Penrose, and non computable action within the human brain..

JoeB-

Hi

Like the title says, this is about Penrose's idea (and proof!) that the action of the human brain can't be simulated algoritmathically.. i.e that no algorithim, or sequence of computable steps can fully describe human insight, or could match the capabilities of humans.

Is this correct?

Penrose uses Godels theorem to help him... he says that although Godels theorem says than no 'system of rules' can be self-complete... i.e there will always be proposals that are true within the system but which cannot be proved to be true in the system.. however human insight can allow us to 'see' the truth of a statement even if it cannot be proved true with the formal system.. and so we are somehow able to 'transcend' the limitations of 'formal rule' systems and gain insights which cannot be gained by computation.

If this is so then AI is doomed... not totally doomed of course but that if we merely run computable algorithims on a PC (or a quantum computer) we can never achieve the non-computable aspects which are (supposedly) proven to operate in the brain. This is regardless of any system of learning, or random imput, or pseudo random input, or combinations of 'top down' and 'bottom up' etc.. nothing will work.


So my specific question is, do people think that Penrose is correct, that the action of the human brain can't be simulated on a PC? (Regardless of any improvement). Or has he made a little error somewhere?

This would seem to me to be very important, if he is correct we should start looking for the non-computable action present in nature right now...

Cheers

Professor_Fink

Essentially, no, it is not a rigourous proof, and no, I don't believe there is any evidence that the brain cannot be simulated algorithmically, and there is plenty of evidence that it can. There is a very nice analysis and discussion here: http://www.scottaaronson.com/democritus/lec10.5.html

Morbert

JoaBallantine can never know that this statement is consistently true.

Professor_Fink

Morbert wrote: »

JoaBallantine can never know that this statement is consistently true.

Nice counter example.

JoeB-

I'm still not so sure...

While I think that it might be possible to simulate intelligence using a computer program I don't think there would be any real awareness or understanding there.

So I think there must be something extra to understanding, the whole conciousness thing isn't the result of a complicated algorithm...

Simulating intelligence would be possible because the human programmer will use his insights to limit the problems that must be solved by the program,.. and even still I don't think it would even be possible to pass the Turing Test, and even if that were done I would have to consider very carefully before I would allow for any 'real' understanding' or intelligence.

So I'm not sure an 'algorithm' can ever be said to (conciously) understand Godels theorem anyway (or even the concept of 'natural numbers'), in which case the whole point is moot...

Surprisingly I have sort of come to the conclusion that it may be impossible to build a robot that genuinely understands maths... a simulation of a hurricane is not a hurricane, and likewise a simulation of 'awareness' or 'understanding' is still only a simulation with no real understanding there at all!

Cheers

Professor_Fink

So, how do you know we aren't living in a simulation?

Morbert

JoeBallantine wrote: »

I'm still not so sure...

While I think that it might be possible to simulate intelligence using a computer program I don't think there would be any real awareness or understanding there.

So I think there must be something extra to understanding, the whole conciousness thing isn't the result of a complicated algorithm...

Simulating intelligence would be possible because the human programmer will use his insights to limit the problems that must be solved by the program,.. and even still I don't think it would even be possible to pass the Turing Test, and even if that were done I would have to consider very carefully before I would allow for any 'real' understanding' or intelligence.

So I'm not sure an 'algorithm' can ever be said to (conciously) understand Godels theorem anyway (or even the concept of 'natural numbers'), in which case the whole point is moot...

Surprisingly I have sort of come to the conclusion that it may be impossible to build a robot that genuinely understands maths... a simulation of a hurricane is not a hurricane, and likewise a simulation of 'awareness' or 'understanding' is still only a simulation with no real understanding there at all!

Cheers

The trouble is that awareness itself might simply be an emergent phenomenon; a side effect of 'big brain' evolutionary strategies. Reproduction and mutation has changed our intellectual limits over many generations, so our intelligence is certainly a wonder of nature, but we don't yet have a scientific reason to suppose that intelligence can't be artificially created. It will certainly be very difficult; the brain and mind are not analogous to hardware/software, and silicone might not be robust enough to accommodate neural pathways.

JoeB-

i don't know....


I don't think awareness is an emergent property, I think it's something else, something fundemental.

I didn't always think like this.. I used to think that it could be reproduced on a computer.. but I now think (know?) that this would only be a simulation, not true conciousness or awareness. (It might be possible to build a conciousness detector?)

