Paper: The Room Hypothesis of Common Sense
The Room Hypothesis of Common Sense states that artificial common sense can be modeled using extra constraints on predicates similar to those used in Lojban.
These extra constraints assigns and uses sub-types on objects that the agent thinks about. The "room" refers to a finite number of objects for which speech-acts can determine whether common goals that the agent tries to achieve will fail.
In the view of The Room Hypothesis, common sense is closely linked to Zen Rationality, an extension of instrumental rationality with the ability for higher order reasoning about goals.
With other words, common sense is a way for a zen rational agent to "factor out" common terms in its utility function into a background theory of efficient higher order behavior. This factorization leads to evolved concepts that corresponds to speech acts in natural language.
This experiment is to test this hypothesis structurally, instead of using machine learning. The motivation is to derive which kind of constraints that occur naturally, such that these constraints can later be translated into machine learning problems.