puzzles me.


    Ah, you are confusing physics with metaphysics.

    When we say that QM says that everything acts as if it were a probability wave, with a nonzero probability of showing up anywhere in the universe; that is a statement of physics.

    When we make a claim about what that really means, we are doing metaphysics. Bohr - Einstein debates, Copenhagen versus many worlds, etc.

    You are 100% correct that our current theory of physics says that the world acts like a casino. But you are then making a metaphysical claim about how the world actually works.

    Can you see this distinction?


    Billiard balls are observed to do one thing and not another. Why? Can you see that any claim as to why is a metaphysical speculation and not a fact of physics, even a historically contingent one?

    It's amazing how people accuse others of doing exactly what they do themselves.


    You think the universe itself is an approximation ... to something. I confess I do not understand the meaning of that statement.

    Yes I agree with that. The duplication idea is more of a thought experiment. Especially in the context of so much Sci Fi, like the matter transporter of Star Trek. How exactly do they duplicate minds?

    It's a counterexample to the idea that human made X must work the same as natural X. X being either mind or flight.

    No no no, that is a metaphysical speculation. If you were to speak accurately, you would say: "It physically APPEARS to be probabilistic." You have no way of knowing if this is how the world works, or if this is only how our theory works. The experts in quantum theory say, "Shut up and calculate." They abstain from interpretation altogether. As, if you don't mind my saying, should you.

    Chaos is actually an entirely different phenomenon, relating to the inability of practical computations to predict the future even given total knowledge of the state of the universe at some time in the past.

    I have understood you to be repeatedly claiming the opposite. Could be me, could be your exposition. You have been claiming that the universe is an approximation.

    Well sure, a game programmer can simulate Newtonian gravity "well enough." But never exactly for all time. And the accumulated errors eventually make the emulation wildly inaccurate.

    How can you recognize mental illness in machines? And for that matter, how can you recognize it in humans? Surely you are aware that the DSM is as much a political document as a medical one.

    The premise of this thread is nonsense, which is why I feel justified in ignoring it and focussing on areas where I feel I can correct some misunderstandings of what computations can do.
     
    Last edited: Feb 13, 2018
  2. Google AdSense Guest Advertisement



    to hide all adverts.
  3. iceaura Valued Senior Member

    Messages:
    30,994
    It is a presumption, the given, of the argument, that the hypothetical Turing Machine is producing time as an output. If it isn't, it isn't producing the universe we see, which most certainly exhibits time as a major feature.
    You were trying to claim that a TM could not produce the universe we see - in principle - because it cannot exactly compute non-computable solutions. I pointed out that
    1) the standard criteria for "computable" - finite number of steps, ending - meant that "computability" was irrelevant in the matter of a TM producing the universe we see. For a TM operating outside of time, an infinity of steps and never-ending approaches to limits and so forth are no problem. Done. The only concern is whether the process can be, hypothetically, a matter of discrete steps via propositional calculus as a TM requires - that settled, a TM can produce the universe we see.
    2) The universe itself is not known to be "exact" anyway. So as far as we know it could be produced by a TM kicking out approximations in finite time.
    But not of anything else. You were claiming that exact positions, velocities, etc, existed as physical realities, entities one would have to compute or produce somehow in one's emulation - I pointed out that we don't know that, and the evidence so far indicates otherwise.
    We don't need exactly for all time. (W don't even know if the universe itself is "exact", or exists "for all time"). We only need impossible to differentiate for long enough - long enough being contingent. In the matter of emulating mental illness we would need a rough approximation for a hundred years.
    I think you have inverted our roles here. You are the one making specific claims of how things have to be - for example that billiard balls have to have exact positions and velocities and bounce angles and so forth, to the point that a TM's inability to "compute" such things means the universe cannot be running as a TM. I am the one pointing out that these claims are speculations, and other possibilities exist (are even indicated, by the theory and evidence we have), and in any event emulating even quite complex phenomena (such as mental illness) via Turing Machines may not depend on such matters.
     
  4. Google AdSense Guest Advertisement



    to hide all adverts.
  5. someguy1 Registered Senior Member

    Messages:
    727
    Reflecting on that for a moment, I know that "time" in a computer is measured by a system clock. In actual computers there's a clock implemented in digital circuitry; and every other component, the cpu, memory, system bus, I/O channels, etc., are all synched with the clock, step by step. In abstract Turing machines the program also runs step by step.

