computer models

[Noah Helman]

i am confused about what type of computer model would be acceptable.
what would be the starting point for a model? let's say, for example,
you wanted to allow some sort of computational evolution subject to
some selective fitness pressure: you would need to start from some
model of a cellular system, which by assumption might be contrived to
show the behavior we are looking for. would we be allowed to assume,
for example, that there's a signaling system in place to detect a
stimulus, but that we're changing the dynamics of the environmental
stimulus and we want to see if the system can adapt using some
computational randomization of the connections in the signaling
pathway? to me, it seems like many possible computer models could be
contrived to show either result. what did you have in mind, exactly?

personally, i do believe that simple models can show emergent
behavior, and i suspect this has already been demonstrated many times
(though I haven't looked yet), but i am guessing that would not be
sufficient for you.

[bkl...@gmail.com]

I think you are right, computer models can be programmed to do the
simple things you mention. Puzzle-solving, optimizing, and testing
random recombinations, while useful, do not enable a model to escape
from its original bounds. They explore a prescribed range of
possibilities. Such exploration does not demonstrate OEEI, although
sometimes exaggerated claims are made for such models.

I think simple models (like Conway's Game of Life) can produce
emergent phenomena (like gliders). But without a feedback loop between
the instructions and the phenomena, the instructions will never
change. Without changed instructions, there's no evolution.

[Mike]

bkl,

1. What do you mean by "without a feedback loop...the instructions
will never change" when you are referring to Conway's Life? Which
instructions are you talking about? Are you talking about Conway's
rules?

2. I am having great difficulty understanding what could possibly be a
computer demonstration that would satisfy your criteria. There are
many demonstrations of evolutionary programming that evolve solutions
to problems, and which continue to evolve better solutions given more
running time. So if that does not satisfy your criteria, what would?
Can you describe a hypothetical demonstration that would satisfy your
criteria, in concrete terms? As an example of "concrete terms" here
is a hypothetical demonstration I suspect would satisfy your
criteria. If it does not, I would very much appreciate an alternative
example at a similar level of discriptive detail.

Example:
A computer program evolves chess playing strategies. Over time,
measurements of the program's performance are taken by assigning the
program a chess rating in the standard way (quarentine is not violated
since we test using copies of the program without modifiying the
original). On several subsequent measurements, the program's rating
improves, with no obvious maximum achievable rating. A maximum rating
would be "obvious" if there was a simple way to prove that the program
could not possibly surpass that rating given enough running time to
evolve better strategies. Thus, without the seeker supplying such a
proof, the program would satisfy the seeker's criteria of no obvious
maximum level of performance.

Mike

[RussAbbott]

It is well known that it is possible to simulate Turing computation
within the Game of Life. Given that, any theoretically possible
computation can be performed. Likewise, one could generate and run
every Turing machine with every input, including a Turing machine that
simulated the Game of Life within which this same demonstration were
being run. That is presumably increasingly complex and without a
complexity limit. Would it qualify? If not, why not?

-- Russ

[Keith C]

bkl,
You wrote, " But without a feedback loop between

the instructions and the phenomena, the instructions will never
change. Without changed instructions, there's no evolution. "

[B]What you are ignoring here is that mutation + natural selection IS
the feedback loop.[/B]
Mutation repeatedly generates modified instructions and selection
keeps those which function adequately.
In time, the instructions which have been retained will be different.

[bkl...@gmail.com]

I agree that mutation + natural selection can set up a feedback loop.
But I am not aware of any version of Conway's Game of Life that
includes these features. My comment was about emergent properties
(like gliders) only.

A chess program that gets better at chess is doing what it was
programmed to do. When life develops new features, it is doing what
simpler life was not programmed to do, according to mainstream theory.

In other discussions I give concrete examples that would seem to
exhibit OEEI-QS. If a computer model, without guidance, discovered the
presence of a phone modem and wrote the needed software and "escaped,"
that would be pretty convincing. However, the danger in such examples
is that a computer modeller will target one and program a path to it.
That kind of evolution would be neither Innovative nor Open-Ended. But
such "cheating" may be hard to police.

[Mike]

Thank you for the computer modem example, bkl. It is very concretely
stated. But even in that, you describe a subjective judgement that
you will have to make to determine if your criteria are met: whether
the programmer "targetted" the innovation. I am worried that for any
innovation produced by a program, you could still find a way to argue
that it was in fact targetted by the programmer, and thus avoid paying
the prize. Can you define criteria for succeeding that will not be up
to your subjective judgement?

Mike

[ctyankee]

Apparently the challenger has not, cannot, and will not define the
challenge! This thing seems more like a psych experiment than a bona
fide issue. save your brain cells.

visit http://Push.PickensPlan.com/

[Keith C]

bkl,
My comment about mutation and selection constituting a feedback loop
is quite general and not limited only to Conway's 'Life'.
That means that instructions can change, and this eliminates one of
your objections completely for all cases.

