Can artificial intelligences suffer from mental illness?

How do you generate a random number? Is it even possible?

 

Log in or Sign up to hide all adverts.

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

 

Log in or Sign up to hide all adverts.

Define random and prove a true random number generator exists.

Eventually you have to base it on some "random" event such as the arrival of the next cosmic ray. But isn't that the result of the laws of physics? It only seems random because our understanding of nature isn't good enough. If we knew enough, we could predict the probability of that arrival. And it would no longer be random.

The point is that we do not know whether or not there is any randomness in the universe. Therefore we can never build a mechanism that generates random numbers. We have no idea even if there is such a thing.

Of course we can produce numbers that we call random by virtue of our lack of knowledge. A coin toss is 50-50 because we don't know the exact flip force of your thumb, and the air pressure, and so forth. If you could control all those factors, a coin flip would be perfectly deterministic. I hope you agree with that. Coin flips are as deterministic as billiard balls. It's the domain of classical physics. No quantum weirdness.

Same with the arrival of the next cosmic ray in a detector. Maybe it's "really" random. Or maybe we just don't know enough physics.

The claim that a "random" quantum process is truly random is a metaphysical assumption; and not a fact of physics.

And getting back to the topic at large: If there is a single random process in the world; then the world can NEVER be the output of a computation of ANY type, finite or infinite. Because you are claiming that some aspect of the world is truly random!

Randomness is a very nihilistic assumption. If the arrival of a single cosmic ray at a detector is random; so is the ground beneath you.

 

Log in or Sign up to hide all adverts.

There are a few highly qualified maverick physicists that do believe that at 'rock bottom' everything is deterministic. Example; Gerard 't Hooft: http://arxiv.org/abs/1308.1007

Most though subscribe to the view quantum phenomena such as radioactive decay are inherently totally random. No experiment has invalidated that assumption - notwithstanding it's possible to modify e.g. decay rate via quantum Zeno effect: https://en.wikipedia.org/wiki/Quantum_Zeno_effect
But that's external meddling. The natural process is still truly random.

But what does any of this have to do with the human mind being subject to TM bounds or not? I agree more or less with some others here that consciousness is an emergent property. Of a sufficiently sophisticated implementation of an interconnected network of TM's. One that works fine without any need for perfect computational accuracy.
Sure just speculation. But the onus of disproof is on the part of anyone claiming perfect computational accuracy is somehow an essential ingredient of human mind/consciousness.

 

Interesting looking links. I'll check them out. Hopefully we can agree at least that the existence of true randomness is currently a matter of debate and not "settled science," as they say.

Because if there is anything that we can ALL agree a computation can NOT do, it's exhibit randomness. That's the one thing machines can't do. Someday we might have infinitary physics that lets us have infinitary computations. Someday we might discover that our subjective minds are nothing more than Turing machines. Anything is possible.

But surely nobody is going to claim that a machine running a program can produce a truly nondeterministic result.

So if there is true randomness in the world, the the world is not ANY kind of computation, finite or infinite. If there is true randomness, then the world can't operate according to rules, period. This seems to me to be a very big problem for anyone believing in randomness.

This seems very obvious to me so if there are differences of opinion I'd like to know what I'm missing. Randomness is the direct opposite of a computation or rules.

I have no problem with that.

A network of TMs doesn't have any more computational power than a TM as far as I know. And remember, by substrate independence, implementation does not matter. A sophisticated implementation computes exactly the same thing as a simple-minded implementation. The Euclidean algorithm to find the greatest common divisor of two integers works the same on a supercomputer or with pencil and paper. That's part of the essential nature of algorithms.

I apologize if I seem pedantic about this subject but I hope to promote clarity of thinking about what machines do. Implementations and networks won't help. Something else might bridge the gap to consciousness but it won't be an implementation or a network. To the best of my understanding.

Ok I have no problem with that. But you can no longer call it a computation. Call it something else. Sounds like you have a speculation about a "fuzzy computation." There are fuzzy sets and fuzzy logic, maybe someone's looked into this idea.

Oh ok! If I can just get people to separate what we know from what we don't know, that's a contribution.