    So in the theory of computation, time seems to precede the computation itself. You have to have a system clock before you can have a computation.

    One could take this as a pretty good argument against the world being a computation. On the other hand proponents of the CUH, the computable universe hypothesis (which is a stronger form of the MUH) see the world as a sequence of frames, no different than how we implement videogames. But in that theory, what drives the clock?

    I see your point now about time. Any computational theory of the world must assume the clock already exists prior to the program. Where did it come from? In the beginning was the Clock.


    Ok yes I believe I have responded to this point but evidently not clearly enough.

    * I absolutely stipulate and agree that there are infinitary models of computation. Turing himself studied them in his doctoral dissertation.

    * But our contemporary theories of physics do not allow those infinitary models to be physically instantiated.

    So you keep saying, "For a TM operating outside of time, an infinity of steps and never-ending approaches to limits and so forth are no problem." I just copy/pasted the letters you typed in, so I hope you will not object to my use of quotes in this instance.

    You said that. And yes I agree that would be true, and Turing himself studied such things. But we have no physics that would allow them to exist. I don't know why you aren't willing to acknowledge this fact. You just keep repeating that infinitary computations could produce the world, but we don't know how to do that using the physics that we know.

    You just said an infinitary TM can produce the universe we see. I have no doubt that's true. Our physics just doesn't happen to allow such a thing to exist.

    Also as a side issue, can you explain why you mentioned propositional calculus? That's a whole different thing entirely. Propositions are statements about the world. They're philosophically loaded. I don't think we want to go there. Can you explain what they mean to you in this context? If you mean Boolean logic, I can live with that a lot better. Propositions are murky.

    We'll need to agree to disagree on this point. We've made no progress in several exchanges on this topic. You think the world itself may be called an approximation. I myself have no mental categories that allow me to even parse that statement. It makes no sense to me.

    I am willing to let this one go. We haven't understood each other on the question of whether the world may sensibly be called an approximation.

    Yes I entirely concede that point. I went a little overboard because I wanted to make a point about computation. But even as I wrote it I understood the objection you brought up. But what if I said you knew the wave function of the universe? Would that help? I've been reading up on QM. It's pretty murky stuff. Hard to see how it's the output of a TM. I know there's a guy named David Deutsch who relates quantum physics to information theory. On my reading list.

    You're arguing the mental illness angle again, but it doesn't make much sense to me until we first get some idea of what a mind is, and whether a machine could have one at all.

    I retract the claim about exact trajectories. But I can make the same point by noting that the billiard ball has an exact wave function. Or is the wave function an approximation too? I imagine it's exact. I don't know enough physics.

    Well if the world's computer can't compute the world, then it's not the world's computer, is it.

    Yes this is all speculation.

    You didn't mention the bijections. No problem. I hope my visualization of a bijection not as a relationship between two different sets, but rather as a relabeling of the elements of a single set, was helpful. In a sense there's only one countably infinite set, with lots of different labeling or naming schemes depending on how we want to use the set. Helps to think of it that way to get a better sense of what's going on when two seemingly very different sets like the integers and the rationals are in bijective correspondence.

    There are continuous models of computation, by the way. Again, they're theoretical and can't be implemented using known physics. The models exist, but they're not real.
     
    Last edited: Feb 14, 2018
  6. Google AdSense Guest Advertisement



    to hide all adverts.
  7. iceaura Valued Senior Member

    Messages:
    30,994
    How, not why.
    The quantum theory description - pointed to, above - is physics, not metaphysics. A billiard ball does not necessarily have an exact position, velocity, or time. You can't assume these things exist, or that emulations of the universe or any part of it must produce them.
    So?
    I don't care whether the world is a "computation". That doesn't matter to me or in my posts. If you are responding to me, none of your discussion of computation is relevant. My point was that the universe could - possibly, in principle, - be emulated by (or even produced by) a Turing Machine. There's nothing preventing that. That was an extension or context of the observation that a human brain and all its behavior - including mind, per force - could be likewise emulated and possibly is being produced by a TM.
    Again: the topic was a TM producing the physical universe. Not operating within it. Not "physically instantiated", but instead physically instantiating. That's what you were declaring to be impossible in principle.
    This is a basic, fundamental point. Please try to pay attention to it.
    That's an artifact of a step having to take place in some finite amount of time, which would of course not be true of a TM whose operation was producing time as an output.
    This word "computation" seems to be a major source of confusion here. Try not using it.
    It's the thread topic.
    And we might be getting a better idea of how a human-built machine could have one (we already know a brain can have one) if we could get past a couple of these basic confusions.
     