Your remaining objection seems to relate to objectives. Specifically,
you say that improvement in a chess-playing machine is not 'evolution'
if it remains a chess-playing machine.
This is a restriction imposed by the evaluation procedure provided in
the feedback loop. If this only responds to chess performance, then
the simulation will remain a chess-playing device.
If other game playing abilities are rewarded, then the result could be
a very general and versatile game playing machine. Alternatively, it
could result is several different species of machines, each adapted to
a specific game.
It all depends on what determines 'survival'.

[MangoCats]

This "escape" would be an example of a faulty quarantine, wouldn't it?

On Jul 23, 11:01 am, bkl...@gmail.com wrote:

[Paul M]

Hello,

I did submit a solution, mostly arguments what is possible and what
isn't and an general idea about simulation, however since I found out
late about this challenge and I hardly had enough time to submit my
solution I did not come here before. Now after reading through this
forum I get a much better idea what the seeker wants to find out but
since its already past deadline I will post here some arguments which
can perhaps be treated as an extension of my submission or at least as
attempt to answer the question.

Chess playing program plays chess cause that is what he is designed to
do but that shouldn't prevent it from producing innovations only those
innovations will consider a particular area - a chess game. For
example it can come up with a novel opening which would give it
benefit, or some novel way to use certain pieces for defending the
king, etc.
Would those not qualify as innovations?

Even if the answer is no I would like show that it is possible for
computer program to devise new code which will lead to the proposed
example of innovation - escaping through the modem. Of course for this
to happen certain conditions have to be met - there has to be a
working modem on both connected computers and the second computer has
to be able to download the code.

Here is a theoretical model of such a demonstration:
The computer which runs the demonstration has M bits of memory, the
code is executed directly after each restart and the hardware is setup
in such a way so as to reset if execution of the code produces any
error or after a certain preset constant time T (to prevent hanging).
On each reset the hardware flips one random bit in the memory. To make
the process more efficient we don't run the same simulation twice, so
if a certain state of memory Mx has already been executed we don't run
it again and instead flip another random bit until we find a novel
state.
This demonstration will be able to produce any desired effect which is
allowed by hardware and for which there exist a program P (length of P
<= M and running time of P < T). Whats more it will produce this
effect after finite time.

If as we assumed there is a modem and it is possible to escape through
that modem this demonstration will come up with all possible programs
which code the modem in the desired way and then use it to transmit
themslves to another computer. If there exist programs P meeting
length and time constrains which can escape into the internet and
infect 100 000 personal computers this demonstration will produce them
after finite time.
Since M and T are arbitrary any program written so far by humans can
be reproduced in such a manner after enough time.

Does it prove that open ended evolution in a quarantined system is
possible?

It can be argued that it is not really evolution or that it is not
open ended.

If open ended is defined as in the challenge it should meet the
criteria since it can produce many innovations, but it can be argued
that there is a limit in that after trying all 2^M combinations no
further innovation is possible.
All computer models running on finite resources face this limit, we
can however modify our demonstration to remove it.
Each time the demonstration reaches the point where all the 2^M
combinations have been tested we increase M by 1. Now there certainly
is no limit to innovation.

So someone can argue that it is not an evolution cause there is no
natural selection. That is true but is natural selection really
important for what this challenge aims to decide? This demonstration
proves that new instructions can be produced by random change and that
such instructions can lead to innovation. It shows that innovation is
possible in a quarantined (in the meaning that no premade instructions
are added) system.

Evolution can be treated like a special case of such a system, if
completely random fluctuations can produce new code and innovation
then evolution even more so. Evolution simply makes the process much
more efficient and faster at the cost of favoring certain changes over
others.

If however we decide that natural selection is absolutely necessary it
can be added to the demonstration by adding more identical computers
systems. Consider we have N such computer systems. All computers run
programs in parallel, once each computer finishes one cycle the
fitness is assigned to programs. For example lets say we base fitness
on running time t and assign fitness of t if t<T and 0 it t>=T.
After deciding fitness the most fit program is copied to randomly
selected N/2 computers, the 2nd best fit to N/4 and the 3rd to N/8 (so
that after copying half the computers start next cycle with best fit
program, 1/4 with 2nd, 1/8 with 3rd and the remaining 1/8 computers
have programs unchanged).
Now we have natural selection which selects against programs that hang
or programs which have early errors. However since after each cycle
1/8 of programs is allowed to pass to next step no matter how fit they
are it means that this selection wont prevent us from sampling the
whole possible program space and we are still guaranteed to observe
any innovation that is possible and that it will take only a finite
time.

I realize of course that in most cases such demonstration wont be
practical due to the huge time it will take, but the fact that only
practical considerations prevent us from running it should show that
open ended evolution can exist in a quarantined system.

Regards
Paul M