An Information-Theoretic Unification of Intelligence

Liu Peng, Yaochu Jin

Abstract:

This paper introduces a conceptual framework that unifies biological and artificial intelligence under a single principle:
intelligence is a system's capacity to reduce environmental uncertainty through information processing.
The primary value of this framework is not as a direct measurement tool,
but as a lens for analyzing AI paradigms,
identifying current limitations,
and charting future research directions.
By emphasizing the strategic acquisition and effective utilization of information,
our framework offers a principled approach to developing more robust and broadly intelligent systems.
We apply this perspective to diverse AI paradigms,
from expert systems to deep learning, and use it to advocate for advancing AI through enhanced ``information awareness,''
explore the human-AI co-evolutionary dynamic, and outline strategies to amplify AI's real-world impact.