Synergistic human-agent methods for deriving effective search strategies: the case of nanoscale design

摘要

Complex systems are challenging to understand and design, and even more so when considering nanoscale reasoning. This paper introduces a synergistic cognitive and agent-based methodology for deriving effective strategies for human searches in optimization design tasks. The method consists of conducting cognitive studies to determine effective human search approaches, rapidly testing algorithmic variations of human strategies using software agent automation, and finally providing the highly effective agent-refined strategies to humans. The methodology was implemented by developing a graphical user interface (GUI) of myosin biomotors and conducting a baseline cognitive study to determine how users effectively search for optimal biosystem designs. The best human designers typically searched local to their current best solution, utilized univariate searches, and may have learned and applied parametric knowledge. These trends informed rule-based agent strategies, and testing variations of rules resulted in the discovery of highly effective strategies using initial random searches, univariate searches to learn parameter relationships, and greedy local searches to apply knowledge. The GUI was modified to aid users in implementing two of the highest performing agent strategies in a final cognitive study. These users provided with the agent-refined strategy performed better than users with no provided strategy during the baseline cognitive study. When agents and users were provided myosin domain knowledge prior to searching, convergence on high-quality designs occurred earlier, which suggests that even experts in the domain could benefit from the agent-derived strategies. These findings demonstrate the power of synergistic human- and agent-based approaches, in which cognitive-based findings can reveal strategies that are refined by agents that generate search strategies for greatly improved user performance. The synergistic methodology extends beyond nano-based applications and could generally aid designers in discovering effective decision-making approaches across a broad range of domains.