Semantic Mutation Testing For Multi-Agent Systems

This paper introduces semantic mutation testing (SMT) into multi-agent systems. SMT is a test assessment technique that makes changes to the interpretation of a program and then examines whether a given test set has the ability to detect each change to the original interpretation. These changes represent possible misunderstandings of how the program is interpreted. SMT can also be used to assess robustness to and reliability of semantic changes. This paper applies SMT to three rule-based agent programming languages, namely Jason, GOAL and 2APL, provides several contexts in which SMT for these languages is useful, and proposes three sets of semantic mutation operators (i.e., rules to make semantic changes) for these languages respectively, and a systematic approach to derivation of semantic mutation operators for rule-based agent languages. This paper then shows, through preliminary evaluation of our semantic mutation operators for Jason, that SMT has some potential to assess tests, robustness to and reliability of semantic changes.