Hi all,

I liked very much both Virginia's and Maarten's messages. I should  
perhaps have emphasised more the use of ANY agent language, whatever  
it is. I strongly believe this trend will lead to much more  
sophisticated social simulations.

Yes, Maarten is right that Jason itself doesn't have an in-built  
notion of time, but it isn't difficult to get around that for doing  
simulation. Besides, the MAS-SOC platform (http:// 
jasss.soc.surrey.ac.uk/8/3/7.html) handles that nicely, but  
unfortunately it's not publicly available for download yet.

Also, my personal view on Brahms is that, even though the theoretical  
foundation *is* different, what it in practice offers for programmers  
isn't conceptually TOO different from what other (e.g., BDI) agent  
languages do (if you put on top it an organisational model), and  
indeed suitable for the same class of problems. But I'm only saying  
this to get Maarten furious! :D

The important thing I think we are all saying is: "shop around" for  
the best agent language for your project, but DO use an agent language!

All the best,

Rafael


On 13 Jan 2007, at 18:26, Maarten Sierhuis wrote:

> I second Rafael's points about using agent-oriented languages for  
> simulating complex agent behavior. However one has to be careful in  
> selecting cognitive architectures as a model for large scale  
> societal behavior. Cognitive models are based on a theory of human  
> cognition, based on the notion that humans are bounded rational and  
> goal-driven. But more importantly, these architectures are based on  
> a theory of human memory that has never been proven to be correct.  
> More so, many scientist (e.g. neuroscience, anthropology, cognitive  
> science) have in recent years developed counter theories to the  
> theory of the human mind as a "symbolic copy machine." I am not  
> saying don't use BDI languages as a model for simulating social  
> phenomena, I am saying be careful in using the fact that BDI is  
> based on Bratman's Planning Theory which is based on folk  
> psychology and on the premise that humans are planning agents (but,  
> alas, not every human activity is goal-driven).
>
> Another point of caution is the fact that BDI languages, such as  
> Jason and others, are no simulation languages. There is no notion  
> of time and a simulation-based scheduler (Rafael, please correct me  
> if I am wrong). If you use something like Soar or ACT-R there is  
> only time in the form of the engine simulating retrieval of "facts"  
> from "memory" in milliseconds (it takes a long time to simulate one  
> day with a clock that goes in milliseconds, let alone a week, month  
> or year). The language itself has no time predicates for the agents  
> available.
>
> Myself, I have been trying to argue for years (at least since 1992)  
> that using agent-oriented simulation languages (as opposed to Java,  
> C++ or C#) for simulating complex agent organizations is really the  
> right thing to do. If you want to simulate goal-driven human  
> cognition at the millisecond level, indeed Soar and ACT-R might be  
> the right tools. However, if you want to simulate territorial self- 
> organization of a large number of agents, I would question using a  
> goal-driven human cognitive model (even for simulating humans,  but  
> definitely for simulating other life-forms). If you do, you better  
> have a validated theory that your simulated life-form really  
> "works" according to a goal-driven bounded rational cognitive  
> model. For simulating humans there are other theories around that  
> put this assumption in question. Personally, I think that this is  
> one of the main reasons why the social simulation community is  
> still using "low-level" languages (C++, Java, C#) to simulate  
> agents. These "low-level" languages are not based on any social  
> theories and thus its users need to first develop their own theory  
> about agent behavior before creating a model. However, the limiting  
> factor of these languages is the fact that they are based on  
> Turing's theory of programs as a sequence of instructions, instead  
> of parallel activated processes.
>
> As a plug for my own work on agent-oriented *simulation* languages;  
> Brahms is a BDI-like simulation language that is not based on  
> Bratman's planning theory (that's why I call it BDI-like), but is  
> based on the theories of situated action, activity theory, situated  
> cognition, and perception-action behavior (and is thus not goal- 
> driven but "belief-driven" and activity-based). You can model agent  
> organizations (as multiple inheritance groups of agents) and  
> located behaviors (situation-action rules) as subsumed activities  
> in a subsumption architecture (a la Brooks' hierarchically  
> networked behavioral perception-action models). You can model  
> situatedness of agents in a geography (such as a grid) model,  with  
> the state of the world modeled as facts that agents can detect and  
> act upon. Brahms is an even-based simulation language and agents  
> can have beliefs about time (either real time or simulated time).  
> You can set a simulation clock grain size as low as a second to as  
> long as you want. Last, agents can also forward reason (deduce new  
> beliefs based on other beliefs) as in a BDI model and communicate  
> beliefs (via a time-taking communicate activity) to other agents.  
> Brahms does not rely on any specific model of human or any other  
> life form memory or cognition.
>
>
> Doei ... MXS
>
> ______________________________________________________________________ 
