emergent/swarm/evolutionary systems etc

[Peter MacKenzie]

Hello, I'm Peter, and I'm very much a newbie.

New in the sense of being a novice at everything to do with programming,
rather than just new to Python, so please be very explicit in your replies
(should you deem me worth of such attention). I've long had a passing
interest in programming from a macro-conceptual level, but only very
recently have I tried to learn the details of how to impliment my ideas in
code. Since I'm so inexperienced and learning soley from on-line
documentation in my sparse spare time, I'm finding the learning curve rather
steep, and would appreciate it if you didn't assume that I 'know' anything
(especially with formating issues).

What instigated my uptake of programming was a recently sparked, and rapidly
growing interest in emergent processes, and particularly of computer
applications for them. I've not covered more than 20 hours of good study on
the subject (accounting for liberal procrastination ), but I like to jump in
the deep end with things, and the logic of emergence really 'clicks' with my
brain. I have, however covered biological evolution in considerably more
depth, so I'm already quite familiar with the concepts involved.

What I've been trying to do with Python is to set up something resembling
John Von Neumann's conception of an imperfectly self-replicating program:

(3.2.1 Self-Reproduction
Nearly all of the practical implementations of articial worlds to be
described are related
in some way to the seminal work of John von Neumann. In the late 1940s and
early
1950s, he devoted considerable time to the question of how complicated
machines could
evolve from simple machines.9 Specically, he wished to develop a formal
description of
a system that could support self-reproducing machines which were robust in
the sense
that they could withstand some types of mutation and pass these mutations on
to their
ospring. Such machines could therefore participate in a process of
evolution.
Inspired by Alan Turing's earlier work on universal computing machines
[Turing 36],
von Neumann devised an architecture which could full these requirements.
The ma-
chine he envisaged was composed of three subcomponents [von Neumann 66]:
1. A general constructive machine, A, which could read a description (X) of
an-
other machine, X, and build a copy of X from this description:
A + (X) ; X (3.1)
(where + indicates a single machine composed of the components to the left
and
right suitably arranged, and ; indicates a process of construction.)
2. A general copying machine, B, which could copy the instruction tape:
B + (X) ; (X) (3.2)
9 Von Neumann had diculties in dening precisely what the term
`complicated' meant. He said \I
am not thinking about how involved the object is, but how involved its
purposive operations are.
In this sense, an object is of the highest degree of complexity if it can do
very dicult and involved
things." [von Neumann 66].
3.2. PREVIOUS WORK 47
3. A control machine, C, which, when combined with A and B, would rst
activate
B, then A, then link X to (X) and cut them loose from (A + B + C):
A + B + C + (X) ; X + (X) (3.3)
Now, if we choose X to be (A + B + C), then the end result is:
A + B + C + (A + B + C) ; A + B + C + (A + B + C) (3.4)
This complete machine plus tape, [A + B + C + (A + B + C)], is therefore
self-
reproducing. From the point of view of the evolvability of this
architecture, the crucial
feature is that we can add the description of an arbitrary additional
automaton D to
the input tape. This gives us:
A + B + C + (A + B + C + D) ; A + B + C +D + (A + B + C + D) (3.5)
Furthermore, notice that if the input tape (A + B + C +D) is mutated in
such a
way that the description of automaton D is changed, but that of A, B and C
are
unaected|that is, the mutated tape is (A + B + C +D0)|then the result of
the
construction will be:
A + B + C+ D + (A + B + C + D) mutation
; A + B + C + D0 + (A + B + C+ D0)
(3.6)
The reproductive capability of the architecture is therefore robust to some
mutations
(specically, those mutations which only aect the description of D), so the
machines
are able to evolve. Von Neumann pointed out that it was the action of the
general copy-
ing automaton, B, which gave his architecture the capacity for evolving
machines of
increased complexity, because B is able to copy the description of any
machine, no mat-
ter how complicated [von Neumann 66] (p.121). This ability is clearly
demonstrated in
Equation 3.5 above.

