Sam Altman thinks AI will have ‘novel insights’ next year

In a thought-provoking essay released on Tuesday titled "The Gentle Singularity," OpenAI CEO Sam Altman articulated his ambitious vision for the transformative role of artificial intelligence over the next decade and a half. Altman’s writing exemplifies his characteristic futurism, emphasizing the potential of Artificial General Intelligence (AGI) while also tempering expectations regarding its imminent arrival. In his essay, Altman outlined a future where AGI significantly alters our understanding of work, energy, and societal structures. Notably, he forecasted that by 2026, we might witness the emergence of AI systems capable of generating novel insights. This assertion, while somewhat ambiguous, aligns with OpenAI's recent endeavors to enhance AI models to produce innovative and valuable ideas. In April, OpenAI introduced its o3 and o4-mini AI reasoning models, with co-founder Greg Brockman highlighting their capability to generate new, helpful concepts. Altman’s latest musings suggest that OpenAI may intensify its focus on developing AI that can indeed create original insights within the upcoming year. This pursuit is not exclusive to OpenAI; many competitors are similarly dedicated to training AI models that assist scientists in forming new hypotheses and driving groundbreaking discoveries. For instance, Google unveiled a paper in May detailing AlphaEvolve, an AI coding agent that purportedly generated unique solutions to complex mathematical problems. Additionally, a startup called FutureHouse, backed by former Google CEO Eric Schmidt, claims its AI tool has already made a genuine scientific discovery. Anthropic also launched a program in May aimed at bolstering scientific research with AI support. Should these initiatives prove successful, they could revolutionize critical areas of scientific inquiry, including drug discovery and materials science. Altman's previous blogs have hinted at OpenAI's strategic direction; for example, he predicted that 2025 would mark the rise of AI agents, a claim that preceded the rollout of their initial three AI agents: Operator, Deep Research, and Codex. However, the challenge of developing AI systems that can produce truly novel insights may be more complex than simply creating agent-like capabilities. The scientific community remains cautious about AI's potential to generate original insights. Earlier this year, Thomas Wolf, Chief Science Officer at Hugging Face, expressed skepticism, arguing that current AI models struggle to ask profound questions, a vital element for scientific breakthroughs. Furthermore, Kenneth Stanley, a former OpenAI research lead, noted that today’s AI lacks the ability to formulate novel hypotheses. Stanley is currently assembling a team at Lila Sciences, a startup that recently secured $200 million to create an AI-driven laboratory focused on enhancing the hypothesis generation process. He acknowledges the intricacies involved in equipping AI with a sense of creativity and relevance. As we look ahead, whether OpenAI can indeed develop an AI model capable of delivering genuine novel insights remains an open question, but Altman's essay provides a tantalizing glimpse into the company’s future direction.