Imagination

Abstract

Abstract This chapter explores David Hume’s views on the faculty of imagination. Hume had a particularly rich and ambitious take on imagination as a kind of mental laboratory in which new composite ideas could be created by fusing together old ideas in creative ways. This chapter explains how Hume’s theory, if successful, could rebut some of the most serious objections raised against empiricism by critics such as Jerry Fodor. It then discusses how generative adversarial networks can be seen to realize some of Hume’s most speculative and ambitious ideas. The chapter also discusses more recent and powerful generative architectures, such as those behind DALL-E 2 and MidJourney. It reviews how such architectures can be modified to favor novelty in their outputs to model more human-like kinds of creativity, and they can even be used to create simulated input for offline learning or complex planning for future outcomes, as is found in DeepMind’s Imagination-Augmented Agents (I2A) architecture.