Imagination-based Artificial Intelligence

Abstract

This paper highlights imagination-based artificial intelligence as a new and promising approach to artificial intelligence (AI). The characteristic features of imagination-based AI are a framing of target tasks in terms of affordances and an identification of those affordances using internal simulations. Systems incorporating the imagination-based approach have attained comparable performance levels to deep-learning-based models while having lower data requirements and greater explainability. Work remains to be done to develop the breadth of applicability of the imagination-based approach, and this class of AI systems raises philosophically interesting issues regarding the possibility of imagination in artificial systems.

Keywords: artificial intelligence; imagination; simulations; affordances; deep learning.

DOI: https://doi.org/10.21747/21836892/fil42a13