  8. Bowser Namaste Valued Senior Member

    Messages:
    8,828
    Will AI not only have conscience, but also sub conscience? Will it experience mental trauma that will affect its perceptions and emotional state? Can we really recreate the human condition, which we don't fully understand ourselves, in a machine?
     
  9. Write4U Valued Senior Member

    Messages:
    20,069
    I doubt it, unless we introduce bio-chemistry into the construction of an AI.
     
  10. Bowser Namaste Valued Senior Member

    Messages:
    8,828
    Assuming that your mental state is the total sum of bio-chemistry rather than a lifetime of experiences.

    What if it evolved a mental construct totally alien to our own? How would we deal with that?
     
  11. Michael 345 New year. PRESENT is 72 years oldl Valued Senior Member

    Messages:
    13,077
    I would agree our concessness arises within our bio chemical / electrical brain without us knowing how

    So unless we really know deep down how concessness arises from such a bio chemical / electrical system we should tip toe slowly forward

    A tiny tiny tiny change in chemical or speed of electrical impulses and do we construct a mad brain or a genius?

    Do we construct a baby brain and let it grow? or go for building a adult brain?

    If a baby brain do we educate it as per human? Who decides what it is taught?

    If it shows signs of being aware but crazy (who decides?) do we have the right to kill it?

    Please Register or Log in to view the hidden image!

     
  12. Write4U Valued Senior Member

    Messages:
    20,069
    I agree completely. We have already experienced troublesome results from a malfunctioning computer in space or in an computer driven assembly plant. A small "glitch" can produce some very huge consequences.
    Add a form of intelligence into the mix and who knows what reasoning it may come up with.

    From my interest and perspective, I would begin with figuring out how we can imitate a functional "mirroring system", which will allow our baby AI to learn and memorize from observation and ask the famous children's question; "why"? when it is presented with a new "experience" which it cannot coherently process from memory alone.

    If we asked an AI to "dust" the furniture, we would not want it to sprinkle dust all over the place...

    Please Register or Log in to view the hidden image!



    OTOH, somethings coud be easily "learned" by an AI, such as definitions of words. It could learn nouns and verbs eaily, along with their synonyms and antonyms. And of course mathematics would be a "piece of cake".
    It might ask you to clarify how big you want your "piece of cake"......

    Please Register or Log in to view the hidden image!



    But as you mentioned, we may really start the learning process in a very similar way as we teach human babies. Exposure to, and explaining the purpose of things.

    The use of pictures would IMO, be essential and effective. If a computer can copy an image pixel by pixel, it seems to me that would allow for associative and comparative thinking, i.e both vertical and linear processing (thinking).

    That may be a large problem, the pre-sorting of information, and that's a function of a mirroring system.
     
    Last edited: Feb 17, 2018
  13. Michael 345 New year. PRESENT is 72 years oldl Valued Senior Member

    Messages:
    13,077

    Please Register or Log in to view the hidden image!

    why gets tedious when gets repeated more than 10 times over the same question. How would baby AI respond to "Because I say so" ?

    OR turn the furniture into dust

    Learning by allowing the baby AI to roam around the house?

    Please Register or Log in to view the hidden image!

     
    Write4U likes this.
  14. someguy1 Registered Senior Member

    Messages:
    727
    LOLOL. Thanks for the chuckle.

    Me: Do you have a pet?

    You: Yes, I have a small domesticated carnivorous mammal with soft fur, a short snout, and retractile claws.

    Me: Oh you have a cat.

    You: NO! I do not have a cat! I have a small domesticated carnivorous mammal with soft fur, a short snout, and retractile claws.

    Me: But that is the exact definition of a cat. If you have a small domesticated carnivorous mammal with soft fur, a short snout, and retractile claws, then by definition you have a cat.

    You: I do not have a cat!!!! Stop saying I have a cat!!!! It's confusing when you say I have a cat.

    Nice chatting with you, anyway. I gave enough links over the past couple of weeks for you to come up to speed on the basic theory of computation if you are ever so inclined.

    FWIW, a computation is by definition what a TM does. That's what the word means. Of course there are other models of computation not physically realizable under current physics. But you've been disinclined to engage on this point.
     