From Artificial Evolution to Artificial Life
Timothy John Taylor
Ph.D
University of Edinburgh
1999)

-The formatting on the equations is a bit messed up, but I hope you get the
idea.

Although the concept is quite straight forward, and I have my own ideas
about what specific features I'd like to include, I'm usure as to how to
implement it at the level of coding. Basically, I wish to create:
1. An entirely human-designed evolutionary engine.
2. A file that it '1' will acess, modify, test against a goal to retrive a
score, then save with the modifications intact and ready for the next run or
loop-cycle.
3. A buffer that will store a population of these files.
4. A death engine that will remove low scoring files (probably as a part of
'1')

These concepts aren't new, and are already well publicised. Variants of
them are used for determining coefficient values for engeneering, medical
etc research, but I wish to create a program that's more open ended than a
number-finder. The Neumann model is incomplete, since it not only fails to
specify a method for mutation beyond the vague decription given, but also
fails to account for the vanishingly small chance that completely random
mutation of a program would produce any intelligible results relative to the
processor's capacity for run cycles (overlooked, perhaps, because of an
over-riding focus on making the evolutionary engine robust to mutation). To
solve this, I propose an 'incubator' approach, where the engine would, as
part of its evolutionary toolkit, and particularly at the beginnng of the
process before the code becomes complex enough to allow the likelyhood of
non-fatal mutations, scavange code from external sources.

Regardless of any issues of copyright etc, this has the advantage of basing
the now-less random mutations on code that is known to be operable, rather
than the infinite monkies approach of random alphanumeric generation. To
incorporate this and other revisions and refinements that I've just spotted,
the program structure would look like so:

An entirely human-designed evolutionary engine, which:
1. Allows the user to set a goal.
2. Takes random sections of code from a library.
3. Places them in a text file.
4. Applies a 'scatter shot' of random alphanumeric changes (with either a
random or a user specified % of alpahnumerals changed)
5. Runs the file.
6. Scores the file on how many valid lines it has, if it fufilled the goal,
how long it took, etc.
7. Saves the score to the file (in the title?).
8. Repeats for a user specified number of times.
9. Selects the best programs, deletes the rest
10. Returns to '3' and repeats until stopped.

At the moment, I'm not even sure if python can create, chop and change
files, so that the effects of a program being run can persist after it's
closed down. The library modules are still largely incomprehensibl e to me,
both in their function and their implementation. I'm a very long way from
understanding how to apply the randomised aspects of the file changes, or
much else for that matter, and I can already see a bunch of things that
would need a lot of tweaking, even if the structure is sound (this is the
first proper program I've planned, which I'm sure means that the structure
is almost certaintly gibberish).

Other things that would be handy:

A bit to make sensible sections of code less vunerable to mutation, and
concentrate changes in 'junk' areas.

Something to decrease or prevent the addition of scavanged code when the
proportion of scavanged code to random alphanumeric changes and working
code, or vice versa (with working code forming the fulcrum of the dynamic).

A 'skills' counter/protector, that would allow files to retain redundant
code if their goal was changed, rather than necessarily losing it to the
pressure of clock cycles and the equivalent of genetic drift. (For this,
the 'bastion' module looks like it might be useful, though I'm not terribly
sure about how it works, or 'exactly' what it does)

An random goal changer, to provide an impetus for constant evolution without
the need for user input. (There's only so good you can be at any one thing)

An option to give the files an internal copy of the evolution engine, which
could then be modified by itself as it alters the entire file. (This may
prove difficult, as although this would allow the files to develop more
efficient strategies for the process of evolution itself, it would make them
more vunerable to fatal mutations. A possible solution would be to have a
'nuclear membrane' - a guard to prevent mutations that are particularly
likely to be fatal eg. mutations that alter significant chunks of the
engine. It would also be possible to have multiple engines in each cell,
which could self-assess and compare scores, replacing engines that
underperform through damage or 'under-evolution' (radiation resistant
bacteria use a similar technique)
=============== ===
All of this seems like a rather tall order, especially as I'm still
struggling with some of the things from the beginners guide. Although I
should probably buy a book and work my way through lots of examples of
increasing complexity, I like to tackle things head-on and learn as I go.
Even just writing this message has helped me to clarify to myself what it is
I want to do.