> _
>
>  Dr. ing. Maarten Sierhuis                							USRA/RIACS
>  Senior Scientist                              							Mail Stop B269-1
>  Human-Centered Computing           						NASA Ames Research Center
>                                                      								 
> Moffett Field, CA 94035
>  e-mail: [log in to unmask]
>  Phone: (650) 604-4917
>  Fax: (650) 604-4036
>
>  http://homepage.mac.com/msierhuis
>  http://www.agentisolutions.com
>  http://ic.arc.nasa.gov/projects/brahms/index.html
>  http://www.riacs.edu
> ______________________________________________________________________ 
> _
>
> This communication is intended for the use of the addressee only  
> and may contain information that is privileged or confidential.  If  
> you are not the addressee, you are hereby notified that any  
> dissemination, distribution or use of this communication is  
> prohibited.  If you received this communication in error, please  
> destroy it, all copies and any attachments and notify the sender as  
> soon as possible. Any comments, statements or opinions expressed in  
> this communication do not necessarily reflect those of NASA or USRA/ 
> RIACS, its subsidiaries and affiliates.
>
>
>
> On Jan 13, 2007, at 8:36 AM, Rafael H Bordini wrote:
>
>> Dear Rui,
>>
>> First, apologies for the delay in replying, I was away with little  
>> internet access for a couple of weeks.
>>
>> Indeed your message led to an interesting discussion. Personally,  
>> I was delighted with your mention of using Jason, of course, and  
>> indeed the other agent programming languages and platforms you  
>> mentioned. There has been much progress on those recently, and  
>> I've very recently heard of quite a few people interested in using  
>> various different agent languages and platforms to develop  
>> simulations with agents that are more elaborate than what is  
>> typically obtained with agent-based simulation toolkits or with  
>> more suitable code than obtained by those who developed  
>> simulations programming in, say, C or Java. As Scott rightly  
>> pointed out, what is special about agent-based models is that it  
>> allows users to observe processes such as agents learning from  
>> each other, the emergence of social relations (as reflected, e.g.,  
>> in an agent's "mind"), etc.
>>
>> Personally, I think you are in the right direction by using a  
>> cognitive agent architecture (with the support of a programming  
>> language tailored for that) to improve your model in the way you  
>> mentioned below. Obviously you might stumble across fancifully  
>> worded, but otherwise quite empty, criticism of BDI-based agent  
>> programming from people who have absolutely no alternative to  
>> offer you when it comes to programming complex agents, and indeed  
>> very little knowledge of what's going on in that research area. In  
>> any case you should be aware of that and I must also say that you  
>> are likely to find difficulties in this approach precisely because  
>> there isn't much past experience you will be able to rely upon.
>>
>> Another thing I should point out is that most agent programming  
>> languages aren't yet well equipped with constructs for some social  
>> aspects of multi-agent systems such as norms. If you make a final  
>> decision to use Jason, the best thing will be for you to use the  
>> Moise+ organisational model by Hubner/Boissier/Sichman which has  
>> already been combined with Jason, see: http://moise.sourceforge.net/
>>
>> Best of luck with your research,
>>
>> Rafael
>>
>>
>> On 5 Jan 2007, at 16:47, Rui Lopes wrote:
>>
>>> Hi!,
>>>
>>> My name is Rui Lopes and I'm a M.Sc Student of Computer Science  
>>> in the University of Coimbra. I am working in ECoS, CISUC (http:// 
>>> cisuc.dei.uc.pt/ecos/) for one year and hale, ina  project on  
>>> Models of Territorial Self-organisation - MATer (http:// 
>>> cisuc.dei.uc.pt/ecos/view_project.php?id_p=54). Now I am also  
>>> working on my dissertation on the Emergence and Propagation of  
>>> Norms and Beliefs and recently I have joined this list. The aim  
>>> of dissertation is to "upgrade" the simulation that results from  
>>> MATer with agents that recognize and reason about beliefs and  
>>> norms to study the emergence of norms and beliefs and how those  
>>> propagate in real world territories.
>>> After some research I am decided to build BDI based agents. MATer  
>>> uses Repast and I'm thinking in use Jason or Jadex for the  
>>> reasoning of the agents. I have also looked at others like 3APL,  
>>> Jess, etc but they seem less appropriate.
>>> I have read on several frameworks for modeling norms and beliefs,  
>>> but most of them are only theoretical models. I didn't understand  
>>> quite well if modal or deontic logic are an obligation in such a  
>>> system and if there are proper tools to embed this kind of  
>>> reasoning into a toolkit like repast.
>>> I would appreciate some counseling on the toolkits/languages to  
>>> use on my simulation as also as references and readings on  
>>> implemented models of norms and beliefs.
>>>
>>> Thank you very much in advance,
>>> Rui Lopes
>>>
>>> ----------------------------------------------------------------
>>> This message was sent using IMP, the Internet Messaging Program.
>>
>>
>>
>> ----------------------------------------------------------------
>> Dr. Rafael H. Bordini             Department of Computer Science
>> [log in to unmask]            University of Durham
>> http://www.dur.ac.uk/r.bordini    South Road, Durham DH7 6TE, UK
>> +44 191 334 1727 (phone)          +44 191 334 1702 (fax)
>