Our mission is to ensure that AGI (Artificial General Intelligence) benefits all of humanity.
Systems that point toward AGI are beginning to emerge, making it important to understand the moment we are in. AGI is a weakly defined term, but generally, we refer to it as a system capable of tackling increasingly complex problems at a human level across various fields.
Human beings are tool-builders, driven by an inherent desire to understand and create. This drive leads to progress, making the world better for all of us. Each generation builds upon the discoveries of previous ones to create increasingly capable tools—electricity, the transistor, the computer, the internet, and soon, AGI.
Over time, in fits and starts, the steady march of human innovation has brought previously unimaginable levels of prosperity and improvement to nearly every aspect of human life.
In one sense, AGI is just another tool in this ever-growing scaffolding of human progress. In another sense, it marks the beginning of something entirely new. The economic growth ahead seems astonishing, and we can now imagine a world where we cure all diseases, have more time with our families, and fully realize our creative potential.
In a decade, perhaps everyone on Earth will be capable of achieving more than even the most impactful people today.
We continue to see rapid progress in AI development. Here are three observations about the economics of AI:
- AI Scaling: The intelligence of an AI model roughly corresponds to the logarithm of the resources used to train and run it. These resources are primarily training compute, data, and inference compute. It appears that spending arbitrary amounts of money results in continuous, predictable gains, with scaling laws accurately predicting this over many orders of magnitude.
- Decreasing Costs: The cost of using a given level of AI decreases about 10x every 12 months, leading to much more widespread use. This is evident from the token cost of GPT-4 from early 2023 to GPT-4o in mid-2024, where the price per token dropped by approximately 150x. This rate of reduction is much stronger than Moore’s Law, which historically saw a 2x decrease every 18 months.
- Super-Exponential Value: The socioeconomic value of linearly increasing intelligence is super-exponential in nature. As a result, we see no reason for exponentially increasing investment to stop in the near future.
If these observations hold true, the societal impact will be significant.
AI agents are now being rolled out, and they will eventually feel like virtual co-workers.
Consider a software engineering agent, one we expect to be particularly important. Imagine it can eventually do most tasks a software engineer with a few years of experience could do, for tasks lasting up to a few days. It won’t generate the most groundbreaking ideas and will require significant human supervision, but it will excel at certain tasks and be surprisingly ineffective at others.
Still, imagine this agent as a relatively junior, virtual coworker. Now imagine 1,000 or even 1 million such agents. Envision these agents across every field of knowledge work.
In some ways, AI may turn out to be like the transistor—an economic discovery that scales well and permeates nearly every corner of the economy. We don’t think much about transistors or transistor companies, yet their impact is widely distributed. Similarly, we expect our computers, TVs, cars, toys, and more to perform near-miracles thanks to AI.
The world won’t change all at once—change rarely happens in that way. Life in 2025 will be much like it was in 2024. People will still fall in love, create families, argue online, hike in nature, etc.
But the future is coming in a way that will be impossible to ignore, and long-term changes to our society and economy will be profound. We will find new ways to be useful to each other, new ways to compete, but they may not resemble today’s jobs.
Agency, willpower, and determination will likely be incredibly valuable. The ability to decide what to do and how to navigate an ever-changing world will have enormous value; resilience and adaptability will be essential skills to cultivate. AGI will amplify human willpower, enabling individuals to have greater impact than ever before.
The impact of AGI will be uneven. While some industries will remain largely unchanged, scientific progress will likely accelerate beyond what we can imagine today, potentially surpassing every other area of change.
The price of many goods will eventually fall dramatically. Currently, the cost of intelligence and energy constrains much of the economy, but these prices will likely drop, while the price of luxury goods and limited resources like land may rise.
Technically, the road ahead is relatively clear. However, public policy and collective opinions on how to integrate AGI into society are crucial. One reason we are launching products early and often is to give both society and technology time to co-evolve.
AI will seep into all areas of the economy and society, and we will expect everything to be “smart.” Many of us believe people will need more control over technology than we have historically provided, including open-sourcing more, and balancing safety and empowerment will require careful trade-offs.
While we must avoid recklessness, and some decisions around AGI safety will likely be unpopular, we believe that, directionally, empowering individuals as we approach AGI is key. The alternative could be a scenario where AI is used by authoritarian governments to control their populations through mass surveillance and loss of autonomy.
Ensuring that the benefits of AGI are widely distributed is critical. Historically, technological progress has improved metrics like health and economic prosperity over the long term. However, increased equality is not guaranteed by technology alone, and achieving it may require innovative solutions.
Specifically, the balance of power between capital and labor could be disrupted, requiring early intervention. We are open to unconventional ideas, such as giving a “compute budget” to enable everyone on Earth to use a lot of AI. Alternatively, simply driving down the cost of intelligence may achieve similar results.
In 2035, everyone should have access to the intellectual capacity that only a few possess today. Unlimited access to genius should be available to anyone who can imagine it. Right now, much talent is constrained by a lack of resources, and unleashing that talent will lead to tremendous creative output and benefit for all.
Thanks especially to Josh Achiam, Boaz Barak, and Aleksander Madry for reviewing drafts of this.
By using the term AGI here, we aim to communicate clearly and do not intend to alter or interpret the definitions and processes that define our relationship with Microsoft. We fully expect to be partnered with Microsoft for the long term. This footnote may seem unnecessary, but we pre-emptively address any potential misinterpretations that might arise in the media.
