Newsgroups: comp.ai,comp.robotics.misc
Path: cantaloupe.srv.cs.cmu.edu!das-news2.harvard.edu!news4.ner.bbnplanet.net!news3.near.net!paperboy.wellfleet.com!news-feed-1.peachnet.edu!gatech!howland.reston.ans.net!tank.news.pipex.net!pipex!news.mathworks.com!uunet!in1.uu.net!ncrgw2.ncr.com!ncrhub6!daynews!intruder!news
From: David E. Weldon, Ph.D. <David.E.Weldon@DaytonOH.ATTGIS.COM>
Subject: Re: Logic, representation, and robotics  (Was: What's the best AI for a robot)
X-Nntp-Posting-Host: 149.25.61.42
Message-ID: <DBvvFw.C00@intruder.daytonoh.attgis.com>
Sender: news@intruder.daytonoh.attgis.com (News administrative Login)
Reply-To: David.E.Weldon@DaytonOH.ATTGIS.COM (WELDOD)
Organization: AT&T Global Info Solutions
X-Newsreader: DiscussIT 2.0.1.2 for MS Windows [AT&T Software Products Division]
References: <3uc2lg$k8l@Mercury.mcs.com>
Date: Mon, 17 Jul 1995 23:19:08 GMT
Lines: 62
Xref: glinda.oz.cs.cmu.edu comp.ai:31628 comp.robotics.misc:277


}==========Jorn Barger, 7/16/95==========
}
}In article <3tvddl$gv8@zonker.cs.ucsb.edu> on comp.ai,
}Steve E. Chapel <schapel@zonker.cs.ucsb.edu> wrote:
}>From the recent posts on first order logic and bilattices, I think 
}the
}>reasoning/planning engine would incorporate first order logic 
}with an
}>incompletely specified partial order for the truth values. [...]
}
}I answered:
}>(My real problem with FOL, etc, is that from a programmer's point 
}of
}>view it's not usually the right tool for the job...)
}
}
}What sort of rules does a robot need?  
}
}Is logic the most efficient way to handle them?
}
}"If there's an obstacle, go around it" looks like an if-then formula,
}but it's not-- it's really a *when-then* statement, involving
}*reasoning about time* that doesn't come free with FOL.  (If-thens,
}for example, are true whenever the first part is false; when-thens 
}are
}something else altogether.)
}
}But the symbol-system of FOL (etc) was born on the printed page,
}and translating it to RAM can result in a non-optimized image of
}that page-- a series of data-structures, each representing one 
}'rule',
}arranged in a sequence like the rules on the page...
}
}And you should distrust any theorist who claims that indexing the
}rules is a separate problem-- if you look at it deeply, when you've
}optimized the indexing of the rules, what you have may not look
}any longer like FOL at all...
}
}A really important sort of reality-test for AI programmers is to
}contemplate their programs as mathematical abstractions, with all
}the symbolic labels stripped.
}
}Because they all have to use the same basic buildingblocks, and
}a new approach will have to assemble those buildingblocks in a
}new way-- but to see what's new you have to look behind the
}symbolic labels...
}
}So I'd like to suggest that the data structure you want for a robot
}is probably more like a list of goals, and a list of obstacles for
}each goal, and a list of strategies for each obstacle.
}
}But there's not much FOL left, at this point... is there?
}
So far as I know, none of the above says anything one way or the other about
FOL.  FOL is simply a model for thought processes that helps organize
algorithms and assists, in a major way, to avoid inconsistencies and errors. 
Performance issues, data structures, goals, building blocks, indexing, and
conversion to machine language are all independent of (orthogonal to) the use
of FOL in a computer language.
}

