The Agents Are Coming
Donald F. Mango
If you have seen the enormously popular software The Sims, you were probably intrigued, but you may not have realized that you were looking at part of a scientific revolution involving adaptive agents. Adaptive agents are software entities that, when placed in a computer environment, monitor the state of that environment, and, armed with rules of behavior, interact with it. In the case of The Sims, the agents are people of various types, placed in a house or hotel or city, interacting in ways both mundane and hilarious. One of the key elements to the popularity of The Sims is its complete unpredictability. There is literally no way to predetermine the aggregate result of placing a certain set of Sims in a certain environment with certain initial conditions. The only way to find out what will happen is by watching the situation unfold.
The adaptive agent paradigm (AAP) is gaining momentum in many areas of science: financial markets, water policy, organizational and network theory, and the military. In fact, leadership methods used in Project Iraqi Freedom were developed using adaptive agentsthe "army of one" philosophy. To jump start your exposure, see the outstanding set of papers presented at the National Academy of Sciences 2001 Sackler Colloquium, "Adaptive Agents, Intelligence, and Emergent Human Organization: Capturing Complexity through Agent-Based Modeling," online at www.pnas.org/content/vol99/suppl_3. Another good site is "Agent-Based Computational Economics" at www.econ.iastate.edu/tesfatsi/index.cfm?fa=ace.
In the financial and economics world, AAP is developing alongside behavioral finance. Both schools are challenging the neoclassical economics assumptions of completely rational, homogeneous representative agents, operating in equilibrium, maximizing their utility, resulting in analytically solvable models for the economists. AAP involves concepts of bounded rationality (i.e., potentially inconsistent behavior), heterogeneous agents, and self-organized or emergent complexity (unpredictable patterns and aggregate results) where problems require computational solution.
The philosophy of science in general has not caught up with this notion of computational solution. It potentially represents a third valid "symbol system" for research and scientific communication (the others being Verbal and Mathematical).
Impact on Actuarial Science
What might this mean for our science? Here are some speculations:
1. Forced group interaction (similar to multiplayer gaming): the days of the isolated practitioner are limited.
2. Software as research product: software will have formal recognition as a communication medium. The CAS may institute calls for software alongside calls for papers.
3. Policy analysis: regulators could use this as a means of testing the impact of policy changes (e.g., fair value accounting).
4. Test Impact of Changes: Allow testing of the aggregate effects of changes in rates, class plans, laws, or agent compensation in order to determine beforehand the likely impact of changes.
The difficult part in all adaptive agent modeling is modeling the behavioral rules of the agents themselves. For example, in order to model the policyholder retention impact of changes in rates, you might need a demand curve. While it may appear daunting, recent research in behavioral finance indicates that good results are achievable from modest assumptions. Most people's decision processes can be modeled reasonably well by a few simple rules. The key task for actuaries is to begin formulating those rules.