Medium Dependent Intelligence, Uncopyable Intelligence, and Transhumanism

[AlexanderHowl]

Yes, I think that a human brain (and any digital brain whose physical structure mimics the human brain and whose underlying processes mimic the function of human nerve cells) would be by our current definition a hypercomputational device. We are capable of leaps of intuition that a Turing machine could not reasonably mimic. We are also capable of realtime 3D visualizations and realtime modifications of our 3D visualizations that I do not think that a Turning machine could mimic in a sensible fashion.

The advantage of the human brain is 500 million years of neural evolution that allows for a hypercomputational device that software cannot mimic. I do think that a physical digital brain, as in a brain that uses digital signals and digital storage rather than chemical signals and chemical storage, could also form a hypercomputational device similar to the human brain. The pruning process that each such brains undergo during the equivalent of adolescence (in order to remove the unnecessary neural connections) would preclude a copying of the hardware to create a copy of the individual.

I do think that it would take superscience to copy hundreds of billions of digital nerve cells and their tens of trillions of digital nerve connections without major divergence. I also think that it would superscience to copy the programs of hundreds of billions of digital nerve cells without error within any reasonable timeframe. I do think though that you could copy the basic structure of an existing brain in order to create a 'brain child' whose initial thought processes would be similar to the thought processes of the 'brain parent' before experience causes divergence (similar to how the thought processes of a human child are similar to the thought processes of their parents until adolescence).

[vicky_molokh]

Originally Posted by AlexanderHowl

Yes, I think that a human brain (and any digital brain whose physical structure mimics the human brain and whose underlying processes mimic the function of human nerve cells) would be by our current definition a hypercomputational device. We are capable of leaps of intuition that a Turing machine could not reasonably mimic. We are also capable of realtime 3D visualizations and realtime modifications of our 3D visualizations that I do not think that a Turning machine could mimic in a sensible fashion.

The advantage of the human brain is 500 million years of neural evolution that allows for a hypercomputational device that software cannot mimic. I do think that a physical digital brain, as in a brain that uses digital signals and digital storage rather than chemical signals and chemical storage, could also form a hypercomputational device similar to the human brain. The pruning process that each such brains undergo during the equivalent of adolescence (in order to remove the unnecessary neural connections) would preclude a copying of the hardware to create a copy of the individual.

I do think that it would take superscience to copy hundreds of billions of digital nerve cells and their tens of trillions of digital nerve connections without major divergence. I also think that it would superscience to copy the programs of hundreds of billions of digital nerve cells without error within any reasonable timeframe. I do think though that you could copy the basic structure of an existing brain in order to create a 'brain child' whose initial thought processes would be similar to the thought processes of the 'brain parent' before experience causes divergence (similar to how the thought processes of a human child are similar to the thought processes of their parents until adolescence).

So in your setting, the Church-Turing Thesis is self-evidently (or at least provably, or failing that, non-obviously but objectively) false?

[AlexanderHowl]

Yes, at least when it comes to noncomputational cognition, though I posit that humans are capable of approximating computations of greater size than any Turing machine is capable of computing. For example, I posit that it is possible for a skilled human being to approximate 100,000,000,000! within 1% of its true value with only a few months of work using a notepad and a few pencils while it would take a Turing machine the mass of the galaxy a billion years to reach that level of accuracy.

When it comes to cognition, a simulation of intelligence is not a replacement of intelligence. Hardware is the key to intelligence, the software simply runs the foundational elements of the hardware that allows for the evolution of intelligence (the software in the case of biological hardware is the genetic code while the software in the case of digital hardware is the binary code). In effect, intelligence is a third element that arises from the interaction of hardware and software and, because it is capable of hypercomputation, is superior to any Turing machine.

[TGLS]

I dunno. WolframAlpha was pretty speedy.

[whswhs]

Originally Posted by AlexanderHowl

Yes, at least when it comes to noncomputational cognition, though I posit that humans are capable of approximating computations of greater size than any Turing machine is capable of computing. For example, I posit that it is possible for a skilled human being to approximate 100,000,000,000! within 1% of its true value with only a few months of work using a notepad and a few pencils while it would take a Turing machine the mass of the galaxy a billion years to reach that level of accuracy.

