Altman: Three observations on AI economics

On February 10, OpenAI CEO Sam Altman shared his three observations on the economics of artificial intelligence (Three Observations) on his personal blog. He believes that the scaling law of artificial intelligence will continue to be effective, and there is no reason to stop exponential investment in AI in the near future. Increasing equality is not determined by technology, and doing so may require new ideas, especially since the balance of power between capital and labor can be easily disrupted, which may require early intervention.

"Our mission is to ensure that AGI (general artificial intelligence) benefits all mankind." Altman said that humans are creators of tools and have an inherent motivation to understand and create. Electricity, transistors, computers, the Internet, and, soon, AGI, each new generation of technology builds on the discoveries of previous generations to create more powerful tools. In one sense, AGI is just another tool in the scaffolding we are all building together to advance humanity. In another sense, it’s a beginning, and it’s hard not to say “this time it’s different.”

Artificial intelligence continues to advance rapidly. Altman shared three observations on the economics of artificial intelligence.

First, the intelligence of a large AI model is roughly equal to the resources used to train and run it, which are mainly training compute, data, and inference compute. It seems that you can spend any amount of money and get consistent and predictable returns, and the scaling laws for making predictions are accurate over many orders of magnitude.

Second, the cost of using a particular level of AI is falling by about 10 times every 12 months, and lower prices will bring more applications. This can be seen in the token cost of GPT-4 in early 2023 to GPT-4o in mid-2024, with the price per token falling by about 150 times during this period, and Moore's Law changing the world by 2 times every 18 months is incredibly powerful.

Third, the socioeconomic value of linearly growing intelligence is inherently super-exponential. One consequence of this is that there is no reason to stop exponentially growing investments in the near future.

If these three observations continue to hold true, Altman said, the impact on society will be significant. Start rolling out AI agents now, and these agents will eventually feel like virtual colleagues. For example, software engineer agents will eventually be able to do most of the work that software engineers with several years of experience at top companies can do. But the agent will not have the greatest new ideas, and it will require a lot of human supervision and guidance.

"In some ways, AI could be like the transistor in economics, a major scientific discovery that scales well and permeates almost every corner of the economy."The world won't change all at once, Altman said.In the short term, life will remain largely the same, and people will spend 2025 the same way they did in 2024, but the future will be coming to us in a way that cannot be ignored, and the long-term changes to society and the economy will be huge.

We will find new things to do, new ways to be useful to each other, and new ways to compete,but they may not look much like today's jobs.

Agency, willpower, and determination are more valuable.Deciding correctly what to do and figuring out how to navigate a changing world will be of great value. Resilience and adaptability will be useful skills to cultivate.AGI will be the greatest lever ever exerted on the human will, giving individuals more influence than ever before.

Technologically, the path ahead of us looks fairly clear. The impact of AGI is expected to be uneven, and while some industries will see small changes, scientific progress is likely to be much faster than it is today. Public policy and collective opinion on how AGI will be integrated into society is very important. Many expect to give people more control over the technology than ever before, including more open source, and accept a balance between safety and individual empowerment. "While we don't want to act recklessly, and there may be some major decisions and restrictions related to AGI safety that will be unpopular, directionally we think it is important to lean more towards individual empowerment as we get closer to achieving AGI."

The historical impact of technological progress shows that most indicators we care about, such as health outcomes and economic prosperity, will get better on average and in the long run, but increasing equality does not seem to be determined by technology, and doing so may require new ideas. In particular, the balance of power between capital and labor can be easily disrupted, which may require early intervention.

"We are open to some strange-sounding ideas, such as providing some 'computing budget' so that everyone on the planet can fully use artificial intelligence. We can also see many ways to achieve the desired effect as long as the cost of intelligence is continuously reduced." Altman said that in 2035, anyone will be able to mobilize the intelligence equivalent to all human beings in 2025, and everyone should have unlimited talents to guide what they can imagine. Now a large number of talents do not have the resources to fully express themselves. If we change this situation, the world's creative output will bring huge benefits to everyone.