But to say that conciousness is fundemental is a bit spiritual, like conciousness is 'out there' and our brains have evolved to tap into it. But this still feels better to me than to say it is simply a result of complicated computations.

I think it is a very hard problem.. and man may not make much progress even in a hundred centuries.. sure, we will make progress in mapping the brain, maybe even corralating brain states with feelings, or maybe even causing feelings by manipulating brain states!, but why it feels like something may always be a mystery.

It might still be possible to build concious machines without understanding why they work... (or possibly re-animating a dead brain, gory as it sounds)

But conciousness is something wonderful, I wonder what the limits would be for a machine specifically designed to achieve it.

I think this problem is like the reason why the laws of nature are there at all, why there is something rather than nothing? It is impossible to answer, or it feels impossible... but we can understand the laws of nature all we want, but will we ever be able to say why they exist at all?

(God is no answer to me by the way.. he just seems incredibly out of date, and he doesn't seem to fit the bill at all well.. just having a universe that exists is enough)

I think Penrose is on to something.. difficult for me to understand, but I'd say a mathemathical robot would understand him, and the various coding strategies for specifing machines, systems and statements, and so a robot like that may never accept that a given algorythim underlies its own behaviour.. this is a contradiction or leads to an absurdity.. and so such a algorithimically based robot can never exist. Sounds plausible to me.

But maybe we are algorithmic but our controlling algorthim is so complex it could never be specified? I don't really believe this but then I don't know what I believe...

I used to think it was impossible to determine if we were living in a simulated universe... but that was before I came across the idea of non-computable things.. that a universe could be completely determinstic yet practically, and in principle, completely unforecastable.

Such a universe couldn't be simulated I don't think..

Professor_Fink

JoeBallantine wrote: »

I don't think awareness is an emergent property, I think it's something else, something fundemental.

The problem is of course that there is absolutely no evidence to support that view.

JoeBallantine wrote: »

I didn't always think like this.. I used to think that it could be reproduced on a computer.. but I now think (know?) that this would only be a simulation, not true conciousness or awareness. (It might be possible to build a conciousness detector?)

Well, how do you define consciousness? A sufficiently detailed simulation would of course be able to replicate the functions of the brain, and so should answer questions and respond to stimuli in the same way as a 'concious' human. So what leads you to believe that it is somehow less concious?

JoeBallantine wrote: »

sure, we will make progress in mapping the brain, maybe even corralating brain states with feelings, or maybe even causing feelings by manipulating brain states!, but why it feels like something may always be a mystery.

Actually, we can already do that rather crudely. We know what regions are responsible for different processes, and for many, we have a fair idea of the evolutionary path they followed. In fact, we can to a certain extent understand why some emotions "feel" the way they do. For example, we know that seratonin is important in regulating certain moods.

JoeBallantine wrote: »

But conciousness is something wonderful, I wonder what the limits would be for a machine specifically designed to achieve it.

You seem to be deliberately constructing a definition of conciousness that includes a "is not part of a simulation clause", which seems both rather arbitrary, unnatural and also untestable.

JoeBallantine wrote: »

I used to think it was impossible to determine if we were living in a simulated universe...

It is.

JoeBallantine wrote: »

but that was before I came across the idea of non-computable things.. that a universe could be completely determinstic yet practically, and in principle, completely unforecastable.

Such a universe couldn't be simulated I don't think..

Not true. If the universe were to use non-computable numbers (for example) in the laws of physics, then it would be possible to build a device which tested these laws. Such a device would act as an oracle for calculating these numbers, so could be used in conjunction with a turing machine to expand it's computational power. Closed time-like curves are an example of a system which allows efficient computation of NP-complete problems (which I realise are computatble) and so can be used as an oracle with which to boost computational power.

The other thing you have to bear in mind, is that the universe running the simulation may have very different laws of nature than our universe, so we could not rely on observations about our own physics or computational powers to inform us about this second universe, which by definition must have more computational power than our own.

JoeB-

Professor_Fink wrote: »

Well, how do you define consciousness? A sufficiently detailed simulation would of course be able to replicate the functions of the brain, and so should answer questions and respond to stimuli in the same way as a 'concious' human. So what leads you to believe that it is somehow less concious?