If anybody has even bothered to read this far, I'd ask you please to give
some thought to these ideas (of myself and others), and perhaps to give me
some pointers in the right direction. Thankyou.

[Cameron Laird]

In article <c4**********@n ews.service.uci .edu>,
Josiah Carlson <jc******@uci.e du> wrote:

rules. Given that I have only a basic foothold on the language, does
anybody foresee difficulties for me learning enough to impliment simple and
experimentally flexible sim-city style simulations (minus fancy graphics and
llamas) in no more than 2 months (to allow for time to conduct actual
experiments + field observations etc)? I would be able to engender aid from
various staff, and the university library should carry titles on the
subject. Failing that, I could do it the old fashioned way and buy a how-to
book, but I'd like some opinions on the difficulty of the goal from people
who've already trancended the non-programmer/programmer barrier.

Two months is a pretty tight schedule. If you're on your toes, I would
bet you could learn enough of the langauge to support your ideas in 2
months. Actually programming the thing in 2 months; I wouldn't be able
to make that kind of judgement about your abilities.

[Cameron Laird]

In article <c4**********@t itan.btinternet .com>,
Peter MacKenzie <pe*******@btin ternet.com> wrote:

[Cameron Laird]

In article <c4**********@h ercules.btinter net.com>,
Peter MacKenzie <pe*******@btin ternet.com> wrote:

Spreadsheets do seem to be an attractive option, but the benefits are not
without their detractors:

+Easy to work with
+Require no/limited skill acquisition
+Flexible

-Cells can only hold single items of data (which can be worked around by
using arrays of cells to hold multiple attributes for each location)
-Require no/limited skill acquisition (being familiar and honest with my own
psychologica l composition, I know that the only way I'm likely to develop a
fair degree of programming competence is if there's a driving pressure to do
so. It's something I'd like to learn, and this gives me the excuse/leverage
to do so.)

Unknowns: Time series graphical output would be necessary, even if it's
very primitive. Do you know if the spreadsheet could be set up in such a
way that cells would change colour depending on their values, or if the
graph making facilities would be able to create reasonable representations
of said values so that a series of graphs would be capable of showing up
phenomena with fluidic (wavelike, turbulent, etc) characteristics ?

I'm afraid the temptation to take the hard route my prove too great
(psychologic al paradoxes: I always feel good about feeling so terrible about
these things after I've passed the point of no return in the undertaking,
and the enormity of the task at hand sinks in - it's a whole adrenalin
thing), but I'd still like to make a comprehensive assessment of my options
before I commit to anything.

Spreadsheets can do anything. Python can do anything.

To first approximation, at least. They both have developed
enough to have far more capabilities than you'll exhaust in
a couple of months.

The key questions are: which better suit your psychology?
Which will get in the way less? Which support "libraries"
of related material in your "domain" (geography)?

You're not experienced enough with software yet to judge
the first of these questions well. Whatever first impres-
sions Python or Excel make on you are likely to dwindle to
insignicance after a few more weeks of software exploration.

One of the dimensions your comparison of the two approaches
doesn't cover is robustness of expression and abstraction.
You aren't in a position to appreciate this yet, but Python
beats spreadsheets all to pieces in these regards. Crudely,
you can hand a Python solution to someone else, two years
from now, and he'll understand what you've done, and how to
modify or validate or ... it. I argue strongly that that is
*not* true for spreadsheet solutions. I therefore regard
spreadsheet approaches, except in specialized circumstances,
as anti-scientific, because they don't promote the free
exchange of ideas.

There's a rich literature on simulation done with computers,
some of it specifically by those coming from geography. It
would be no particular problem to dissipate the entire two
months just reading up on what you plan to do. You need to
figure out a very circumscribed goal, and ask experts on what
you should do to achieve it.