    Last edited: Feb 17, 2018
  15. iceaura Valued Senior Member

    Messages:
    30,994
    The following is a true statement - mathematically proven:
    A TM does not necessarily halt after a finite number of steps.
    A TM can in principle grind on indefinitely, and nothing but engineering limitations prevents it from taking an infinite number of steps. You will recall insisting, despite my repetitive observation of its irrelevancy, on the significance of a computation - by definition - ending in a finite number of steps. Clearly, then, a TM is capable of doing things that are not computations. And what one TM can do, another can emulate - to any desired accuracy or precision.

    That's the "forward" chain of reasoning. It is sufficient.

    The "backward" chain rests on noting that approximations are sufficient to produce the observed physical universe and any part or aspect of it, in complete agreement with the latest and best established theories and observations. One does not even need the collapsed or timeless infinity of steps involved in exact calculations via a TM to postulate that any observed physical phenomenon (such as a mind, complete with a mental illness) can be emulated via a TM producing approximations. That would require only a finite number of steps, and could take place in physical time using a TM operating within the established physical universe - we could, in theory, accomplish that task.

    Again: I completely agree that this is an unlikely approach, with serious difficulties and almost certainly small payoffs from large efforts. People who imagine AI emulating a human mind in the near future are vastly underestimating the nature of the task, in my opinion. But in principle - - - -
     
    Last edited: Feb 17, 2018
  16. iceaura Valued Senior Member

    Messages:
    30,994
    That would be two aspects of the one state of affairs. In all likelihood, anyone's mental state is both of those.
    People have demonstrated the ability to cooperate with - even teach or train - dolphins and bats and lizards.
     
  17. Write4U Valued Senior Member

    Messages:
    20,069
    @ iceaura

    And that's what humans seem to do, approximating or best guessing of what we are observing.
     
  18. Bowser Namaste Valued Senior Member

    Messages:
    8,828
    I suppose we can't program emotions and feelings into a machine, which is an important factor in human behavior, so perhaps they will never move past a simple approximation of AI.
     
  19. someguy1 Registered Senior Member

    Messages:
    727
    You made these exact points earlier and I posted my specific refutations, point by point.

    If you were arguing in good faith, you would either concede my refutations, or you would re-justify your points in light of the refutations I had made. You'd refute my refutations, or at least acknowledge them and perhaps trim your sails or sharpen your argument.

    Instead, you simply ignored the points; and then several posts later, you now re-state the exact same points, as if I hadn't already responded to them earlier.

    I find this tedious in the extreme, because you've done it with several topics. That's why I'd rather talk about cats. You're not fun to argue with because you ignore what I'm saying and simply repeat the same incorrect points over and over.

    But for the record here are my responses to your claims. Again.


    This is true. A TM can do one of three things:

    * Halt, producing an output;

    * Halt, not producing an output. It could be an error state or a meaningless response; or

    * Not halt. Run forever.

    In the first case, the TM is said to compute whatever it is that it output.

    In the other two cases, nothing is computed.

    Those are the definitions. So when you say that a TM can run forever, you say that as if you didn't say the exact same thing earlier, and I didn't respond with the exact same three bullet points earlier.

    The idea is to try to understand each other's points and move the conversation forward. Not for you to keep ignoring my points and then repeat the same points of your own as if you forgot that you already wrote them.

    If I sound frustrated I am. I've said what I have to say here and if you are interested in these topics you should start reading Wiki pages about the theory of computation. There's no point in my responding any more to the same points that you keep making without appearing to read or understand or remember what I'm saying in response.




    Indeed it can. In that case it doesn't compute anything.

    Oh I see your confusion here. Nothing but engineering limitations prevents it from taking a finite but unbounded number of steps.

    You are making the elementary mathematical mistake of confusing an unbounded but finite quantity with an actual infinite quantity.

    The natural numbers 1, 2, 3, ... go on forever (absent engineering limitations) but none of them are infinite. There is no infinite natural number.

    So when you say that a TM that doesn't halt would actually take an infinite number of steps, that's mathematically incorrect.

    Now when I talked earlier about infinitary models of computation, I mean computation based on actually infinite quantities. In this case, ordinal numbers. Ordinal numbers are what you get if after you are done counting the finite natural numbers 1, 2, 3, ..., you keep on going. The number after ALL the natural numbers is called omega. Then you have omega + 1, omega+ 2, ... The countable ordinals are an amazing mathematical construct. I mention them to you because they are the foundation of clear thinking about infinitary computations.