Well, the fact that the complicated calculation could be carried out using paper and pen.. and it is unlikely the pen and paper are concious, or the system of the pen, paper and operator... so from where does the conciousness arise? Maybe there is a conciousness somewhere, a disembodied conciousness that is associated with an abacus for example.. so someone moves the beads and conciousness is created. But the conciousness has no mouth or body and so can't make itself known. Sounds a little dodge to me.

But maybe the speed of the calculation is important.. you need to hit a certain speed and complexity for conciousness to arise. Complexity can always be broken down into simple steps, so it is hard to see why conciousness would arise from it. I don't think the speed is any more important.. but what I do think is that maybe the method of running the computation may be very important.. i.e, our brains, which may operate on a quantum level... it is this quantum level interaction that allows for conciousness. As if conciousness is out there already maybe...

Of course I can't answer these questions.. they're fairly hard and mind bending. But I'd like to see the computation that results in a concious feeling of love.

Professor_Fink wrote: »

You seem to be deliberately constructing a definition of conciousness that includes a "is not part of a simulation clause", which seems both rather arbitrary, unnatural and also untestable.

It is probably also untestable to say that conciousness arises from calculation alone. But the point about simulations being carried out on paper, or on beads on a wire, or on a 'billiard ball' computer is good... it would be hard to attribute conciousness to beads on a wire, even if they passed the Turing test. And so, merely appearing to be concious isn't good enough. (You must be made of meat to appear concious at the moment, I'm a substance chauvanist)

Professor_Fink wrote: »

It is.


Not true. If the universe were to use non-computable numbers (for example) in the laws of physics, then it would be possible to build a device which tested these laws. Such a device would act as an oracle for calculating these numbers, so could be used in conjunction with a turing machine to expand it's computational power. Closed time-like curves are an example of a system which allows efficient computation of NP-complete problems (which I realise are computatble) and so can be used as an oracle with which to boost computational power.

I'm not certain about this, but I realise you may be and so I don't want to cause offence by saying you're wrong.. I just don't know. But if it were to be proved that conciousness doesn't arise from simulation or calculation (as I've done above, he he ) then my existence alone would be enough to prove (to me anyway) that the universe isn't simulated. (Unless of course I was simply a brain, and false inputs were fed into my brain, a la The Matrix?, But then I'd exist in a real universe, not a simulated one... as opposed to a simulation living in a simulated universe)

Professor_Fink wrote: »

The other thing you have to bear in mind, is that the universe running the simulation may have very different laws of nature than our universe, so we could not rely on observations about our own physics or computational powers to inform us about this second universe, which by definition must have more computational power than our own.

Interesting... and it lead me to think about simulating our own mini-toy-universe... of course we could use any laws we wanted, including random ones.. but I still don't think we'd be able to 'create' a conciousness within the simulated universe that would wonder and write poetry about the nature of its universe.. and wonder what happens when it dies?

Cheers

aequinoctium

Penrose explains his ideas in 'Shadows of the Mind'.
it's a good read. however i dont think he is taking it so seriously as to put his reputation on the line with it, just some musings.

Professor_Fink

JoeBallantine wrote: »

i.e, our brains, which may operate on a quantum level... it is this quantum level interaction that allows for conciousness. As if conciousness is out there already maybe...

Any quantum operation can be simulated by a classical computer (although not necessarily quickly. Strictly BQP (the class of problems efficently solvable with bounded error by a quantum computer) is a subset of the problems efficiently solvable by a non-deterministic turing machine in polynomial time (we call this an NP machine). Since any NP machine can be simulated by a computer in at most exponential time, it is always possible to simulate quantum systems.

JoeBallantine wrote: »

I'm not certain about this, but I realise you may be and so I don't want to cause offence by saying you're wrong.. I just don't know. But if it were to be proved that conciousness doesn't arise from simulation or calculation (as I've done above, he he ) then my existence alone would be enough to prove (to me anyway) that the universe isn't simulated.

You can't prove that conjecture, because it isn't true.

JoeBallantine wrote: »

Interesting... and it lead me to think about simulating our own mini-toy-universe... of course we could use any laws we wanted, including random ones.. but I still don't think we'd be able to 'create' a conciousness within the simulated universe that would wonder and write poetry about the nature of its universe.. and wonder what happens when it dies?

Cheers

Take a look at Polyworld, which is pretty much the kind of simulation you are talking about. There is a fantastic google tech talk about it.