Reading through *Thinking ... with Python* certainly can be
part of that path.
--

Cameron Laird <cl****@phaseit .net>
Business: http://www.Phaseit.net

[Cameron Laird]

In article <c4**********@h ercules.btinter net.com>,
Peter MacKenzie <pe*******@btin ternet.com> wrote:

I think then that I might stick with spreadsheets as far a possible. Do you
know of any simulation programs that might be more suited? Also, how would
you extract data from a spreadsheet for use in another program? I'm not
familiar with means of linking spreadsheets to external applications, though
I don't doubt that it can be done.

Software people *constantly* talk about "means of linking ... to
external applications"; it's one of our obsessions, for reasons
best left to another time.

Here's a way you can practice thinking about it: a spreadsheet
can write its "answers" to a file--just a simple document, the
sort you might write as a human, yourself. Another program can
read that file. That establishes communication between the
spreadsheet and the other program. Mission accomplished.

Spreadsheets typically have *dozens* of distinct ways of talking
to "outsiders" , although that's often apparent only to software
specialists.
--

Cameron Laird <cl****@phaseit .net>
Business: http://www.Phaseit.net

[Cameron Laird]

In article <c4**********@h ercules.btinter net.com>,
Peter MacKenzie <pe*******@btin ternet.com> wrote:

Hello, I'm Peter, and I'm very much a newbie.

[Cameron Laird]

In article <c4**********@t itan.btinternet .com>,
Peter MacKenzie <pe*******@btin ternet.com> wrote:

(Hmm, this might appear as a double posting, but I don't think my last one
made it through.)

Thanks, but.

("One approach to discussing and comparing AI
problem solving strategies is to categorize them using the
terms ''strong'' and ''weak'' methods. Generally, a weak

[Michele Simionato]

Josiah Carlson <jc******@uci.e du> wrote in message news:<c4******* ***@news.servic e.uci.edu>...

Want to run a file? (be careful though, executing the contents of a
file can be dangerous)...
fil = open('filename' , 'rb')
contents = fil.read()
fil.close()
exec(contents)

or just use execfile('filen ame') or even import filename.
Michele Simionato

[Michele Simionato]

"Peter MacKenzie" <pe*******@btin ternet.com> wrote in message news:<c4******* ***@sparta.btin ternet.com>...

In addition to exams, I also must start a dissertation in July for my
geography hon B.Sc. Although I should ideally have found a subject two
months ago, I've so far lacked any truly appealing project ideas (much to
the consternation of my advisor). Since reading 'Emergence', by Steven
Johnson, and conducting prelimenary research on the matter, I've settled on
the dissertation title: "Emergence theory as an approach to city design".

To this goal, I'd like to use computer modeling to simulate ways in which
the spatial distribution of indicator phenomena in cities (land price/use,
crime, demographic composition etc) is affected by bottom-up, local area
rules. Given that I have only a basic foothold on the language, does
anybody foresee difficulties for me learning enough to impliment simple and
experimentally flexible sim-city style simulations (minus fancy graphics and
llamas) in no more than 2 months (to allow for time to conduct actual
experiments + field observations etc)? I would be able to engender aid from
various staff, and the university library should carry titles on the
subject. Failing that, I could do it the old fashioned way and buy a how-to
book, but I'd like some opinions on the difficulty of the goal from people
who've already trancended the non-programmer/programmer barrier.

I would suggest you to forget about this project and to spend those two
months in learning the basis of Python. IMO learning a programming language
will do to your mind a far better good than learning a spreadsheet.
Having said so, it is ok if you keep this project in the back of your mind
while you learn programming, but do not expect you can finish it in a couple
of months starting from scratch, without guidance and not working on it full
time. OTOH, a couple of months is enough to get the basics of programming
(which is not the same as becoming a programmer). I would estimate the time
to get a minimal understanding of modern programming in one year. Still,
notice that I estimate the time to write your first useful program in Python
in five minutes, so you will get back something from your effort immediately.
This is the nice thing about programming. The bad thing are bugs, that you
will get well before the first five minutes ;)

Michele Simionato

[Peter MacKenzie]

Von Thunen? I had to look that one up, but I realised upon finding it that
I'd been taught it way back in first year geography in high school. It's
not something they refer to anymore, except as an absolute first step to
understanding settlement structure and distribution. Looking over it again,
it's easy to see why. The criteria required for it to work are almost
impossible in the real world, so the models that are taught after it are
generally refinements and novel applications of the basic Von Thunen
concepts.

