2008 Graduate Workshop in Complexity and Computational Social Science
Student Projects
Each student began a research project during the two-week
workshop. Below are brief descriptions of these various projects.
These projects will form the basis for dissertation chapters and/or
journal articles.
Jerusha Achterberg, Anthropology, U. of Washington (jerusha@u.washington.edu)
Jerusha is exploring the dynamics of communicable diseases in
small populations. In particular, she is interested
in whether Tuberculosis (TB) could become endemic in pre-agricultural settings.
Using an agent-based model she modeled a small populations with varying contact
structures tied to a few parameters, such as population size, household
demographics, birth rates, etc. The model demonstrated that under the right conditions,
TB can become endemic even in
a single, small foraging population.
Nadav Aharony, Media Lab, MIT (nadav@mit.edu)
Nadav manipulated the social networks used by Anigraf agents (see Richards) to
improve the behavior of the implied autonomous agent. He found that
small changes in the social networks can lead to large changes in system-wide
behavior. With small numbers of agents, the social networks can be fully
enumerated and the key driving components and neighborhood structures can
be identified. For larger numbers of agents, Nadav is implementing a
genetic algorithm to explore the resulting space.
Matthew Backus, Economics, U. of Michigan (backus@umich.edu)
Matt considers a world in which the actions of others are privately
and imperfectly monitored. Agents are limited in their ability to remember
the past though, conditional on the state of their memory, they can optimize
their behavior. He finds that this approach can start to reconcile the existing work
in public monitoring with the private case.
Christos Ioannou, Economics, U. of Minnesota (ioannou@umn.edu)
Christos is refining our understanding of how various types of noise
can impact strategic behavior. He evolves automata playing the Prisoner's
Dilemma in the presence of either implementation (an error in executing your action)
or perception (an error in the reported action of your opponent) noise.
He finds that noise makes cooperative outcomes less likely in the game. Moreover,
depending on the type of noise, he finds fundamental changes in the types of
strategies that evolve in the system.
Brad Leveck, Political Science, UC San Diego (bleveck@ucsd.edu)
Brad is developing a theory of how "advice" impacts the behavior
of a social system. The model has advisees receiving advice from
multiple advisers. The advisers observe the world and formulate their
advice, while the advisee decides on how best to follow these recommendations.
Over time, advisees update their future decision making based on the realized
quality of the advice.
Preliminary results indicate that advisees tend to focus
their attention on only one of their advisers across a broad
range of scenarios.
Peter McMahan, Public Policy, U. of Chicago (mcmahan@uchicago.edu)
Peter wants to understand the evolution of public-good norms, such as
social prohibitions against drunk driving, littering, etc. Agents classify
external events in the world into equivalence classes, and using these classes they
decide what norms to adopt. Norms diffuse among the population when agents
use their observations of their neighbor's actions to alter their own
behavior. He finds that "societies" can evolve toward either strong- or weak-normed regimes
depending on social ties and starting conditions.
Jacob Montgomery, Political Science, Duke (jmm61@duke.edu)
Jacob is investigating an ecological model of political parties. Two populations of
candidates form two parties and establish a party core position. Activists contribute to the party they favor
and these resources are distributed to the party's candidates based on how closely the candidate
aligns with the party's core position.
Depending on the parameters, the system's behavior can range from stable to highly unstable equilibria.
He is currently exploring the full dynamics of the system in response to various policies and behavioral assumptions.
Julian Romero, Economics, Cal Tech (jnr@hss.caltech.edu)
Julian is looking at coordination behavior on a network. Each player (node) attempts
to have a different "color" than that of each neighbor, for example, a manufacturer might
want to differentiate its products from those of its rivals on a grocery shelf via unique packaging.
In the model, each node must pick a color from a set of allowable colors when it is asynchronously
called upon to choose. Julian implements
different choice algorithms, ranging from random alterations when needed to strategies driven
by quantal response functions. He varies the difficulty of the networks by manipulating the
number and type of connections among the nodes. He finds interesting patterns of solution efficiency
(measured by the number of color changes needed to find an equilibrium) as a function of graph structure and
strategic ability.
Paul Smaldino, Psychology, UC Davis (pesmaldino@ucdavis.edu)
Paul is interested in a dynamic description of the exploratory behavior
of rats (rattus norvegicus). In particular, he is looking at differences
between wild-type rats and Brattleboro rats (which have a defective
gene for the synthesis of vasopressin). Using some novel visualization
techniques, he first characterized the existing data. To model the
resulting observations, he created a autonomous agent moving in
an analogous space, and embodied the agent with biologically-motivated
movement proclivities, such as continuing to move in the same
direction, staying near walls, etc. Using this model he can
begin to match its behavior to the observations, and ultimately
make predictions about rat behavior in novel maze environments.
Justin Smith, Economics, U. of New Mexico (jthsmith@unm.edu)
Justin is attempting to improve learning classifier systems (LCS) by introducing new mechanisms, based
on recent developments in economic theory, into
the credit assignment algorithm. The current LCS
algorithm uses probabilistic bidding that tends to destabilize prices and result in
inefficiencies. Justin introduces 1) a betting brigade mechanism that allows classifiers to pool
their bids and 2) a means by which classifiers can express risk preferences.
He tested this new algorithm on Riolo's test bed and found that the system was able to
clearly identify, and reward, the key classifiers needed to solve the underlying problem.
Daniel Villatoro Segura, Computer Science, Spanish Scientific Research Council (dvillatoro@iiia.csic.es)
Daniel created a model of group formation driven by social norms. Agents acquire resources
and must decide whether or not to share what they have with others. Agents form tighter
connections with those agents with whom they share behavior based on their prior interactions.
He also varies the amount of information that flows among the agents via secondary reports
on the actions of others (either via a white or black list), and finds that while white lists
have little impact, black lists can improve the diffusion of positive norms.
John H.
Miller , miller@santafe.edu.