From: jimruttshow8596
Overview
Agent-based modeling (ABM) and simulation are tools used to explore complex systems. In the context of human systems, their applicability and limitations are distinct from their use in simpler, more predictable domains [05:10:00].
Utility and Limitations
Agent-based modeling and simulation can be useful when dealing with single agency and clearly defined rules [27:25:00]. However, their direct application to human systems is often problematic because most human systems involve:
- Multiple identities [27:31:00]
- Patterns not solely based on simple rule-based decisions [27:33:00]
- Highly interconnected systems where direct cause-and-effect may not be clear [05:10:00]
A significant concern with using simulation in human systems is the tendency to confuse simulation with prediction [27:39:00]. While simulations may be historically accurate, they are often problematic for foresight [28:36:00].
Murray Gell-Mann’s Perspective
Physicist Murray Gell-Mann famously stated that the “only valid model of a human system is a system itself” [28:20:00]. This suggests that while agent-based modeling can provide insights and clues into aspects of a system, it does not offer the predictive element that some might seek [28:24:00].
Ensemble Statistics and System Behavior
While individual trajectories from agent-based models may not be predictive, the statistics on the ensemble of trajectories can reveal crucial information about the system’s underlying dynamics [28:59:00]. For instance, these models can help determine if a system operates in a Gaussian (normal) or a fat-tailed (extremistan) space [29:05:00]. Understanding this distinction is vital for management, as Gaussian and fat-tailed spaces require very different approaches [29:20:00].
In a “Pareto world” (fat-tailed), the best an organization can do is to trigger human beings to a heightened state of alert to examine a situation [29:37:00]. This approach was applied in counterterrorism work for DARPA, where AI was used to trigger a heightened state of alert when a terrorist outrage was more likely, rather than attempting to predict or model the event directly [29:45:00]. This technology is being adapted to create “anticipatory triggers” for issues like potential abuse in elderly care homes [29:58:00].