The case for AGI by 2030

Abstract

Driven by increased computing power and algorithmic advances, AI models have shown rapid progress in recent years, transitioning from basic chatbots to systems capable of complex reasoning and problem-solving. Larger base models pretrained on massive datasets, combined with reinforcement learning techniques, have enabled models to achieve expert-level performance on scientific reasoning, coding, and other specialized tasks. Furthermore, increasing test-time compute allows models to “think” longer, leading to improved accuracy, while the development of agent scaffolding enables them to complete complex, multi-step projects. Extrapolating current trends suggests that by 2028, AI systems could surpass human capabilities in various domains, potentially accelerating research in fields like AI, software engineering, and scientific discovery. However, challenges remain in applying AI to ill-defined, high-context tasks and long-term projects. Growth in computational power and the AI research workforce are also expected to encounter bottlenecks around 2030, suggesting that transformative AI will either emerge by that time or progress will slow considerably, making the next five years crucial for the field. – AI-generated abstract.