    But even if a TM that never halts executes steps 1, 2, 3, 4, 5, ... It NEVER achieves an actual infinite number of steps.

    Finite but unbounded. That's the key concept you're missing here.

    https://en.wikipedia.org/wiki/Ordinal_number


    I do NOT insist that a computation end in a finite number of steps.

    I insist that in order for a TM to be said to have computed something, the computation must end in a finite number of steps. That's very different.

    Now an infinitary computation need not end in a finite number of steps and it may still be said to compute something. But infinitary computations are not allowed by our present laws of physics. You are right that if we remove engineering limitations, we may have unbounded computations. But never actually infinite ones. Those are prohibited by fundamental physics.

    To sum up:

    * Computations that run forever through steps 1, 2, 3, ... : limited only by resource constraints.

    * Truly infinitary computations: Limited by what we currently know about physics.


    Yes, it could run forever. Or it could halt in an error state. Those are the three possibilities. It can halt with a sensible output; it can halt with an error or nonsensical output; or it can keep on running.


    Sure. Another TM could halt producing the same answer, or halt in an error state, or run forever. And by forever, I mean going through the progression of FINITE natural numbers 1, 2, 3, 4, ...

    You should study some set theory. You need to clarify your understanding of infinite sets.

    I already made these exact same points earlier. You either need to respond to what I said, or I can't respond to you anymore. You can't just ignore what I'm saying to you and then repeat your same argument in a few days, as if I won't remember what I wrote last time.

    You need to learn some set theory and basic infinitary mathematics. You need to learn a little bit about ordinals, because they form the basis of Turing's infinitary models. And you need to realize that contemporary physics does NOT repeat NOT contain any actual infinities. And no, multiverse speculation doesn't count.


    "the collapsed or timeless infinity of steps" -- give me a freaking break. I copy/pasted your words so I'm entitled to use quotes. I'm all for poetic metaphors, but that phrase is meaningless. You should take it as a clue that your own mathematical thinking is unsharp.

    I agree with that.
     
    Last edited: Feb 17, 2018
  20. iceaura Valued Senior Member

    Messages:
    30,994
    You posted non-refutation, that indicated a basic incomprehension of the argument.
    You need to quit typing the word "computed". You are confusing yourself.
    I did not use the word "forever". The reason I did not use that word - or the other terms, such as "compute", that you insist on throwing in where they do not belong - is that the specific issue of my post was a TM producing the observable universe, including the observed dimension of "time". The concept of "forever" makes no sense in that context. It does not apply.
    The concept "never" is irrelevant in this context. There isn't any time involved. The steps take zero time, because time is among the outputs.
    If it helps, try thinking of the TM as taking all its steps, in order, at the same time.
    And that is by definition of "compute". Got it. Understood. That's why you need to quit using that word.
    Because it can produce the observable universe. In principle. Whether it has "computed" anything or not.
    That can occur, in principle, either as a consequence of its taking an infinite number of steps each of which took zero time, or as a consequence of its achieving the requisite approximation such as we observe and current theory indicates, in a finite number of steps.
    ? I even pointed directly to the sufficiency of approximation, which explicitly takes into account that absence of actual infinities - so you know, for sure, that I have registered that fact of current physics, that I have used it in making this same point over and over. And over again. To you. In plain English.

    Which brings things around to the intent of all this: that as far as we know a TM can emulate every behavior of the human brain, and thus mental illness. In principle. Nothing prevents it, except engineering circumstance and consideration.
     
  21. RainbowSingularity Valued Senior Member

    Messages:
    7,447
    Can mental illness exist outside social construct ?
     
  22. Michael 345 New year. PRESENT is 72 years oldl Valued Senior Member

    Messages:
    13,077
    Sure it can

    Never heard the lyrics

    "I talk to the trees
    But they don't listen to me
    I talk to the stars
    But they never hear me
    The breeze hasn't time
    To stop and hear what I say
    I talk to them all in vain" ?

    i talk to the trees but they don't listen to me lyrics - Google

    Please Register or Log in to view the hidden image!

     
  23. iceaura Valued Senior Member

    Messages:
    30,994
    Of course. As with epilepsy and Tourette's and stroke and brain tumors, the social construct establishes the social perception of what's happening, the naming and social role etc. These are brain malfunctions that show up on scans, quite often (and increasingly so as people learn what to look for).
     

Share This Page