Transcript Theory and mechanisms of social interactions in the big

Theory and mechanisms of social
interactions in the big data era
(May 6-8, 2013)
Maxi San Miguel [IFISC (CSIC-UIB)]
&
Anxo Sánchez [GISC-UC3M]
Why are we here?
-Robert Axtell (Computational Social Science, George Mason University, Fairfax)
-Guillaume Deffuant (Laboratoire d’Ingeniérie pour les Systèmes Complexes, Aubière)
-Bruce Edmonds (Centre for Policy Modelling, Manchester Metropolitan University)
-Andreas Flache (Department of Sociology, Rijksuniversiteit Groningen)
-Cars Hommes (CeNDEF & Department of Quantitative Economics, Univ. Amsterdam)
-Alan Kirman (GREQAM, Marseille)
-Jurgen Jost (Max Planck Institute for Mathematics in the Sciences, Leipzig)
-Tom Lenaerts (Computer Science Department, Université Libre de Bruxelles)
-Michael Macy (Department of Sociology, Cornell University, Ithaca)
Meeting designers:
-Maxi San Miguel (IFISC, CSIC-UIB, Mallorca)
-Anxo Sánchez (Departamento de Matemáticas, Universidad Carlos III de Madrid, Leganés)
Local (IFISC, CSIC-UIB) participants:
-Víctor M. Eguílluz -José J. Ramasco -Raúl Toral
What is this workshop about?
Predicting vs. Understanding
in
Social Behavior
Two conflicting views (I)
The new availability of huge amounts of data, along
with the statistical tools to crunch these numbers, offers
a whole new way of understanding the world.
Correlation supersedes causation, and science can
advance even without coherent models, unified
theories, or really any mechanistic explanation at all.
Two conflicting views (II)
The focus on data detracts from the need to get a better
theoretical grasp on social systems.
Note that even the very best of these computational
articles are largely focused on existing theories. That's
valuable, but it is only one piece of what needs to be
done.
So, what is the plan?
Brainstorm to discuss whether and how the
theoretical approach to the mechanisms at work in
social contexts can guide research leading to
breakthroughs in the social sciences taking into
account the availability of big data.
Aim: Establish a roadmap for the development
of theoretical social sciences in the big data era
We believe we need to know:
• What theories of human behavior and mechanisms
of social interactions are well established? Which
theories are most desperately needed? How are
these theories related to data?
• Economics, sociology, cognition: Can we really make
progress towards an applicable, unified body of
knowledge identifying what is needed from these
different perspectives?
We believe we need to know:
• Theory and models of emotions, expectations, trust,
meaning, their explanatory power in relation to data, and
data analysis oriented to assessing these behaviors.
• Simple versus elaborate models. Details vs universal
behavior. When can researchers resort to very simple
behavioral rules and when is it needed to include very
detailed models. What can be learned from each of them?
• Proposing theories from observational data, versus
designing data collection to validate theories.
Proposed tentative schedule
• Today, 19 h: Trust, emotions, thresholds, emotional
waves (Alan Kirman)
• Tomorrow, 9 h: Expectations (Cars Hommes)
• Tomorrow, 11 h: Heterogeneity, noise and simulation
(Rob Axtell)
• Tomorrow, 16:30 h: Computational models and big
data (Andreas Flache)
• Tomorrow, 18 h: Context dependence in behavior
and data (Bruce Edmonds)
• Wednesday, 8:30: Interactions, micro-macro and
collective effects (Juergen Jost)
• Wednesday, 10:30: Conclusions (Michael Macy)
But:
We can change it now
We can rearrange it on the fly
So, let’s start!
In this session the discussion is intended to identify key
points for the debate, initial disagreements or alternative
viewpoints and to plan the rest of sessions
Alan Kirman:
Trust, emotions, thresholds, emotional waves
Emotions and trust.
Big data as indicator of emotional waves.
Thresholds: reactions to collective behavior vs reacting to
individual ones.
Aggregate behavior as resulting from self-organization:
emergence vs typical agents.
Cars Hommes:
Expectations
Heterogeneous expectations as a source of instability, in
particular in financial/macroeconomic contexts: models
and paradigms.
Heterogeneous expectations vs heterogeneous agents.
Theory verification by experiments and data.
More on stylized vs detailed simulations.
Robert Axtell:
Heterogeneity, noise and simulation
Sorting noise from actual human variability in large data
sets, in order to incorporate realistic levels of agent
heterogeneity into models.
Tradeoffs in large numbers of simple agents vs a few
complex agents.
Data as an indicator of relevant simulation sizes.
Andreas Flache:
Computational modeling and big data
How can computational modelling be fruitfully related to
the analysis of big data?
Computational modeling to generate different theoretical
scenarios and to confront them with the data.
What and how much do we need to know about the
"content" of social interactions, to predict behavior of
social systems from data on their macroscopic structure
(e.g. complex network structure)?
Bruce Edmonds:
Context-dependence in behaviour and data
It is well known that many aspects of human cognition
and behaviour are sharply context-dependent including:
language, preferences, memory, social norms, visual
perception, judging trustworthiness and assessing
reputation. Thus we would expect to see this in both the
data and in our models of human behaviour. However
how to do this best is not clear, thus the discussion is on
ways forward to possibly: detect the influence of social
context, identify it, exploit it in our analysis and generally
deal with it.
Jürgen Jost:
Interactions, micro-macro, and collective effects
What behavior can be found empirically, how flexible and adaptive is it, how
does it influence and shape the collective processes, and to what degree
does it reflect or anticipate those processes? In my opinion, the focus
should not be on models of individual behavior, but on those processes of
interaction between the different levels.
Can there emerge qualitatively different types of dynamics? If so, what
triggers them, and what are the relevant bifurcation parameters and
threshold variables? Or in social terms, who is following whom, and by what
kind of behavior get individuals into a leadership position?
What types of collective dynamics 'enslave' the individual degrees of
freedom, that is, make the behavior of the individuals simpler than what they
might do in isolation?
Michael Macy:
CONCLUSIONS?
-Perspective paper for SCIENCE?
A discussion of the role of theory in the social sciences could be of great
interest to our readers. I am especially interested in the questions you raised
--"Which theories are most desperately needed? How are these theories
related to data?" The best next step would be for us to communicate after
the meeting in May so that we can discuss what would be included in the
manuscript. (B. Jasny, Science)
-Roadmap for the development of theoretical social
sciences in the big data era?
-List of selected hard and important questions to be
addressed in the near future?
Questions:
• Correct ordering of sessions?
• What is missing? (Tom?)
• What can be shortened?
• Suggestions for efficient discussions?
• Work groups, lunch dinner, evening, Tuesday informal
discussions?
•…?...?...?