Behavior, Computation and Networks in Human Subject Experimentation A NetSE Workshop held at UCSD July 31

1
Behavior, Computation and Networks in Human
Subject ExperimentationA NetSE Workshop held at
UCSDJuly 31 August 1, 2008

• Michael Kearns
• Computer and Information Science, Penn
• Colin Camerer
• Economics, Caltech

2
Observations on Recent Trends and Activities

• Explosion of CS interest in economic/game-theoreti
  c models
• complexity of equilibrium computations
• realization that modern technological networks
  are economic systems
• selfish routing, free-riding in peering networks,
  ISP game theory
• economics as a design principle for complex
  systems
• algorithmic mechanism design
• emerging electronic markets
• sponsored search, Internet advertising, affiliate
  networks
• Algorithmic Game Theory (AGT) scaling up
  classical game theory/economics
• AGT edited volume Cambridge University Press 2007
• Significant CS-Econ interactions over several
  years
• Hal Varian invited lecture ACM EC 2006
• Many CS plenary talks at last two GT World
  Congress e.g. Tim Roughgarden Nash Lecture 2008
• Emerging interdisciplinary resources/programs
• Joint postdocs at Caltech 3 new CS faculty hires
  in algorithmic game theory at Northwestern
• Observation Practically all AGT work and CS/Econ
  interaction is theoretical in nature

3
Goals of the Workshop

• Bring together researchers with behavioral and
  computational interests
• Introduce a behavioral component to AGT
• Examine how AGT can inform/shape behavioral
  experiments
• Discuss opportunities and challenges for
  larger-scale human subject experiments
• CS/AGT strength system design, theories of
  scaling, network models
• BGT strength experimental design and
  methodology, network models

4
Sample Brainstorming Topics

• How do computational considerations modify BGT?
• How do behavioral considerations modify AGT?
• How can we scale up behavioral experiments?
• examination of (social) network effects
• use of web/Internet technology
• development of shared experimental
  infrastructure/platforms
• If there were no technology/methodology limits,
  what experiments would we do?
• How can we mix human subjects and artificial
  agents in interesting ways?
• What are the appropriate (statistical) models for
  collective behavioral data?

5
Workshop Outputs Research Challenges

• Unifying Algorithmic and Behavioral Game Theory
• Subfields both seek to modify classical GT
  towards realism
• Algorithmic GT realism computational
  tractability
• Behavioral GT realism conformance with and
  prediction of actual behavior
• Ideally, two should be unified and inform each
  other
• AGT must consider strategic computation related
  to behavior, not traditional complexity
• BGT must develop computational models of behavior
  (e.g. Camerers cognitive hierarchy)
• Network and Systems Infrastructure for Behavioral
  Experiments
• Internet and Web have opened the possibility of
  scaling up behavioral experiments in sociology,
  game theory, economics, CS,
• Current status many independent groups build
  one-off systems for specific experiments
• Little/no sharing of code, infrastructure, human
  subjects, data, methodologies, etc.
• Should we have a shared Web/Internet-based
  platform/infrastructure for large-scale
  behavioral work?
• How general/specific should it be? What about
  control/management of subjects?
• For more detail/info or to give
  suggestions/ideas
• Email mkearns_at_cis.upenn.edu
• See workshop report to NetSE