I sent the message below on last friday, but it bounced because I
was erroneously removed from the list. I re-send it now. I hope it is
still in tune with the discussion that is going in the list :-)
rosaria conte
=================
Warren Thorngate wrote
>How about suggesting that MABS folk make more use of what
>psychologists and social psychologists and sociologists know from
>experiment and observation about human behaviour in filling their models
>with assumptions? Example: rather than running MABS under the
>simple-to-programme assumption that every agent has the same motives,
>skills, knowledge, etc. -- which 100+ years of research in the social
>sciences has shown to be false -- how about building in human variability
>in all such constructs? This might help to constrain the exponential
>growth in MABS simulations.
I 100% agree especially with the first part of this paragraph. In the
last few years, the simsoc community has invested a lot in the task
to integrate the social scientific and the MAS community, and a
reasonable degree of success has been achieved especially with MABS.
Still, I think Warren is right when he says that the "ground"
assumptions in multiagent systems, and even in multiagent-based
simulation are not fully influenced by social scientists and social
psychologists. The question is why.
In the second part of the above paragraph, Warren suggests an
explanation. The example given indicates both what he considers a
wrong assumption, namely that agents have the same motives, and the
reason for the popularity of this assumption, namely that same-motive
agents are easy to programme.
Now, I agree that the assumption is wrong, and I won't comment more
about this point. Instead, I am not so sure that simplicity to
prgramme is the reason for its success. I bother the readers with
further consideration about this point because I think this point is
strongly related to Alan's further remarks about theory and models,
and to the rest of the discussion.
I think that Social Simulation works can and do model diversified
agents in terms of probabilistic distribution of different beliefs,
strategies, preferences, etc.. This is not very difficult to
programme. Instead, what is
still missing is a theory of social processes which accounts for and
predicts variability.
Despite the evolutionary paradigm - where variability plays a
fundamental role, according to some scientists (including social
scientists), variability is incompatible with theory. In their view,
abstraction implies the opposite of variability. At most, variability
can be postulated, as probabilistic models do, as a starting point,
which does not need explanation. The *sameness* assumption which
Warren complains about is therefore owed more to epistemological
reasons (although probably wrong) rather than to easy programming.
Now I come to the theory/models issue. Probabilistic models are
certainly useful. But they don't help achieve a general theory of
social processes. Quite on the contrary, a general theory of social
processes that accounts for variability could help us predict
differences and their distribution. Not only in the sense of the
evolutionary paradigm (which shows the global benefit of different
interactional strategies). But also, and moreover, in the sense of a
theory of the processes from social structures to agents' motives,
knowledge, etc. A general theory like this would allow to predict
differences among agents, rather than simply and blindly postulate
them.
One last comment about variability and abstraction. As I said, I am
strongly convinced that the sameness assumption is wrong if we refer
to the agents' motives, beliefs, etc.. However, I also think that
there are important uniformities in the way agents work, learn, form
representations, and decide upon them. A general theory of these
aspects of agency, together with a general theory of social processes
is necessary to make predictions from the social proceesses to the
agents' minds, which in my view is the only way to have theory-based
modelling of social matters.
Rosaria Conte
National Research Council, Institute of Psychology, V.LE Marx 15, 00137 Roma.
Division "AI, Cognitive and Interaction Modelling"
PSS (Project on Social Simulation) - voice:+39+06+86090210;fax:+39+06+824737
email: [log in to unmask] - http://ip.rm.cnr.it
University of Siena - Communication Sciences - "Social Psychology"
PLS, NOTICE THAT MY EMAIL ADDRESS HAS CHANGED: YOU MAY WANT TO UPDATE
YOUR ADDRESS BOOK. THANKS.
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|