Sam Altman’s “Intelligence Age”: Superintelligence, Scaling and Work

Sam Altman’s 2024 essay “The Intelligence Age” presents an optimistic view of how AI might transform society. I want to read key passages and explain why I find them compelling. Altman argues that the ability of deep learning to improve with scale is central to the story.

Altman describes society as a kind of accumulated intelligence: each generation benefits from the infrastructure and discoveries built by others. He imagines AI adding to that scaffolding and eventually giving each person a team of virtual experts. I recently used ChatGPT to understand a health concern and prepare myself, while also working with human doctors.

One application that excites me is education. I have two young children, and I hope AI tutors can adapt to an individual child’s pace and preferred way of learning. I learn well through video; another student may benefit from a different format. These are aspirations, not guarantees of reliable tutoring.

Altman envisions shared prosperity greater than anything available today. I am optimistic that AI could help people, although prosperity alone does not ensure happiness. Not having enough money can itself create serious problems.

Altman recounts the chain of discoveries behind computer chips and modern AI. He writes that superintelligence could arrive in a few thousand days, while explicitly allowing that it may take longer. This is a prediction made in 2024, not a confirmed timetable.

His explanation is that deep learning worked and improved as more resources were devoted to it. I see two important forms of scaling: training with more data and compute, and spending more compute at inference using reasoning techniques. The remaining technical questions and future returns to scaling are uncertain.

Altman imagines autonomous personal assistants that coordinate tasks such as medical care. I would like agents to handle work such as reading and summarizing email, with appropriate oversight. He also anticipates AI helping develop future AI systems and advance science.

I imagine two routes to that feedback loop: models generating or labeling data for later models, and AI systems helping researchers find new techniques. I see this as a possible path to rapid improvement, but it is a scenario rather than a demonstrated inevitability.

Altman argues that making compute abundant and affordable requires energy, chips and infrastructure; otherwise AI could become concentrated among the wealthy. He also says the transition will bring difficult risks. Access and how benefits are distributed are central questions.

Labor disruption is one example. Altman expects significant changes but thinks most jobs will change more slowly than some people anticipate. I hope new work emerges as old work changes, and I am enthusiastic about AI amplifying human creativity; those outcomes are not guaranteed.

Altman compares today’s occupations with jobs that have disappeared over time. My own speculation is that human coding as we know it might eventually disappear, while new creative work appears. The timing and outcome remain open questions.