Well that certainly isn't me. I can't even remember where I put my keys. And I don't believe minds are algorithms.

ps -- There are nondeterministic Turing machines. They're purely theoretical and I don't know anything about them. https://en.wikipedia.org/wiki/Non-deterministic_Turing_machine

Also a Google search on "fuzzy computation" and "fuzzy Turing machine" produced many links.

 

Truly random numbers now can be algorithmically generated according to:
http://phys.org/news/2016-05-method-random-cybersecurity.html
http://eccc.hpi-web.de/report/2015/119/
Even if that is slightly hyped, algorithmically generated true randomness 'fapp' has been possible for a long time. Why should it need to be perfectly random anyway?

Just surmise that, via pencil & paper, you could write down TM code(s) implemented in that sophisticated network of TM's, that managed the feat of reproducing internal sensation - say colour.
There is no way you as code writer would ever perceive that sensation via writing down those code lines while labouring away for how ever many years it took.
But with sufficient computational speed, the TM network would. So speed matters.
Think of it the other way around. A landscape painting is nowadays easily converted into a string of digits as say a png file. But you just reading that digitized string of 1's and 0's will never experience the sensation 'landscape painting'. But as a highly integrated network of TM's operating in some hierarchic feedback arrangement - you will see a landscape painting when those 1's and 0's are correctly reassembled on your PC screen. And yes maybe that requires fuzzy logic, maybe not.

 

When I wrote a tiny tiny program, in Visual Basic 6, to teach the names of the bones in the body, along with a test, 100 questions each 4 multi answer, 1 correct, 2 close, 1 wrong,. I needed a random number generator to randomly pick which position each of the answers should go.

From memory spreadsheet XL has a random number generator. I think I used it with a slight tweak and when I was happy it was giving me a random position for each of the questions I run it 100 times and wrote the test into the program placing the 4 answers to each question into the 4 positions the random number generator had selected for each of them

The person taking the test would write their answer in the answer box - click the CHECK button - box would turn green for a correct answer, red if incorrect

The 3 bones in the ear were a bit of a problem. I didn't want to restrict the answer to the sequence (ie from outer to inner) so I had to write a sub routine and / or logic so the order did not matter

Please Register or Log in to view the hidden image!

 

Are we sure that a particle has a duality? Would the OP question become easier to solve if a particle did not have this generally assumed dual quality?

Bohm proposed that a particle itself does not have a wave function, but follows the "guiding wave" in the Pilot Wave model

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

If Bohm was correct, would that solve at least part of the uncertainty problem?

It seems that transferring information involves the use of fixed values even if we cannot be sure of where these fixed values are "embedded" in the equation.

 

Most agree BM's predictions are identical to standard non-relativistic QM. Thus total randomness as physical outcome in QM events i.e. radioactive decay etc. is present in BM also.
And once again, why should perfect randomness be a requirement for human consciousness as a particularly sophisticated multi-level TM implementation?

 

Oh. I certainly do not believe in randomness in human consciousness. I would argue that in the abstract the two states are completely incompatible.
There must be a method to the madness......

Please Register or Log in to view the hidden image!

 

If the brain does not know what information it stores is correct and it does not know that it does not know that and it also does not know what it does not know and also it is subject to becoming defective without knowing and much more what else can it put out but random incomplete answers to random's incomplete inputs?

We are all mad only some of us hide it better

Please Register or Log in to view the hidden image!

 

If we don't understand what it is we are perceiving we call it a "miracle" or in scientific language, a "phenomenon". But this does not imply randomness, but ignorance of the forces involved.

This is why "learning" is essential in being able to interpret the input correctly.

An interesting experiment on kittens showed that if they were raised in an environment which only consisted of vertical objects and then placed in an environment which contained many small tables, the kittens easily avoided the vertical legs of the tables, but would bump their heads against the horizontal edge of the the table, clearly showing that their minds had no experience with horizontal obstacles and actually could not process this information, even as it was clearly visible.

However, after a few such accidental collisions, the kittens began to learn the concept of horizontal obstacles and from that point on would not only weave around the legs, but also would duck to avoid the edge of the tabletops.

IMO, that proves a non-random process of learning from experience.

 

I am not nearly as sure that we need to get "beyond" Turing machines. And I very much doubt that we need a new physics to explain the human mind - some better applied math, combinatorics over time, looks like a more urgent requirement. And of course much more progress in brain anatomy and behavior - biology.

