Complexity Control in Artificial Self-Organizing Systems: The Case of Bottom-Up versus Top-Down Intervention When Managing Pandemic Contagion.

Abstract

We model an adaptive agent-based environment using selfish algorithm agents (SA-agents) that make decisions along three choice dimensions as they play the multi-round prisoner's dilemma game. The dynamics that emerge from mutual interactions among the SA-agents exhibit two collective-level properties that mirror living systems, thus making these models suitable for societal/biological simulation. The properties are: emergent intelligence and collective agency. The former means there is observable intelligent behavior as a unitary collective entity. The latter means the collective exhibits observable adaptability that enables it to reorganize its network structure to meet its objectives in response to a changing environment. In this study, we generate these capabilities in a single, simple case. We do this first by letting a temporal complex network among SA-agents emerge and second by changing conditions in the ecosystem to test adaptability. This latter phase is done by introducing an artificial virus that infects SA-agents during interactions and can remove (or 'kill') the SA-agents. We then study the dynamics of the contagion within the collective as the virus spreads through the population and impacts collective reward-seeking performance. Specifically, we compare two strategies to control the spread of the virus: exogenous top-down control and endogenous bottom-up self-isolation strategies.