Sam Altman Highlights Human Energy Use in AI Efficiency Debate

In a quiet but potent intervention during a recent industry panel, OpenAI CEO Sam Altman reminded the tech world that while AI systems consume vast amounts of energy, humans—our biological predecessors and operators—are no less resource-intensive. "It also takes a lot of energy to train a human," Altman remarked, drawing laughter and then thoughtful silence from the audience. The offhand observation, though seemingly anecdotal, has ignited a broader discourse on the comparative sustainability of artificial and biological intelligence.

Altman’s comment comes at a time when global regulators and environmental advocates are intensifying scrutiny over the carbon footprint of large language models. Training a single AI model like GPT-4 has been estimated to consume energy equivalent to hundreds of homes over a year, according to independent analyses by MIT and Stanford researchers. Yet Altman’s remark redirects attention to the often-overlooked energy costs embedded in human development: the food, transportation, education, and healthcare required to raise and train a single individual to contribute meaningfully to technological innovation.

While sources such as Zhihu discussions on AI model advancements and Altman’s leadership at OpenAI do not directly cite this quote, they provide critical context. One Zhihu thread examining Altman’s controversial departure from OpenAI’s board in November 2023 notes that his leadership style has consistently emphasized long-term, systems-level thinking—even in seemingly tangential remarks. Another thread, analyzing Meta’s SegmentAnythingModels (SAM), reveals a broader industry trend: researchers are increasingly comparing the efficiency of machine perception with human cognition, implicitly acknowledging the biological complexity underlying human learning.

Altman’s statement, therefore, is not a defense of AI’s energy use but a provocative recalibration. He is not suggesting humans are more efficient than machines—he is asking us to stop measuring AI’s cost in isolation. The human brain, while consuming only about 20 watts at rest, requires decades of energy-rich development to reach the cognitive capacity that now trains and fine-tunes AI systems. The average human, from infancy to professional expertise, consumes an estimated 2,000 to 3,000 kilocalories daily, translating to decades of cumulative energy expenditure far exceeding the operational cost of most AI models.

Environmental scientists argue that the real issue isn’t whether AI or humans use more energy, but whether we are optimizing the entire pipeline. Training a neural network may take weeks; training a human takes 18–25 years. Yet humans adapt, generalize, and innovate with minimal explicit instruction—qualities AI still lacks. This asymmetry suggests that while AI may be energy-intensive per training cycle, it may offer long-term efficiency gains if deployed correctly.

Altman’s comment may be the most subtle form of advocacy yet: a plea for holistic metrics. If we are to regulate AI responsibly, we must also account for the energy infrastructure that sustains its creators. The next generation of AI policy, he implies, must consider not just servers and GPUs, but schools, hospitals, and supermarkets. In an era obsessed with carbon footprints, Altman reminds us that the most powerful AI system ever built—the human mind—was trained by nature, nourished by agriculture, and sustained by civilization itself.

As the industry moves toward greener AI, Altman’s observation could become a foundational principle: efficiency isn’t just about reducing power draw—it’s about understanding the full ecosystem of intelligence, biological and artificial, that powers our future.