1) They are measured attributes of physical phenomena in agreement with theory.
2) It wouldn't matter if they were mere speculation - only that they are plausible, actual non-aberrant possibilities. Your claim that it makes sense, as a thought experiment, to (for example) assign an exact position and velocity to a particle at an exact time, is far more speculative and notably less in agreement with observation and theory.

It will violate Bells' Inequality, which no predetermined sequence of cause and effect can do as long as Relativity holds (and current formal logic, in which statements are T/F/Meaningless).

It is certainly the case that it could be - possibly - emulated by one. Nothing about it precludes the possibility.

It's a direct demonstration that there are possible ways in which a universe which handles infinite length computations can nevertheless be emulated as a Turing machine taking steps.

That's all beside the point. We are talking about a computation that produces time - time is output, not input. Planck time would be a product of any computation that produced the universe.

In a finite number of steps.
Which leaves you in the odd position of being forced to acknowledge the possibility - the soundly based speculation - that a non-computable universe could nevertheless be emulated on a Turing machine.
I'm fine with that. It's the possibility of emulation that I need - that's how one could get mental illness in an artificial intelligence.

Our imperfect knowledge of the product of the computation (if any) does not tell us how it is done. That's hardly surprising.

Contemporary physics says nothing much about the subject.

In what way did my likening velocities and positions to temperatures and densities lead you to believe any of those attributes were "artifacts of our minds"? Of course in a sense all of observed reality is a mental creation, but the subcategory we term "real" and not "artifact" is very useful part of that creation, and densities, temperatures, velocities, etc, belong in it separate from "artifact". IMHO.

No, I'm arguing nothing of the kind - closer to the opposite, if anything.

 

Nor does it rule out randomness. It is what it is

Also I doubt you you could in essence not call the phenomenon random

You find out the forces involved. How far back do you trace THOSE forces before you come to something which you consider is NOT RANDOM?

Kittens randomly bumping heads into horizontal obstacles, of which they have had no previous experience, does indicate a lack of knowledge within the brain.

However as the brain learns the brain goes and operates the body in - don't go there mode and avoidence mode

It also writes in sub routines about lessons learnt from random bumping of head. Which I would consider / contend is turning a random event (the bumping of head) into a foreseeable event in situations involving horizontal obstacles AND OBSTACLES CLOSE TO HORIZONTAL

Where does the knowledge come from about OBSTACLES CLOSE TO HORIZONTAL?

My stand would be the brain is operating in best guess mode

And since it does not KNOW if its best guess mode is correct, ie not having been exposed to a 45° obstacle, it really is a random best guess. Until the brain can tick it of being the correct best guess

But here's a kicker. Will it ALWAYS be correct???

Please Register or Log in to view the hidden image!

 

If you can trace causality back, that means the phenomenon is not random. They may be complicated and practically undecipherable or predictable, but should in theory be computable.

Which was the case for the concepts of gods, IMO. Today we call the "wrath of Thor" by the name "thunderstorm" and while Thor has a long mythology, it does not explain how he created thunder and lightning. Today we know and the god Thor is dead.

 

Of course not. Above I cited and example of being wrong, by agreement.
That's why Anil Seth used the phrases "best guess", and "when our best guesses agree, we call it reality"

And his examples of optical illusions show that we are often wrong, because our MNS has cognitive limits, which sends the wrong information to our conscious part, which then makes an incorrect "best guess".

 

? Thor is dead

Please Register or Log in to view the hidden image!

Please Register or Log in to view the hidden image!

? Why do we still have thunderstorms then?

Serious. I am sure you cannot trace the brains output impulses of the brain which caused the body to act the way it acted back to the the originating impulse

In part because the original impulse has already been written over by the time the body acts

To reserrect THAT impulse (or at least a carbon copy or clone) is I contend is extremely hard

And this is hard because you now do not have the original circumstances or the original brain programs

Really no way for the Eureka moment "Ah THAT'S why the body acted the way it did"

Please Register or Log in to view the hidden image!

 

These look like interesting links. I'll add them to my reading queue, which gets longer every day. However I don't need to read these articles in order to know that they can't possibly prove what you claim. That's because nobody can generate perfectly random numbers, for two fundamental reasons.

