From ALife to TownSim (Was: TownSim Models)

From ALife to TownSim (Was: TownSim Models)

Post by Jorn Barg » Wed, 29 Nov 1995 04:00:00



For the last month or so, there's been a lot of discussion on
comp.ai.games about how to build a simple simulation of a town,
like SimCity only focused on individuals...

And lately, the big question there has been 'should it include
genetic algorithms?'  (See below for a recent summary.)

I'm crossposting these comments to comp.ai.alife because the
recent announcement of "Vivarium" for the Mac got me motivated
to download a few of the Mac alife offerings... all of which
seem to use some form of GAs...

Vivarium, for example, gives each creature approx. 15 'rules' like:

  if there's food to your right then turn right

which can mutate into:

  if there's a mouse to your right then turn left

I'd recommend to the TownSim group to track down some of these
if you haven't already, because while they're a lot simpler than
the Town, they're certainly an interesting *near-miss*...
<URL: http://www.cs.wisc.edu/~smucker/EC.html > is a startingpoint.



>1.  You don't want multiple species.
>2.  In real life, nothing evolves that fast.  The town would be dust before
>    anything evolves.
>3.  GAs aren't really all that applicable to a model of a town.  The object is
>    to learn, do, interact, not optimize.
>4.  GAs aren't the magic solve-everything AI concept.  They are turning out
>    like expert systems in the early eighties and neural nets in the late
>    eighties.  A fad, basically.  Perhaps a combination of several techniques
>    would be more effective, but placing your money on this one algorithm for
>    something like a town sim seems a bit silly.

I think these objections are important, but I think the 'combination'
idea is well worth exploring, too...

j

-==---
. hypertext theory : artificial intelligence : finnegans wake . _+m"m+_"+_
                 http://www.mcs.net/~jorn/  !             Jp   Jp     qh qh
          ftp://ftp.mcs.net/mcsnet.users/jorn/            O    O       O  O

...do you ever feel your mind has started to erode?        "Y_  "Y5m2Y"  "  no.

 
 
 

From ALife to TownSim (Was: TownSim Models)

Post by Peter Harris » Sat, 02 Dec 1995 04:00:00




>1.  You don't want multiple species.
>2.  In real life, nothing evolves that fast.  The town would be dust before
>    anything evolves.
>3.  GAs aren't really all that applicable to a model of a town.  The object is
>    to learn, do, interact, not optimize.
>4.  GAs aren't the magic solve-everything AI concept.  They are turning out
>    like expert systems in the early eighties and neural nets in the late
>    eighties.  A fad, basically.  Perhaps a combination of several techniques
>    would be more effective, but placing your money on this one algorithm for
>    something like a town sim seems a bit silly.

Correct.  I am doing an A-Life project, and it must be kept in mind
that a single concept is a solve all.  For example I tried to use a
similar system to optimise school classes - it turned out more
standard methods worked.

The primary power of A-Life is not coming up with the best
optimization for a single case, but a adaptable optimization for a
range of problems.  For example, if you write a standard program it
might be good at solving a specific type of problem.  If however you
develop a Evolutionary Solver the program could produce results for a
wide range of similar problems.

One of the most serious drawbacks to using a computer is that while
the processes used in A-Life could be run on a a-sync parallel
machines (I dare not call it a computer), tradisional computers all
run serially.  Computers simulate what could be done using real
parallel processors thousands, possibily millions of times faster.

PS : I am NOT talking about parallel computers, ie four or five serial
computers working in sync together.  I imagine the beasts I describe
to be a totally different creature to the modern computer.  The
consituant parts would operate in their own time.  Only statistical
average of processing would keep them 'in sync' to arrive at sensible
results.


64 9 827-8721  mobile - 64 2 581 2219
http://www.harrison.co.nz/