Re: [glob2-devel] Globulation Lore v1.0

[Stéphane Magnenat]

On Sunday 25 January 2009 15:41:13 Federico P wrote:
> On Saturday 24 January 2009 13:19:48 Stéphane Magnenat wrote:
> > About the emergent/complex-system-ish theories, my view is the following:
> > You can perfectly have the implementation of the reasoning distributed
> > over the individual physical agents but a unique hive mind abstraction.
> > In the end, until proven otherwise, that seems to be the way that our
> > brains work. This has the additional goodies to allow scenarios where the
> > hive mind is splitted at some point but re-united afterwards.
>
> ha! This comes from starcraft Brood War campaign! ^^
> But we don't have a building like the OverMind which designate the mind
> abstraction, and couldn't create such campaign unless changing a lot the
> whole concepts.

My whole point is that you *don't* need a building. You can distribute the 
implementation of the unified overmind over all the physical units. Research in 
robotics is not exactly there but we are working on it.

> Actually an overmind is exactly a refuse to consider emergency... 

Yes, but considering emergency as a miracle solution to engineer devices shows 
a lack of sense of reality. I know this, I have worked in this field for more 
than two years and co-authored a paper on emergent behaviours for swarm of 
robots [1].

Ants work, for sure, but they are ultra-complex devices with thousands of 
sensors, hundreds of actuators, and one-quarter million neurons. Ants have 
been produced using the physical engine of reality by evolving over the whole 
earth for five billion years. Do not underestimate the amount of FLOPS that a 
small portion of space itself can perform.

As comparison, robots in [1] have two sensors (14 inputs), two actuators, and 
three neurons. Their rather stupid behaviours took weeks to evolve on state of 
the art computers using a custom-tailord simulator [2, 3].

So, as a conclusion, let us say that yes, emergence is a very interesting 
phenomenon, but that our models of it and its usefulness as an engineering 
tool are so weak that we could as well try to become immortal for the same 
price.

So I still vote for the distributed overmind. Have a nice day,

Steph


[1]     Evolutionary Conditions for the Emergence of Communication in Robots
Dario Floreano, Sara Mitri, Stéphane Magnenat, Laurent Keller. Current 
Biology, vol. 17, pages 1–6, DOI: 10.1016/j.cub.2007.01.058, 2007.
"http://stephane.magnenat.net/data/Evolutionary Conditions for the Emergence 
of Communication in Robots - Dario Floreano, Sara Mitri, Stéphane Magnenat, 
Laurent Keller - Current Biology - 2007.pdf"
[2]     http://mobots.epfl.ch/enki.html
[3]     http://lis.epfl.ch/resources/teem/index.php

-- 
http://stephane.magnenat.net