In a recent analysis titled \"More is different for intelligence,\" lessw-blog draws a compelling parallel between the history of software and the future of Large Language Models (LLMs).
\nThe post argues that the industry is currently underestimating the long-term impact of AI by focusing solely on efficiency gains in existing workflows, rather than anticipating the structural changes that occur when the cost of intelligence approaches zero.
The central thesis rests on the concept that a sufficient quantitative change produces a qualitative change. The author illustrates this through the evolution of "computation." Before the software era, computation was a manual task performed by humans, serving as a significant bottleneck for knowledge work. Software did not merely make these human computers faster; it reduced the cost of calculation so drastically that it enabled entirely new economic activities. Real-time supply chain management, continuous financial trading, and large-scale A/B testing were not just faster versions of old processes-they were previously impossible due to the prohibitive cost of manual calculation.
The post suggests that LLMs are poised to trigger a similar transformation regarding "intelligence" or semantic processing. Currently, the technology sector views LLMs primarily as productivity tools-mechanisms to draft code, write emails, or summarize documents more quickly. This phase is analogous to using early computers simply to speed up a room full of accountants. However, the author contends that as the cost of intelligence drops, the economy will move beyond optimization. We will likely see the emergence of complex production graphs and organizational structures that are currently latent-ideas that are technically feasible but economically viable only when intelligence is abundant and cheap.
This perspective is vital for understanding the trajectory of AI adoption. It challenges the notion that AI is merely a "copilot" for human workers and suggests it is a foundational layer for new types of organizations that could not exist under human-only constraints.
For a deeper understanding of how historical computing trends might predict the future of AI-driven organizations, we recommend reading the full article.
Read the full post on LessWrong
\n\n