1) We don't even know, and can't ever prove, that there even are any truly random processes in the world.

2) And even if there are, any physical implementation would be subject to design bias and physical error.

So we could never have a mechanism that generates truly random numbers.

I'm so happy you agree!

That means something else, at least in the US. Better to use a different acronym.

Ah. Because Michael 345 suggested using random numbers to test reality.

It seems to me that for this purpose, "almost random" or "sort of random" numbers won't do. So I asked:

That's when you jumped in. I certainly agree that for most practical purposes, "sort of" random processes work just fine. Coin-flipping is a common example. And taking the low-order bit of the femtosecond timestamp of the next cosmic ray to hit a particular detector is another. Neither are random. Both are "good enough" random.

But Michael 345 wants to test physics. And for that, we need true randomness. Of which there is no such thing.

Yes truly. This is one of my favorite arguments against the computational theory of mind. If a mind is a computation, then it can in principle be carried out using pencil and paper. So if I am sitting at a desk day in and day out, tediously executing the "mind" algorithm one instruction at a time, exactly the way a supercomputer does, only slower; I would therefore create or instantiate a mind. And where would that mind live? Inside my own mind? In disembodied space? In Searle's Chinese room?

Perfectly possible. Run a computation fast enough, and it does something qualitatively different from when you run it slowly. It has "emergent" properties. We often hear this argument.

For all I know, it's true. But one thing I am certain of. Whatever those emergent qualities are that arise from running a given algorithm faster; those qualities can not be computational. Because it's inherent in the nature of computation that implementation details and speed do not matter. We would be quite surprised if running the Euclidean algorithm on a supercomputer not only found GCD's, but also played chess. This would be astonishing. It does not happen.

But you (and many others) say that when we run an algorithm faster, it DOES do new things like develop a mind. If it does, then mind is not computational. Because if it were computational, then mind would be produced by the algorithm at very low speed as well.

Of course. Yes. Intentionality. Searle's point of veiw. It's we humans who give meaning, or "aboutness," to the bitstrings. Bitstrings don't experience anything. And computers flipping bits do not experience anything, or give meaning to the bits. Minds give meaning to bitstrings.

Don't know. Nobody knows. But it can't be computational. If the organization and speed of a computation make a difference, that difference is not computational. It must be something else. And exactly what is that something else?

 

Hopefully by now you've read my responses to Q-reeus, which shed light on the subject of randomness.

XL uses what's called a pseudo-random number generator. These give you a sequence of numbers that pass every known statistical test for randomness; yet they are perfectly deterministic. https://en.wikipedia.org/wiki/Pseudorandom_number_generator

As a really simple example, suppose (as is widely believed but still not proved) that pi is a normal number. Saying pi is normal means that its digits are perfectly distributed in the sense that as the number of digits goes to infinity, each digit occurs exactly 1/10 of the time; and each digit-pair occurs exactly 1/100 of the time; and each digit-triple occurs exactly 1/1000 of the time; and so forth.

Then we could use the digits of pi as a pseudo-random number generator, even though the digits are completely deterministic, and are exactly the same every time we look at them.

It seemed to me that you were proposing to test reality itself by throwing random numbers at an algorithm. For this purpose, wouldn't you need TRUE random numbers and not psuedo-random ones?

To sum this up, there are three levels of randomness:

* Sequences of numbers that are perfectly deterministic, but that satisfy statistical randomness, like the digits of pi or the pseudo-random number generators used in spreadsheets and programming languages;

* "Randomness by ignorance." We don't know enough about the thumb force and air pressure involved in a coin flip, so a coin flip is "effectively random" for most purposes. Likewise the low-order bit of the femtosecond timestamp of the arrival of a cosmic ray at a detector. This kind of idea is often proposed by people claiming to have implemented true randomness, but of course it's no such thing.

* True randomness. We don't know if it exists, or how we would physically implement it even if it did.

Those are pseudo-random. But in fact XL's algorithm for producing those is perfectly deterministic.

 

I'll take the opportunity to ask if Pi is a random number? It can only be approximated but always remains open ended ... .
Is that not why we call it an "irrational number"? Strangely, it seems to pop up everywhere and is not necessarily associated with circles only.