Newsgroups: comp.ai.games
Path: cantaloupe.srv.cs.cmu.edu!bb3.andrew.cmu.edu!newsfeed.pitt.edu!gatech!swrinde!howland.reston.ans.net!math.ohio-state.edu!newsfeed.acns.nwu.edu!ftpbox!mothost!schbbs!news
From: Dave Bingham Visicom <bingha_d>
Subject: Re: Fleet tactics & strategy
Content-Type: text/plain; charset=us-ascii
Organization: MOTOROLA 
Date: Fri, 15 Mar 1996 16:27:47 GMT
Message-ID: <1996Mar15.162747.23037@schbbs.mot.com>
X-Mailer: Mozilla 1.1N (X11; I; SunOS 4.1.4 sun4m)
Mime-Version: 1.0
X-Url: news:pNGpEk7.predictor@delphi.com
Content-Transfer-Encoding: 7bit
References: <314697FC.47E4@dra.hmg.gb> <pNGpEk7.predictor@delphi.com>
Sender: news@schbbs.mot.com (SCHBBS News Account)
Nntp-Posting-Host: 170.1.15.1
Lines: 53

Will Dwinnell <predictor@delphi.com> wrote:
>Simon A Watts <sawatts@dra.hmg.gb> writes:
> 
>>Any suggestions out there as to an AI system can run a *fleet* 
>>or team of units in a war game (actually, a sci-fi starship 
>>fleet battle)?
> 
>.
>You may be able to accomplish this with no central control by using a
>collection of relatively simple control rules local (at the ship
>level).  The generation of complex behaviors on a higher level by
>many interactions of simpler units at a lower level is known as
>"emergent behavior" and you may be able to utilize this as a control
>strategy.
>.
>Will Dwinnell
>Commercial Intelligence Inc.

Will has a good idea here. However, realize that in any large heirarchical
organization, especially a military one, there are hundreds of local rules
followed by each level of the organization. That is what training is all about.
It takes years to inculcated into the command and staff echelons the hundreds
of rules (constraints) under which they must operate.  

Also, 'emergent behavior' is contrary to the real process of action in any
military organization. Orders (goals) are sent down from each level of the
organization. The orders are 'scoped' for the level of the organization to
which they are targeted. Those orders are translated into action by the lower
level.  The lower level adds all its constraints, proposes (acceptable) courses
of action and then selects and executes one. This often translates into more
orders being sent to lower levels.

As a West Point graduate, I would guess there are at least 10's of rules
(constraints) that come into play for a lieutenant. For a general they may be
100's of rules. However, by 'scoping' the orders (goals) to the lower level,
many of these rules may not apply, and the problem simplified.

My understanding of a neural net was that it circumvented the need to
explicitly define these rules, by focusing on the output behavior and capturing
that instead. Sort of a short-cut.  

BTW, the Marines are investigating an automated learning system that 'captures'
the planning (receiving orders, applying constraints, proposing actions,
selecting an action, executing) of its 'best', most experienced officers. This
automated system would then track or mirror the planning of officers new to the
staff and provide a check that they weren't overlooking something critical. A
very interesting problem.

-- 
****************************************************************************
David A. Bingham	        |   bingha_d@motsat.sat.mot.com
Motorola, Phoenix, AZ           |   "Make it so."

