Abstract

Can machines truly think, reason and act in domains like humans? Thisenduring question continues to shape the pursuit of Artificial GeneralIntelligence (AGI). Despite the growing capabilities of models such as GPT-4.5,DeepSeek, Claude 3.5 Sonnet, Phi-4, and Grok 3, which exhibit multimodalfluency and partial reasoning, these systems remain fundamentally limited bytheir reliance on token-level prediction and lack of grounded agency. Thispaper offers a cross-disciplinary synthesis of AGI development, spanningartificial intelligence, cognitive neuroscience, psychology, generative models,and agent-based systems. We analyze the architectural and cognitive foundationsof general intelligence, highlighting the role of modular reasoning, persistentmemory, and multi-agent coordination. In particular, we emphasize the rise ofAgentic RAG frameworks that combine retrieval, planning, and dynamic tool useto enable more adaptive behavior. We discuss generalization strategies,including information compression, test-time adaptation, and training-freemethods, as critical pathways toward flexible, domain-agnostic intelligence.Vision-Language Models (VLMs) are reexamined not just as perception modules butas evolving interfaces for embodied understanding and collaborative taskcompletion. We also argue that true intelligence arises not from scale alonebut from the integration of memory and reasoning: an orchestration of modular,interactive, and self-improving components where compression enables adaptivebehavior. Drawing on advances in neurosymbolic systems, reinforcement learning,and cognitive scaffolding, we explore how recent architectures begin to bridgethe gap between statistical learning and goal-directed cognition. Finally, weidentify key scientific, technical, and ethical challenges on the path to AGI.