Predicting Error-Resilient Collective OutcomesTM in Wireless Networks ACSAC Works in Progress Submission*

Arnold B. Urken
Professor of Political Science
Wireless Network Security Center, Stevens Institute of Technology

Errors caused by breakdowns in network communications channels and/or decision making failures are a fundamental obstacle to producing intelligence in wireless computer environments that is precise, accurate, and reliable. Previous research has demonstrated that collective decisions can be designed to save time and overcome decision making errors despite malicious and inadvertent errors in wireless communications. Whether the decision makers are humans and/or nodes, error-resilient collective outcomes (ERCOs) can be produced to provide a strategic and tactical information advantage in asymmetric warfare.

For example, suppose that 10 sensors vote on a binary choice (between A and B) and submit their votes to a central host. If the aggregation rule is majority and commander has received 6 votes in favor of A, A would be an ERCO because the outstanding votes could not possibly change the collective outcome - even if it changed the collective score. "A" would be a stable outcome regardless of the effects of delay, network link breakdowns, or imperfect sensors. A challenge of ERCO analysis is to determine if and when ERCOs can be produced for more complex decision tasks.

Although previous research(1) has illustrated the potential advantages of ERCO analysis in simple as well as more complex multidimensional decision tasks and alternative voting rules, the comparative static illustrations did not investigate when and how frequently ERCOs can be expected to occur. The current research addresses these issues by conducting Monte Carlo simulations of voting processes. In these pilot experiments, a five-choice agenda is studied with fixed and random variables. The fixed variables include the number of voters, the voting method (rule that determines the number of votes that can be used to express a preference or judgment), and the aggregation rule (plurality and unanimity in addition to majority). The random variables in the simulation include the distribution of preferences, reliability - or competence - of the decision makers, the order in which votes arrive at the central host, the time (represented by a Rayleigh distribution) associated with the arrival of each vote, and the individual vote rate of false positive and false negative decisions. Tied collective outcomes are either not resolved or are .broken. using random techniques.

Preliminary results of these experiments show how much voting information must be received and/or how long one must wait until an inference about the collective outcome can be reached with a high probability. Under certain conditions, the probability of producing an ERCO is very high even though a high proportion of votes are outstanding or only a few seconds has passed.

ERCO analysis can be used to design collective decision system support for humans, sensors in client-server or peer-to-peer networks. The presentation will discuss applications of more recent results to analyze practical security scenarios involving the deployment of sensors and coordinated human management of emergency responses. Future plans for experimental testing and investigation of complex decision tasks will be discussed.

(1) Arnold B. Urken, .Time, Error, and Collective Decision System Support,. in Proceedings of the International Conference on Telecommunications Systems, Monterey, October, 2003.

* The technology and methodology described in presentation is patent pending.