Medium Dependent Intelligence, Uncopyable Intelligence, and Transhumanism

[Anthony]

Originally Posted by Minuteman37

Is there anyone here who works on Video Game Design?

Not as such, but I'm a software engineer.

Originally Posted by Minuteman37

My impression as a laymen is that there are so many different layers of "stuff" involved in game design there isn't usually anyone who understands how the entire thing works from the sub-basement level 1 and 0s to the penthouse suite graphic design level, so you put together a large team of people who each work on their own little part and then you assemble it like a puzzle.

It's been decades since any single human has been able to understand all aspects of the behavior of event a moderately complex program.

Originally Posted by Minuteman37

I imagine AI design is like that but a thousand times worse. You have 100s of doctors, and engineers working for more then a decade to hand craft this AI

AI design would almost certainly involve programming a learning core (which could well be a very large and complex project) and then feeding it data so it will learn.

Originally Posted by Minuteman37

You're dealing with Petabytes of data, one little flaw could manifest while copying and go completely unnoticed. That might not mean anything, but what happens when you copy that copy?

Generally nothing interesting. Data corruption happens even without copying, and there are ways of dealing with it (generally by making multiple copies...).

[Anaraxes]

Originally Posted by Anthony

there are ways of dealing with {data corruption} (generally by making multiple copies...).

Or forward error correction, more efficient and effective.

[Anthony]

Originally Posted by Anaraxes

Or forward error correction, more efficient and effective.

Forward error correction is a type of redundant data transmission (or storage). Redundant data is multiple copies (or semi-copies).

[whswhs]

Originally Posted by AlexanderHowl

But we are not talking about THS, but of a more generic Transhuman setting.

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.

* A major limitation on this is complexity. But a common assumption of transhuman settings is continuing exponential growth in the complexity of computer systems. What required a computer the size of the Pentagon one decade will later be possible to a box you can carry in your pocket.

* I'd also note that there are significant parallels between genetics and code, and between epigenetics and the execution of code; the process that generates the brain might be called a "program."

[AlexanderHowl]

Originally Posted by Anthony

That does not necessarily follow. First of all, 'digital' is the wrong term in that case. Secondly, whether you can copy it or not depends on what part of mind is hardware dependent.

Digital intelligences are just intelligences dependent on digital storage rather than biological storage (the term 'digital' here just refers to the use of discrete units of measure rather than nondiscrete units of measure). There is nothing inherent in the term that requires the products of digital signals to be copyable or transferable.

[Anthony]

Originally Posted by whswhs

* 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.

It's possible to emulate any finite deterministic mechanism; a source of randomness is needed to fully emulate a lot of real world systems. In principle there could be an element that is not finite, or an element that is neither deterministic nor random, though they would be problematic from the standpoint of physics as we know it, and it's not clear what it even means to be neither deterministic nor random.

Originally Posted by AlexanderHowl

Digital intelligences are just intelligences dependent on digital storage rather than biological storage (the term 'digital' here just refers to the use of discrete units of measure rather than nondiscrete units of measure). There is nothing inherent in the term that requires the products of digital signals to be copyable or transferable.

You are using a very odd definition of digital.

[AlexanderHowl]

No, I am using a rather standard definition of digital, at least when it comes to digital data (https://en.m.wikipedia.org/wiki/Digital).

[johndallman]

Originally Posted by AlexanderHowl

Digital intelligences are just intelligences dependent on digital storage rather than biological storage (the term 'digital' here just refers to the use of discrete units of measure rather than nondiscrete units of measure). There is nothing inherent in the term that requires the products of digital signals to be copyable or transferable.

May I suggest you coin a new term for this kind of "digital"? This usage is actively confusing to people who understand the fundamentals of computing.

If an AI is to be intrinsically uncopiable, it can't run on a computer that is only Turing-complete. The hardware needs to be doing something that can't be emulated by software on a Turing-complete machine. That's quite hard, because all questions in digital circuit design can be represented in software, as can all analogue ones that are stable enough to represent with numbers.

This is why people say that uncopiable AI tends to imply something like a "soul," or alternatively, the correctness of Roger Penrose's theories about quantum mechanics and consciousness.

[whswhs]

Originally Posted by Anthony

It's possible to emulate any finite deterministic mechanism; a source of randomness is needed to fully emulate a lot of real world systems. In principle there could be an element that is not finite, or an element that is neither deterministic nor random, though they would be problematic from the standpoint of physics as we know it, and it's not clear what it even means to be neither deterministic nor random.

In philosophy, I believe that's called "agency causation": For example, the belief that a particular decision made by a human being is an act of the human being as a whole, but cannot be equated to any specific sequence of physical motions of the molecules or particles that constitute that human being. Or if you prefer, "free will."

Nonfinite deterministic systems are all through classical physics. Laplacian determinism, for example, works with differential equations, based on functions of real numbers, and real numbers as such are inherently impossible to express digitally.

[Anthony]

Originally Posted by AlexanderHowl

No, I am using a rather standard definition of digital, at least when it comes to digital data (https://en.m.wikipedia.org/wiki/Digital).

All digital data is by its nature storage system agnostic and can stored on any sufficiently large digital data store. Now, you can have a system that uses an internal digital representation but lacks the ability to output the data (in which case it cannot be copied), but we would not normally call that digital storage (it's only storage if you can retrieve your data...).