XPost: comp.ai.alife   
   From: christian.guttmann@csse.monash.edu.au   
      
   Hello,   
      
   Thanks, David, Martijn and Anthony, for your responses.   
      
   > >Hello,   
   >   
   > >I am currently looking into communication and collaboration among   
   > >agents. Does anyone have an educated answer about the following question.   
   >   
   > > > Is collaboration possible without communication?   
   >   
   > >By "without communication", I mean: no exchange of any information   
   > >whatsoever - not before collaboration, not during collaboration, not   
   > >through the environment, not through signals and not through direct means.   
   >   
   > >In case, anyone has a positive answer, it would be helpful if there is   
   > >also a publication about this matter.   
   >   
   > >I am looking forward to responses.   
   >   
   > >/Christian   
   >   
   > It depends on what exactly you mean by collaboration and how strictly   
   > you interpret the "no exchange of any information" caveat.  If you   
   > mean information in the sense of Shannon, the answer is probably no.   
   > But if you allow that agents can simply observe each other's behaviour,   
   > then I'd say yes.   
      
   My understanding of collaboration is, that at least two agents engage   
   in a task-oriented activity and have an understanding that reaching an   
   individual or collective goal is unattainable by an individual agent.   
   Additionally, there must be at least one collective activity which   
   requires the contribution of all participating agents in this   
   collaboration. Every participating agent would have some understanding   
   about its own willingness, ability and effects towards the   
   collaboration.   
      
   My definition only requires at least one collective activity to be a   
   collaboration. I suppose, my original question was: Must this   
   collective activity be a communicational act or can it be another act?   
      
   Why is the answer in Shannon's sense "no"? His model suggests to have   
   a simple Sender - Receiver channel and the only influence to this   
   transmission is noise. So what you say is, if there is no such channel   
   then there is no way that the potential parties would even know about   
   collaboration.   
      
   > Suppose you and I are walking towards each other on a footpath/sidewalk   
   > heading for a collision.  We both notice this and deviate to avoid it,   
   > perhaps using a model of expected agent behaviour to both diverge to our   
   > respective left (or maybe right in the USA).  I'd call that cooperation   
   > without communication.  Here cooperation is just avoiding interference,   
   > while collaboration usually means working to jointly achieve a shared   
   > goal.  But you could even argue that a shared goal "avoid collision" has   
   > arisen by chance in this situation, so it is a case of collaboration.   
   >   
      
   I interpret the "no exchange of any information" caveat strict. This   
   includes behavioural observation. Especially in your "collision" case,   
   to reach the goal of not colliding, both parties must indicate   
   (communicate) to each other (through behaviour) which way to go.   
      
   > More generally, provided they can observe each other's behaviour, agents   
   > can learn to cooperate or collaborate without communication by building   
   > and refining models of each other's behaviour, knowledge and motivation,   
   > then using these to adjust their own behaviours.   
      
   This is very interesting! Who has published about such models of each   
   other's behaviour, knowledge and motivation (preferable not only   
   theory, but some evaluation)? I am not only interested in "behaviour   
   observation models" of other agents but also in models where the only   
   commonality may be language and protocols. I got the impression that   
   many Multi Agent systems are developed under the assumptions that half   
   of other agent's model is already implemented by the designer. Could   
   you confirm that?   
      
   cheers, Chris   
      
   [ comp.ai is moderated.  To submit, just post and be patient, or if ]   
   [ that fails mail your article to , and ]   
   [ ask your news administrator to fix the problems with your system. ]   
      
   --- SoupGate-Win32 v1.05   
    * Origin: you cannot sedate... all the things you hate (1:229/2)