Measuring Machine Intelligence Using Black-Box-Based Universal Intelligence Metrics

Measuring the machine intelligence quotient (MIQ) of intelligent agent-based systems (IABSs) is very important based on the increasing number of intelligent systems applied to real-life problem solving. The most important property of an intelligence metric must be its universality. Developing universal intelligence metrics is difficult based on the very large diversity of intelligent systems. A feasible approach for ensuring the universality of measuring machine intelligence consists in using black-box-based methods able to measure the central intelligence tendency in problem solving. This paper represents a guide for choosing the most appropriate black-box-based intelligence metric for measuring the intelligence of developed IABSs, classification of IABSs in intelligence classes and detection of the IABSs with statistical low and high outlier intelligence. In research where the performance of heuristic and metaheuristic algorithms is studied, the performance indicator is frequently calculated as the mean or the median of experimental evaluation results. There is no consensus agreement regarding which of them is more appropriate. In some cases, both of them are reported. The manner in which it should be decided which of them to be used is scientifically grounded in this paper.