Conniving With Continuations: Representing Goals in a Domain‐Specific Language of Thought

Wanting composes flexibly with knowing: we can want to know, want to know what someone wants, and so on. In this paper, we develop a goal representation that allows for this type of rich integration between goals and other theory-of-mind concepts. We extend the memo programming language, which is specialized for theory-of-mind reasoning, with a new syntactic construct called "wants" that represents goals. To implement "wants," we borrow an idea from programming language theory: the notion of a "continuation," which allows reasoning about possible future states of a program. We show that the new "wants" construct can be composed with other memo primitives to represent a wide variety of interesting goals, and we discuss what the design of this construct can teach us about the concept of desire more broadly.