This Working Paper reviews the current state of the art in data infrastructure and artificial intelligence approaches that could be valuable for social and behavioral modeling. Among the newer machine-learning methods, adversarial training and fuzzy cognitive maps seem to have particular unrealized potential. The Working Paper then discusses the troublesome theory-data gap: the mismatch between measurable data streams and meaningful explanatory theories to frame the data. The Working Paper identifies this issue as a key barrier to meaningful social and behavioral modeling. It then discusses the need to move from purely data-driven work to theory-informed work, and to tighten the iterative loop between theory and data analysis. Closing the theory-data gap is a general problem. The Working Paper illustrates three example models that attempt to integrate theory-informed and data-driven modeling: a network model, factor tree model, and a fuzzy cognitive map model. The first model addresses meme transmission. The last two address the public support for terrorism.The Working Paper identifies key questions and challenges along the way. These are questions that a notional social and behavioral modeling research community will need to tackle as it grows.
Osoba, Osonde A. and Paul K. Davis, An Artificial Intelligence/Machine Learning Perspective on Social Simulation: New Data and New Challenges. Santa Monica, CA: RAND Corporation, 2018. https://www.rand.org/pubs/working_papers/WR1213.html.
Osoba, Osonde A. and Paul K. Davis, An Artificial Intelligence/Machine Learning Perspective on Social Simulation: New Data and New Challenges, Santa Monica, Calif.: RAND Corporation, WR-1213-DARPA, 2018. As of September 08, 2021: https://www.rand.org/pubs/working_papers/WR1213.html