Abstract

Doubts are raised about Dretske’s assumption that an entity can't have a representational state that governs its behavior in virtue of its content unless that internal state has been acquired via appropriate interaction with its environment. The doubts hinge on a subtle distinction between a system's acquiring an internal representational state and a system's internal state acquiring the property of being representational. Employing this distinction, it is suggested that we can pre-load machines with states "destined" to acquire specific, predictable and efficacious representational properties. The picture that results renders Dretske's general naturalistic approach to mental representation consistent with a tempered version of the classical dream in AI.