It seems as if the only way you could verify that you were that close would be to perform the actual computation, and that is by definition something that a Turing machine could do.

Historically, the Turing machine as defined in the original paper was not an actual or proposed machine. In fact, it was not even conceivable that it could be constructed; by definition, it has a tape of infinite length, and we don't have infinite raw materials with which to make such a tape, or infinite manufacturing capability to work with them. The Turing machine is an idealized abstract model, like an ideal gas or an infinite flat plane. And what it's a model of is not necessarily a machine at all; it's a model of the capabilities of an idealized human logician, and the limits it proposes are equivalent to the limits imposed on human logicians by Gödel's theorem.

So I think your human mathematician who's estimating factorial one hundred billion, if they're using any sort of logically verifiable process, is doing something that a Turing machine could also do. And if they're not using such a process, I'm not sure that what they're doing is properly called "mathematics."

If I wanted to look for things that could not be done by a Turing machine, I think I'd look for things that living organisms don't do by logic; for example, for pattern recognition and object perception, which seem to be massively parallel rather than sequential. That's not to say I'm sure they can't be done sequentially; consider for example that classical analog television involves converting a two-dimensional image into a one-dimensional sequential scan, which works because we can scan really fast relative to the response time of the human eye. But that seems to be the kind of thing that it's challenging to program.

[ericthered]

It seems to me that the thing that (currently) separates humans from computers is the ability to create the algorithms. Running them is the easy part.

Now, there is a theoretical algorithm that makes algorithms, but that's something that we've been working on hard on and off for decades, and its not proven, but its suspected to exist.

[AlexanderHowl]

Originally Posted by TGLS

I dunno. WolframAlpha was pretty speedy.

While I admit that I engaged in hyperbole, how do you know that the results of WolframAlpha are correct in this case (or within 1% of the correct value)?

[Anthony]

Originally Posted by AlexanderHowl

While I admit that I engaged in hyperbole, how do you know that the results of WolframAlpha are correct in this case (or within 1% of the correct value)?

Well, assuming I don't trust WA, I'd write a computer program to do it for me. There's lots of evidence that computers are better at bulk computation than humans.

[sir_pudding]

Originally Posted by AlexanderHowl

Yes, I think that a human brain (and any digital brain whose physical structure mimics the human brain and whose underlying processes mimic the function of human nerve cells) would be by our current definition a hypercomputational device. We are capable of leaps of intuition that a Turing machine could not reasonably mimic. We are also capable of realtime 3D visualizations and realtime modifications of our 3D visualizations that I do not think that a Turning machine could mimic in a sensible fashion.

The advantage of the human brain is 500 million years of neural evolution that allows for a hypercomputational device that software cannot mimic. I do think that a physical digital brain, as in a brain that uses digital signals and digital storage rather than chemical signals and chemical storage, could also form a hypercomputational device similar to the human brain. The pruning process that each such brains undergo during the equivalent of adolescence (in order to remove the unnecessary neural connections) would preclude a copying of the hardware to create a copy of the individual.

I do think that it would take superscience to copy hundreds of billions of digital nerve cells and their tens of trillions of digital nerve connections without major divergence. I also think that it would superscience to copy the programs of hundreds of billions of digital nerve cells without error within any reasonable timeframe. I do think though that you could copy the basic structure of an existing brain in order to create a 'brain child' whose initial thought processes would be similar to the thought processes of the 'brain parent' before experience causes divergence (similar to how the thought processes of a human child are similar to the thought processes of their parents until adolescence).

Then human-equivalent AI is equally impossibe, since it would require the same computations and be capable of the same tasks as a human brain, by definition.

[Johnny1A.2]

Originally Posted by whswhs

It seems to me that the principles that apply in THS will generalize to most transhuman settings. Suppose that a human brain, or a cognitive system in silico, processes information and makes decisions in a particular way not because it has been programmed to do so, but because of its distinctive physical structure.

* In principle, it's possible for a Turing-capable mechanism to emulate the functioning of any other mechanism, using a software description of the relevant hardware.

Lots of things are possible in principle that don't work in practice, for a whole slew of potential reasons.