Monitor Network Members, Last week we convened the first Monitor Networks Conversation in Cambridge, with Josh Epstein as our guest. Josh's talk excited the 50 people present about the practice and promise of agentbased modeling, so we thought you might like to hear a few of the highlights. Before we began, we did a little "Wisdom of Crowds" exercise, asking each person to pick the outcome of the election he or she thought most likely. We asked them to pick the winner of the popular vote, the winner of the electoral vote, or to choose "Disputed Election," meaning some outcome reminiscent of the 2000 decision process. The question was asked as a forecast, not a preference. The outcomes are shown below: POPULAR Bush Bush Kerry Kerry Disputed Election
ELECTORAL Bush Kerry Kerry Bush
% 28 5 31 15 20
To me, the most interesting result is that the two "clean win" scenarios garnered only 59% of the vote--in other words, either a dispute or a split between the popular vote and Electoral College is the expectation of 40% of the people who attended. Given the reports about pre-positioning of legal teams in swing states by both parties, and the funding of militant 527 organizations, this points to a scenario for November that may even rival the Yankees-Red Sox series. Josh spoke for an hour or so. I won't try to do justice to either his style (and energy level that passeth understanding) or his substance, but here are some takeaways: o Agent-based modeling (ABM) is a technique that is ideal for modeling social systems because it doesn't require "averaging" of the behavior of the individuals in a group--each actor ("agent") can be uniquely specified, and the interactions among these heterogeneous populations can be simulated. ABM can be applied across a wide range of domains: Josh showed models of economic agents competing for "sugar", a stand-in for whatever the rewards might be in a particular system (e.g. demand in a market, compensation in an organization, dates in a social system, attention in a voting body...); of the contagion of smallpox in a town; of the paths visitors take through Disneyworld; and of the structure members of an organization would put in place in response to different market conditions. ABMs allow for a different style of explanation: rather than being satisfied with equations that accurately describe aggregate behavior, by specifying the agents' decision rules it is possible to "grow" the emergent outcomes; this means that modelers can play with the assumptions about the individuals and see the implications for the system. For example, investigators might be able to research the effect of changing compensation or tax systems in silico, with much more granularity that the typical "elasticity" approach economists have used for the last century. And, if you’re trying to design the tunnels of the stadiums for the Sydney Olympics (a real case) the average throughput is determined by how smoothly individuals can pass one another – and so is the chance of getting trampled.
o
o
The next week, we asked those who attended to share their reactions and thoughts around potential business applications where agent-based modeling could support executive level decisions. Below are some of the responses: –1–
47afadb3-f06d-4b23-a8ac-284a75ae5c30.doc 16-Nov-08 11:27
Supply Chain Supply chains need agility, adaptability, and alignment. What better way to understand the dynamics of options under consideration than through ABM? (N.B. Some major players -- both consumer goods companies and software providers -- have begun doing just that.) Mergers & Acquisitions The merging of two organizations is famously painful. Could ABM provide insights about reducing the pain? (N.B. A model created for Hewlett Packard “grew” certain aspects of their experience after merging with Apollo.) Competition How do you anticipate and respond to your competition? ABM could provide the organization with a more complete view of the competitive landscape and potential shifts. (N.B. Icosystems, a company in Cambridge doing ABM, has created a simulation of the ISP world, in which competitors “grew” pricing strategies) Branding & Messaging By projecting shifts in public opinion, media coverage and attitudes now could you use the information to craft more effective communication and branding messages? (N.B. Affinova uses techniques related to ABM (“genetic algorithms”) to “breed” successful packages and advertisements.) Traffic Patterns How could an ABM of Boston’s traffic have effected the outcome of America’s most complex, technologically challenging and most costly highway project, The Big Dig? (N.B. Funded by the Department of Energy, Los Alamos National Laboratory created a simulation of traffic in Albuquerque, NM. It showed – convincingly – that adding a light rail system would make pollution worse, because more trips would be taken in cars with cold engines.) Retail Location Retail stores have the daunting task of determining where crowds will gather. How, using an ABM, could Starbucks determine if putting 4 coffee shops on the corners of a single busy intersection make money? (N.B. Icosystem has simulated shoppers paths in supermarkets, successfully growing their patterns.) Healthcare Could agent based modeling shed light on ways to improve patient care while maintaining or improving profitability in long term care facilities? (Ok, I’m out: a proposal to develop this kind of model never got funded.) Our plan for the Monitor Networks Conversation series is to bring together insightful, provocative speakers and curious members of our network to engage in an exchange of ideas and experiences. To make this work for everyone, we don’t want to stop at providing the input – we want to create the conversation. We’ll be developing a platform to support this better, but for now, if you have reactions, let us hear from you. We hope you’ll be able to join us for future Monitor Networks Conversations. Our next meeting will be in January of 2005 and will feature Sherry Turkle of MIT. You’ll be receiving more information from us as we get closer to the event but in the meantime, save the date and feel free to share in the conversation. Best regards, Chris Meyer chris_meyer@monitor.com
47afadb3-f06d-4b23-a8ac-284a75ae5c30.doc
–2–