From Symbolic AI to Generative Models: Tracing the Shifting Paradigms of Artificial Intelligence

Abstract

This review traces the evolution of Artificial Intelligence (AI) through its transformative paradigm shifts, beginning with the rule-based Symbolic AI approach. Rooted in logic and explicit knowledge representation, Symbolic AI excelled in problems requiring defined rules and reasoning, powering early expert systems. However, its limitations emerged when facing real-world complexities and uncertainty. This led to the rise of Statistical AI, leveraging data and probabilistic models to learn patterns and make predictions without explicit programming for every scenario. While excelling in tasks like classification and regression by identifying statistical relationships, this data-driven approach often operates as a 'black box', lacking the explainability characteristic of symbolic systems. The most recent paradigm shift has ushered in Generative Models, capable not just of analyzing data but also creating novel content, spanning diverse domains like text, images, and code. These models demonstrate unprecedented versatility and creativity, though they demand immense computational resources and raise new challenges regarding control, bias, and truthfulness of generated outputs. The transition across these paradigms has profoundly expanded AI's capabilities, moving from narrow, rule-bound applications to broad, data-intensive, and creative tasks, fundamentally altering its potential and societal impact. This review argues that these successive shifts represent not merely technological advancements, but fundamental changes in the computational understanding and implementation of intelligence, each addressing limitations of its predecessor while introducing new potentials and challenges.