Path to Artificial General Intelligence: Past, present, and future

Abstract

During the past decade, there has been remarkable progress in Artificial Intelligence (AI). More recently, the emergence of Generative AI was an inflection point in cognitive pattern understanding and generation across multiple modalities including speech, text, imagery, and vision, where AI systems are increasingly matching or surpassing human performance on a growing array of cognitive tasks. These models have been seamlessly integrated into numerous applications and products, reaching hundreds of millions of users. As a result, discussions regarding the achievement of Artificial General Intelligence (AGI) have shifted from theoretical speculation to a plausible near to mid-term objective. In this paper, we present a comprehensive review of the evolution of AI from its inception to the present day. We then examine how advances in computational infrastructure, algorithms, and large-scale modeling are converging to drive the generative AI revolution and shaping the trajectory toward AGI, potentially within the next 5-to-10 years. Specifically, we analyze recent progress in compute capabilities, learning algorithms, and model architectures across a broad spectrum of cognitive tasks. We also share our perspective on the key challenges that remain to be solved, and discuss the critical risks that must be addressed to ensure the safe and beneficial development of AI systems that may eventually exceed human-level performance in perception, reasoning, and general cognition.