Of course, the most interesting part isn't that complex patterns can emerge from simple rules, it's in discovering the rules for changing the rules of the game itself. People are always coming up against the limits of whatever system they find themselves within. Some then seek ways around those limits (Who?). Sometimes they are successful (Why?), and sometimes their success leads to a cascade of changes that change social rules themselves (When?). Maybe their attempts can be modeled as being completely random processes, but my gut feeling is that intelligence matters to this search, though clearly system structure constrains which attempts actually succeed.
Frank
Frank Lenk
Director of Research Services
Mid-America Regional Council
600 Broadway, Suite 200
Kansas City, MO 64105
www.marc.org
816.474.4240
[log in to unmask]
816.701.8237
-----Original Message-----
From: News and discussion about computer simulation in the social sciences [mailto:[log in to unmask]] On Behalf Of Dan Olner
Sent: Friday, June 27, 2008 2:28 AM
To: [log in to unmask]
Subject: [SPAM] - Re: [SIMSOC] How intelligent are our agents? - Bayesian Filter detected spam
Hi all
Some interesting points from the last few mails; just a few comments -
Frederic says: it depends on the aim of your model; and making more intelligent agents won't necessarily make your model better.
I always find myself thinking about 'the game' -
http://www.icosystem.com/game.htm
In this game - both the virtual and the real-world version - two global results occur from simple interactions of only two rule-sets. That result would be *lost* by making agent behaviour more complex.
And vitally, in the real world it doesn't matter that the humans playing it are capable of more complex behaviour. The point is, they're *enacting* the rules of a simple game, and those rules exactly mirror what we see in the model. When modelling this - and in the case of reality - we can ignore everything else about those people, whilst they're playing the rules of that game. The same could be said for, e.g., markets. People play games, and the rules of those games can cut through great swathes of meat-world complexity. (Dynamic and Platonic at the same time...)
Leigh says: "Some ABM researchers have indeed concentrated on simple agents in an attempt to understand the manner in which complicated phenomena can arise from repeated interactions AMONG agents rather than from any complexity inherent in the structures of the individual agents per se."
So 'the game' comes into the 'complexity through simple interaction' box. (Admittedly, it's a trivial sort of emergence, but it's good for illustrating the point - and how many other ABMs can you get colleagues to play in the football field??)
So when we're discussing intelligence, we shouldn't just be talking about the 'black box' of encapsulated agent intelligence. E.g. the Balinese rice-growing water temple system -
http://press.princeton.edu/titles/4831.html
- as a whole has a form of intelligence: it manages irrigation for a whole region.
So: stupid agents can make intelligent systems, intelligent agents can make stupid systems, intelligent agents can play 'games' with simple (or more complex) rules can make either kind of system. Which is good, because then we can justifiably say they can be modelled! (As Stephen Lansing did in Bali.)
Incidentally, it's quite a political hot potato: the 'stupid agents make intelligent systems' argument has been used by Burke, Dicey and Hayek, amongst others, to argue that people shouldn't get ideas above their station. As Andrew Gamble puts it, 'ignorance is a necessary component of order.' Which puts a different light on the intelligent systems argument. But that's going off on a tangent somewhat...
Dan
-----Original Message-----
From: News and discussion about computer simulation in the social sciences [mailto:[log in to unmask]] On Behalf Of Frederic Amblard
Sent: 26 June 2008 23:13
To: [log in to unmask]
Subject: Re: [SIMSOC] How intelligent are our agents?
Hello Peer (good to have news from you) and all others, I wonder whether the question is really significant in order to characterize agents in a model. I mean that you will have difficulties anyway to answer to this question and once answered I am not sure it will be of any interest for your model or your work ...
The difficulty to answer to the question is that each time you're searching for a definition, you gain three other concepts you have also difficulties to qualify (btw what does it mean exactly for an agent to be social (only communication ? representation of others (what is representation of others...) ?) ?
What does it mean to be proactive (I have a problem I am not sure to "persistently pursue goals" myself, at least consciously ... Machado:
"travelers, there is no path, paths are made by walking" ) ?
