Now, *why* Turing-completness *wouldn't* be enough?

[vicky_molokh]

Greetings, all!

This is a tangent of a recently derailed thread. It is a question that relates to Ghosts and/or *AIs, but it does not relate to the sticky identity debate (any tangents related to it should be redirected there instead).

It has sometimes been said that the possibility of Ghosts* and AIs is superscience. Now, I think everybody agrees that no matter what, the process of scanning a brain or writing the AI is going to be pretty complicated. However, I wonder what basis there is for claiming that it is not just TL11+ instead of TL9½ as presented in THS, but actual outright superscience?

We know that Turing-complete machines can duplicate the behavior of any algorithm (program) taken from another machine. We also know that what is typically described as the mind is a self-modifying algorithm. We also have Turing-complete machines even today. The efficiency might be not enough to run a Ghost / SAI on a sufficiently small computer at a given TL, but the point of Turing-completeness is the ability to duplicate the processes of another hardware unit.

The most common argument is the postulation of the existence of Souls which perform some of the functions normally attributed to material hardware, combined with the postulation that a Soul cannot be scanned. The un-scan-ability is a very important point, because otherwise there will be lots of computer Soulless yet intelligent entities - Ghosts.

Still, even assuming un-scan-ability, there is no reason why a Turing-complete machine cannot be made that behaves indistinguishably from a 'soulful' mind. The creation of such an AI would be very difficult, working almost from scratch, essentially a hyper-advanced technology of solving the black box problem with something resembling a mind. However, it would still not require breaking any fundamental laws of nature.

It is interesting that there are arguments for the superscience-ness of AI / Ghosts from people who don't seem to believe in souls. It would be interesting to hear them explained in an organized form.

Thanks in advance!

* == Again, skip the debate whether the new Ghost is the same person as the old meat-, err, biosophont.

[Anthony]

Technically, any deterministic physical system that follows physical laws as we understand them can be modeled with a turing machine, but random (e.g. quantum) systems require the turing machine to have access to a source of randomness, and, of course, such an emulation will start to have random divergence. This isn't a real problem, it's not that technically difficult to have a pure noise generator, and as far as we can tell brain function isn't dependent on randomness anyway, but it's worth pointing out.

The major concerns with brainscanning, from a realism point, are that we don't really have a good idea how to scan a brain, and we don't have a good idea how large such a scan would have to be (presumably fewer than 10^26 bits, but if we can't compress that by quite a bit, the scan likely isn't practical). As far as we know, neither violates the laws of physics, so it's not TL^, but it might be higher TL than THS claims.

[vicky_molokh]

Originally Posted by Anthony

Technically, any deterministic physical system that follows physical laws as we understand them can be modeled with a turing machine, but random (e.g. quantum) systems require the turing machine to have access to a source of randomness, and, of course, such an emulation will start to have random divergence. This isn't a real problem, it's not that technically difficult to have a pure noise generator, and as far as we can tell brain function isn't dependent on randomness anyway, but it's worth pointing out.

That is indeed an issue I have considered in the previous discussion. Since those random variables are not dependent on the inputs provided to the machine (otherwise they wouldn't be variables), they are neither outputs nor derived internal variables. Which means they have to be treated for all intents and purposes as inputs. So yes, we do have to have a RNG substitute for such an emulation. The interesting thing is that even if we provide arbitrarily defined inputs into the supposedly random slots, we will still get a 'legit' emulation, because it is possible for the original system (with true or 'true' RNGs) to undergo a situation where those randoms match the ones we provide in the arbitrary example; sure, the probability of such an eldritch event might be unspeakably low, but it's still not impossible.

[sir_pudding]

Originally Posted by vicky_molokh

However, I wonder what basis there is for claiming that it is not just TL11+ instead of TL9½ as presented in THS, but actual outright superscience?

I'm fairly sure that this is a minority opinion in these here parts with only two adherents. One is a dualist and thinks something along the lines of "since know souls exist any technology that produces sapience without them is superscience".

The other I think is just arguing because he's ornery. :)

Outside of here, you have have a philosophical argument about whether you can ever prove a machine "thinks" or merely simulates "thinking". Which gives rise to the original meanings of hard/soft AI and thought experiments like the "Chinese Room". I get impression that most AI researches think the question is irrelevant. If you can't tell the difference between simulated thinking and real thinking, who cares? I'm not sure any one can prove that I can think (and I'm well aware that some people are pretty sure that I don't).

[Rendu]

Originally Posted by vicky_molokh

We also know that what is typically described as the mind is a self-modifying algorithm.

How exactly is this "known"?

[vicky_molokh]

Originally Posted by Rendu

How exactly is this "known"?

If this isn't true, we have to ditch Psychology, Neurology, and, in THS, Memetics. If a mind is not a finite list of 'instructions' for 'calculating' a 'function' (i.e. a set of inputs, and a set of outputs that depends on them), then what is it?

It is an interesting thing to question, but I'm not sure an alternative is even possible.

[sir_pudding]

Originally Posted by vicky_molokh

It is an interesting thing to question, but I'm not sure an alternative is even possible.

I tend to agree (as long as you don't start trying to use GURPS terms for real life again :) ) but the vast majority of people in the world do think otherwise.

[vicky_molokh]

Originally Posted by sir_pudding

I tend to agree (as long as you don't start trying to use GURPS terms for real life again :) ) but the vast majority of people in the world do think otherwise.

And this is the thread where I am interested in hearing them out. This is why I asked for organized explanations of their position(s) instead of the bits and pieces typically found in tangents.

[sir_pudding]

Originally Posted by vicky_molokh

And this is the thread where I am interested in hearing them out. This is why I asked for organized explanations of their position(s) instead of the bits and pieces typically found in tangents.

"They" in this case is combatmedic, AFAICT.

[vicky_molokh]

Originally Posted by sir_pudding

"They" in this case is combatmedic, AFAICT.

Aren't there more?

As far as I understand combatmedic's (rather coherent for this sort of discussions, I should add) position, souls store an essential part of the algorithm (and possibly are also are partly responsible for some other input, though this one is just speculation on my part) that can not be readily scanned (and cannot be imitated by AIs at TL10).

But then there's ZMC who thinks that souls are required for AIs (if I remember his position correctly), and other less recent protests against the possibility of infomorphs.