Excellent posting on Naqniq about US government systems designed to forecast dissent. It tracks the history of ICEWS and includes some critical comments.
According to Wired, Darpa believe that “increasing sophistication of agent based social simulations…[combined with the>…explosion of new data sources” means we are on the verge of breakthroughs in computational models for the social sciences. “We may revolutionize the social sciences along the way.”
Well, yes, they may. They revolutionised society with the internet, after all.
Attempts to build such a system date back at least to 1976 . (“This report summarizes the progress toward the development of an integrated crisis warning system. During the first phase of this contract year, a fully integrated system comprised of (1) quantitative military, political, and economic crisis indicators; (2) quantitative indicators of U.S. military, political, and economic interests abroad; (3) a unified multi-method forecasting capability; and (4) a computer base was developed. The range of crisis indicators found within the integrated system includes internal (domestic) and external (international and global) static and dynamic military, political, and economic indicators. The system is also comprised of quantitative indicators of U.S. national interests. The unified multi-method forecasting capability requires the system to generate different kinds of forecasts or warnings via different methods for different events and conditions. The systems computer base must be capable of efficiently storing, retrieving, processing, and displaying large quantities of information. An integrated system should enable a user to generate forecasts or warnings on regional, country-by-country, or national interest bases. It should also permit more specific, country-by-country scanning, as well as examination of specific indicators in the context of specific situations over extended or relatively brief periods of time. Progress regarding the development of the prototype crisis early warning system may be understood in terms of the identification and specification of quantitative indicators for crisis warning, an associative forecasting capability, and a computer base”)
Then there was VISTA: “based on a multi-agent network model that incorporates multiple interacting and adaptive elements (agents) that represent different entities (e.g., Red and Blue forces) and regions of a city. The agents react to events depending on their characteristics, history, and connectivity, and system behavior emerges from this agent interaction. Through this simulation, the VISTA system enables exploration of potential outcomes and consequences of hypothetical actions and events, and the determination of when conditions are ripe for emerging threats.”
And, from the company that brough you VISTA, also ACUMEN: “based on the integration of social network analysis (SNA) with multi-agent methods that are sensitive to spatial and temporal effects to simulate state-failure-related dynamics at state and province levels. Theories from psychology, social psychology, sociology, organization science, political science, and economics contributed to the construction of the ACUMEN data requirements and underlying simulation engine….ACUMEN modeled political, military, social, religious, and insurgent groups as agents, along with their relationships regarding hostility, support, membership, and more. ACUMEN modeled the profiles of agents and geographic regions (at the state and province levels) within specific test states using a set of social, political, economic, health, and demographic indicators. In all there were 150 indicators for the state, 60 indicators for each province, and 30 indicators for each agent. ACUMEN specified agents’ geographic relationships (at the state and province level), along with a database of actual events within the state taken from open-source databases and news agencies…. In ACUMEN, nine different indicators of state failure were modeled. These included: lack of state legitimacy; lack of essential services; level of foreign military activity; level of criminal activity; level of terrorist activity; level of corruption; level of tension across groups; level of hostility across groups; and potential for province secession. These nine indicators were weighted so analysts can tailor their analysis to their particular sets of concerns or areas of interest. The overall failure indicator can be applied not only at the state level, but also the province level. Moreover, drill-down analyses and forecasts can track the behaviors of individual component factors within the set of nine. ”
And this blog has drawn attention to Tangram, which focuses perhaps more on tactical analysis., and the Sentient World Simulation. Plus MNE4, though I mention this with some trepidation as Im still not sure what exactly it was, despite reading quite a lot of material about it.
Theres also machines which simulate our environment (eg the Earth Simulator), but they focus more on climate etc than on human activity. However the two are just differnt ends of a spectrum and meet somehwere in the middle around issues of food, economic activity, wealth and water.
Seems to be a recurring theme at the moment, approaching the same questions from the academic, artistic, game and now government angles. The dream that we can predict (and therefore influence) the future.
JASSS had an excellent piece on the difficulties of sharing simulations. At the moment, all these things seem to be built in silos (which is not a bad thing since it allows you to compare thir results more independently) but think of all that duplicated effort collecting information!
Incidentally theres a new software package out called SimplePie which seems to be a good but simple RSS feed aggregator: an excellent base for anyone wanting to star their own massive model of world opinion or events. Collects data from RSS feeds and even provides parsing tools to help you to munge it.
Its a tremendous opportunity. In fact, its a whole set of tremendous opportunities: to use a new technology to illuminate corners of our existence, by taking a new look at existing data. But I suspect well get more, and better, results by starting small, like the Steven Levitt and David Phillips examples I quote in the last link. Now they are interesting and potentially useful.