OpenAI CEO Sam Altman Predicts AI Will Generate Novel Insights by 2026

In his latest essay titled “The Gentle Singularity,” published on Tuesday, OpenAI CEO Sam Altman outlined his vision for how artificial general intelligence (AGI) will transform the human experience over the next 15 years. Altman, known for his forward-thinking and strategic insights, often envisions a future where AGI reshapes sectors like work, energy, and the social contract. However, his recent essay includes specific predictions about what OpenAI might achieve in the near future. One significant claim in the essay is that by 2026, the world could witness the emergence of AI systems capable of generating novel insights. This notion, though somewhat broad, aligns with recent statements from OpenAI's leadership. For instance, in April, co-founder and President Greg Brockman introduced the o3 and o4-mini AI reasoning models, which he noted were the first to produce new and valuable scientific ideas. Altman's latest post indicates that OpenAI will intensify its efforts towards developing such AI models. However, OpenAI is not alone in this pursuit. Competitors are also investing heavily in this area. Google, for instance, unveiled AlphaEvolve in May, an AI coding agent that has reportedly generated innovative solutions to complex mathematical problems. FutureHouse, a startup backed by former Google CEO Eric Schmidt, claims that its AI agent has made a genuine scientific discovery. Additionally, Anthropic launched a program in May aimed at supporting scientific research through AI. If successful, these advancements could revolutionize key aspects of the scientific method, particularly hypothesis generation and experimentation. Industries such as drug discovery, material science, and others rooted in scientific exploration stand to benefit significantly from such automation. This isn't the first time Altman has foreshadowed OpenAI’s upcoming projects through his blogging. In January, he suggested that 2025 would be the year of AI agents, and OpenAI followed through by releasing its initial trio of AI agents: Operator, Deep Research, and Codex. Each of these agents was designed to handle different tasks, from managing conversations to assisting in deep research and code generation. However, creating AI systems that can generate truly novel insights presents a greater challenge than building agentic systems. Skepticism within the scientific community highlights the difficulty. In an earlier essay, Hugging Face’s Chief Science Officer, Thomas Wolf, argued that current AI systems lack the capability to formulate great questions, a crucial step in scientific breakthroughs. Kenneth Stanley, a former research lead at OpenAI, echoed similar sentiments in an interview with TechCrunch, stating that today’s AI models do not have the capacity to generate novel hypotheses. Stanley is now leading a team at Lila Sciences, a startup that secured $200 million to establish an AI-powered laboratory focused on enhancing the creativity and innovation of AI models. According to Stanley, the challenge lies in teaching AI to recognize what is creative and interesting, a task that is far from trivial. While it remains to be seen whether OpenAI can achieve the goal of producing AI systems capable of generating novel insights, Altman’s essay serves as a compelling preview of the direction OpenAI intends to take. If these systems become a reality, they could fundamentally alter how scientific research is conducted, paving the way for rapid and groundbreaking discoveries.