🚀 go-pugleaf

In <1992Sep7.232522.20395@unixg.ubc.ca> ramsay@unixg.ubc.ca (Keith Ramsay) writes:


>First, let me emphasize that the object of this exercise is simply to
>refute this supposed theoretical barrier to making a self-replicating
>machine. Huge *practical* obstacles exist. But I think we can
>distinguish between practical obstacles, and ones relating to some
>notion that in order for a machine X to construct a machine Y, X must
>be more complex than Y.

      In fact, in life, it seems that _less_ complicated forms
create _more_ complicated forms. Else how could evolution occur?
But that's another question. The idea has to be that X & Y are
the same, and further that Y can create another copy, Z etc.

>First of all, you need to have a reasonably "general purpose" machine,
>one which is capable, with the appropriate instructions, of building a
>very wide variety of different machines. Almost all of the difficulty
>in making a self-replicating machine lies here. But I believe I can
>establish the plausibility of it.

      First, this isn't what one would expect. X's complexity will
be at a minimum if it only has to create a copy of itself. Generality
of function can only arise because of increased complexity. It's not
some quality, but a sort of quantitive term describing complexity.
To give X any function it doesn't need to copy itself would be a waste.

>On the one hand, current technology allows us to do almost anything
>via remote control which could be done in person (although less
>conveniently). There are perfectly decent robot arms, for example,
>with lots of freedom of movement, sensors, and so on (the problem
>being, that they are difficult to control!). But for this to work,
>human cleverness is required, or some form of AI. On the other hand,
>there are more specialized and autonomous machines which can do
>certain construction tasks in a reasonably general way. The computer
>aided manufacturing people have systems which don't need very much
>more than a description of the shape of a mechanical part in order to
>make such a part. Both of these approaches are somewhat in the desired
>direction.

      Making a machine more complex does not necessarily move it closer
to being reproductive. The fundamental difference between machines and
living things is that living things shaped themselves, while machines were
shape _by_ living things. Thus machines are an exterior part of living
things. The notion that they can be complexified and 'given life' by
living things is what is in question.

>Granted, that we do not yet have a machine which can take the
>blueprint for another machine of a reasonably general type and build
>it from scratch. The point is, however, that no discussion merely in
>terms of "complexity" or anything like that suffices to show this
>cannot be done.

      Agreed. But no amount of technological enthusiasm is sufficient
to show that it _can_ be done.

>The nanotechnology super-optimists are optimistic
>about the possibility, because they think that it will be possible to
>get nano-manipulators to build machines in a nearly atom-by-atom way,
>and also because they believe we can imitate how cells do it. I think
>that they are overoptimistic about this, but again don't see any
>fundamental reason why a "machine" cannot perform the type of
>construction-from-template that cells do by transcribing DNA.

      The first step is to understand how the DNA transcription process
is controlled. Molecular biologists talk about a "chicken and egg"
problem between proteins as specified by DNA, and
proteins controlling when & how subsequent proteins are created.
This obviates the problem of having to find another, non-DNA control
process -- which would lead to infinite regression, and conflict with
accepted physical principles. Thus they imagine a 'loop'
structure for cellular control, rather than a hierarchical structure.
However, no system in the biological realm is 'loop structured' but this
one. The cell in all other respects is hierarchical/distributed;
multicellular organisms also, and social systems. This loop at the
bottom of things seems paradoxical. Not only is it unclear how it
could have evolved, it is unclear that it could work. The buck
must stop somewhere. (Oh yeah, there is the 'neural net' model of
the brain, but IMO that is bogus also. But that just a theory,
and another thread.)

            ...
[3 machine strategy for overcoming the problem deleted]
            ...

      Having read your 3 machine strategy a number of times, and being
unable to decide whether or not it made sense, I'll just say that
I don't see how adding another machine C can solve the problem,
since you're just extending the circle. It strikes me as the same
technique Escher used in his visual illusing of infinitely ascending
stairs: it makes sense step by step, but not as a whole.
Besides, living cells do not do it that way. Cell A creates
Cell B, which is a copy of itself. No intermediate Cell C is required.

>>      2. Obviously, a simple machine cannot do this. If a complex machine
>>can do this, what is the complexity threshhold?

>I think the above answers this. The notion that there is some
>"complexity threshold" is, I suggest, misleading, in that it suggests
>that the obstacle to be overcome is in some sense "raw" complexity:

       The obstacle to be overcome is the need for control of the
copying process while the copying is taking place. Whatever is exercising
this control cannot be simultaneously being copied. In a computer,
a PROGRAM can copy itself because the program is not actually doing the
copying. The CPU (with its _own_ program) is doing the copying. And
the CPU is a physical process, rather than stored information, and to
copy that physical structure would require another level of control,
and so on.

>>What defines the point
>>at which its complexity is sufficient to copy itself? Keep in mind that
>>the more complex the machine is, the more difficult the task of copying
>>itself. So where does the advantage of complexity overtake the
>>disadvantage?

>The advantage is able to overtake the disadvantage owing to the fact
>that the complexity of describing how to build an information storage
>device with a certain capacity x, and the complexity of describing how
>to copy information from one such device to another, does not grow as
>fast as x. And, of course, due to the fact that one can describe, by
>means of data, how to build machines.

     Interesting argument, but it raises questions. A description, in data,
of how to build a machine is still not sufficient for building
it. The more complex the structure you're building, the
more complex must be your tools of construction. E.g. you
need a better robot hand to construct a better robot hand, nevermind
to manufacture information storage devices.

>The complexity threshold is mainly due to the problems of carrying out
>instructions on how to build things. How complex this is depends upon
>the environment. Viruses can be comparatively simpler than bacteria,
>because they rely upon having the "environment" of cells which supply
>most of the tools they need to replicate. To make a device which could
>replicate itself on the surface of Mars would be exceedingly
>difficult: one would have to be able to instruct it on how to find and
>use metal ores (or else make the machine without using any metal!) and
>so on. But this is not some grand theoretical problem:

      Ah, but there's the rub. Machines are always in an alien
environment, because they didn't evolve into the universe as lifeforms
do. The suitable environment for a machine is a human-created
storehouse full of parts bins. Machines are materials
_pressed into service_, and have that thermodynamic property of
falling apart -- even when supplied with plenty of energy -- unlike
life. This is intimately related with why they cannot copy themselves.
Copying itself is how life manages to avoid this problem. It's a good
trick for that, when you think about it. The individual organism wears
out, but the copies it makes of itself are always brand new.

>>      3. If the problem had been solved, it would be equivalent to a
>>major cosmological statement, not some obscure result stuffed away in
>>someone's desk somewhere.

>What kind of "major cosmological statement" do you think would be
>involved?


      The ability to create life from 'non-living' materials seems
cosmologically significant to me. Such machines would no longer require
maintenance. We could be such machines.


>Keith Ramsay
>ramsay@unixg.ubc.ca


      John E. Snodgrass [snodgras@cyber.net]