A far more efficient way, at least for me to qualify models in social simulation is to use distinctions like reactive / cognitive agents ; symbolic / numeric (for interaction, representations ... ) ; and also aims of the modelling approach : prediction / describe / understanding...
Concerning the answer by Helder Coehlo, I do agree that surely the simulation community is not aware enough of what can be consider as some extension of the agent paradigm for building models. But I am not sure (the point was underlying) that building "intelligent" agents will make your models better. Modelling aims at abstracting reality not at reproducing it ... and so on ... But indeed it is very interesting to have new possibilities, something like new words to build models and in particular to represent part of the target system that were not taken into account before simply because you don't know how to express it with your current language (you all experimented the gap in expressivity between mathematical language and algorithms).
Last point concerning "intelligent" approaches or at least "intelligent agents" approaches is that you have difficulties to understand and interpret what your model is doing, then I am not convinced that such approach can actually be used for purposes of understanding, probably much work also to do in order to represent simulation outputs (indicators and so on) when you have intelligent agents.
atb
Fred
Leigh Tesfatsion a écrit :
> 26 June 2008
>
> Dear Helder Coelho and Other SimSoc Participants:
>
> RE: How intelligent are our agents?
>
> At 03:55 AM 6/24/2008, Helder Coelho wrote:
>> When discussing agent intelligence and architectures two books are
>> important for me, the one (Chapter 4) by Mike Wooldridge Introduction
>> to MAS (2002) e and Russell&Norvig AI, A Modern Approach (2003), and
>> they are general, but only followed by paradigm 1. You are right. I
>> use also another book (Chapter 2) by Wooldridge Reasoning about
>> Rational Agents (2000), which is clear enough about the structure of
>> the BDI architecture, and it helps me to explain the issue of
>> intelligence to students.
>
> SimSoc participants interested in this discussion might also be
> interested in the following special report put out in the IEEE
> Spectrum by the Institute of Electrical and Electronic Engineers,
> hardly a wild-eyed bunch.
>
> IEEE Spectrum Special Report: The Singularity
> http://spectrum.ieee.org/singularity
> June 2008 Issue
>
> "The singularity is supposed to begin shortly after engineers build
> the first computer with greater-than-human intelligence. That
> achievement will trigger a series of cycles in which superintelligent
> machines beget even smarter machine progeny, going from generation to
> generation in weeks or days rather than decades or years. The
> availability of all that cheap, mass-produced brilliance will spark
> explosive economic growth, an unending, hypersonic, technoindustrial
> rampage that by comparison will make the Industrial Revolution look
> like a bingo game." (quote from the article by G. Zorpette)
>
>> The division by M. Luck says nothing about intelligence, but in my
>> opinion there is no difference at all. Yet, and currently, the
>> paradigm 2 (simulation community) is not aware of that and agents are
>> very simple (eg. bit strips), without any concern about intelligence
>> (see the last book by Epstein, 2008).
>
> If "simulation community" is interpreted as encompassing the
> agent-based modeling (ABM) community as a whole, then I think the
> above assertion (that such researchers are "without any concern about
> intelligence") is an inaccurate and misleading assessment.
>
> Some ABM researchers have indeed concentrated on simple agents in an
> attempt to understand the manner in which complicated phenomena can
> arise from repeated interactions AMONG agents rather than from any
> complexity inherent in the structures of the individual agents per se.
>
> Other ABM researchers, however, have focused on the use of ABMs to
> study issues arising for real-world systems (e.g., design of markets
> with good performance characteristics). In the latter case there is
> most definitely a concern that the agents in the ABM appropriately
> reflect the characteristics of their empirical counterparts, including
> their intelligence characteristics. Whether this objective is
> achieved can be debated, of course, but I would certainly say that the
> concern is there.
>
> Best wishes,
>
> Leigh Tesfatsion
>
> Leigh Tesfatsion Department of Economics
> Tel: (515) 294-0138 Iowa State University
> FAX: (515) 294-0221 Ames, Iowa 50011-1070 U.S.A.
> [log in to unmask] http://www.econ.iastate.edu/tesfatsi/
>
|
