OpenAI CEO Sam Altman has outlined his vision for the future of artificial general intelligence (AGI) and its potential impact on society in a detailed blog post. The post, published on Altman’s personal blog, presents three key observations about AI economics and their implications for technological development.
According to Altman, AI model costs are expected to decrease significantly, with a projected tenfold reduction every 12 months. He cites the example of GPT-4, where the price per token dropped approximately 150 times between early 2023 and mid-2024. This rate of cost reduction substantially outpaces Moore’s Law, which traditionally predicted computing power doubling every 18 months.
The OpenAI chief executive explains that artificial intelligence scales logarithmically with training resources, including compute power and data. He notes that continuous and predictable gains appear possible with increased investment, following established scaling laws across multiple orders of magnitude.
Altman emphasizes the super-exponential nature of AI’s socioeconomic value as intelligence increases linearly. This relationship, he argues, will likely drive continued exponential investment in AI development.
The blog post describes a future where AI agents could function as virtual coworkers, particularly in knowledge work sectors like software engineering. While these AI assistants might not match senior human expertise, they could perform tasks equivalent to those handled by professionals with several years of experience.
Regarding societal impact, Altman compares AI’s potential influence to that of the transistor, suggesting it will gradually integrate into various aspects of the economy. He stresses the importance of ensuring broad distribution of AGI benefits and addresses potential concerns about the balance between capital and labor.
The OpenAI CEO acknowledges potential challenges, including the need for policy frameworks and public discourse on AGI integration. He advocates for increased user control over AI technology and suggests the possibility of providing universal “compute budgets” to ensure widespread access to AI resources.
Looking ahead to 2035, Altman envisions a world where individuals can access AI capabilities equivalent to the collective intellectual capacity of everyone in 2025. He emphasizes the importance of making these resources available to unleash global creative potential.
Photo by Steve Jennings/Getty Images for TechCrunch. License: CC BY 2.0