To take a couple of examples, the bid/rent model of land use is a modern
version that incorporates lines of communication (roads, rails, etc) when
considering land uses, which convert the concentric structure of the Von
Thunen model into something more reminiscent of a spider web.

Models that address sociological phenomenon bring in another layer of
refinement. Particularly notable are those models that focus on 'quartered'
cities, with self-contained micro-cities serving distinct groups. These
structures were commonplace during the colonial era, when trade cities were
often a multi-nucleated patchwork of white, indigenous and 'miscellaneous'
societies.

It's common, when addressing any location, to apply a Von Thunen framework
to the layout of the area and use any discrepancies with the idealised
theory as a focus of the study. Generally though, it's something so
ingrained in the geographer's conscious that it never really gets much
thought.

Cameron Laird <cl****@lairds. com> wrote in message
news:10******** *****@corp.supe rnews.com...

In article <c4**********@n ews.service.uci .edu>,
Josiah Carlson <jc******@uci.e du> wrote:

rules. Given that I have only a basic foothold on the language, does
anybody foresee difficulties for me learning enough to impliment simple and experimentally flexible sim-city style simulations (minus fancy graphics and llamas) in no more than 2 months (to allow for time to conduct actual
experiments + field observations etc)? I would be able to engender aid from various staff, and the university library should carry titles on the
subject. Failing that, I could do it the old fashioned way and buy a how-to book, but I'd like some opinions on the difficulty of the goal from people who've already trancended the non-programmer/programmer barrier.

Two months is a pretty tight schedule. If you're on your toes, I would
bet you could learn enough of the langauge to support your ideas in 2
months. Actually programming the thing in 2 months; I wouldn't be able
to make that kind of judgement about your abilities.

.
.
.
Bluntly, I'd bet against it. There's a daunting amount
of new material you'll need to learn. My advice: we can
help you refine your project so that the computing part is
less overwhelming.

'You read von Thunen, by the way?
--

Cameron Laird <cl****@phaseit .net>
Business: http://www.Phaseit.net

[Peter MacKenzie]

>You might like "The Outsider's Guide to Artificial Intelligence"

<URL: http://robotwisdom.com/ai/index.html >.
Yes. It is interesting. I've been playing with the idea of graphical
representations for programming 'phrases' for a while, and the reference to
LISP brought it to mind. Although LISP doesn't look that much better than
Python code, are there any programs out there that let you program, um,
programs, using various shapes, colours etc? Just thinking about it brings
up all manner of difficulties that would be encountered if you tried to
create such a thing, but it would be nice if there was some immediately
obvious graphical connection between pieces of code (so beginners like me
didn't keep trying to put the square code through the round code ;-) ).
Just a thought of the 'in the shower' variety.

I also liked the idea of the metagame project
(http://satirist.org/learn-game/projects/metagame.html), though I don't see
myself taking on a challenge of that magnitude in the reasonably foreseeable
future.



Cameron Laird <cl****@lairds. com> wrote in message
news:10******** *****@corp.supe rnews.com... In article <c4**********@t itan.btinternet .com>,
Peter MacKenzie <pe*******@btin ternet.com> wrote:

(Hmm, this might appear as a double posting, but I don't think my last onemade it through.)

Thanks, but.

("One approach to discussing and comparing AI
problem solving strategies is to categorize them using the
terms ''strong'' and ''weak'' methods. Generally, a weak

.
[much more]
.
.
You might like "The Outsider's Guide to Artificial Intelligence"
<URL: http://robotwisdom.com/ai/index.html >.
--

Cameron Laird <cl****@phaseit .net>
Business: http://www.Phaseit.net