Abstract
Some social robots use simulations to make sense of their environment, and it has been claimed that the use of simulations allows these robots to adapt better to novel scenarios. This paper formulates a logic to model reasoning about simulations run by artificial intelligence (AI) systems, with the goal of assessing the prospective capabilities of simulation-based robots. The semantics for sentences about simulations are modelled by a possible-worlds framework with a variably-strict accessibility relation. It is argued that worlds in simulation logic are best represented by logically incomplete sets of sentences with limited closure under classical entailment. The accessibility relation is intended to delineate worlds in which the explicit stipulations of a simulation hold, as well as the relevant features of our world. Possible constraints are discussed to align the accessibility relation with its intended interpretation. The notion of relevance is observed to be crucial for simulation logic, and represents a present limitation for the generalisation of simulation- based robots. Possible ways of overcoming this limitation are suggested, as well as possible avenues for philosophical investigation.