How Trees Teach Us About AI Safety
A LessWrong article argues that trees, which are mostly air, offer a counterintuitive lesson for AI safety: true safety comes from structural resilience, not merely from efficiency or sustainability. The draft oversimplifies this by focusing on resource efficiency rather than the article's core point about failure modes and robustness.
A recent LessWrong article draws an analogy between the composition of trees and AI safety. Contrary to the typical response that trees are 99% air, the author highlights that trees are mostly made of carbon dioxide from the air; the solid mass comes from CO₂, not from soil. This insight leads to a deeper lesson about AI safety: just as a tree's strength comes from a surprising source (air), AI dangers may also arise from seemingly benign or ubiquitous inputs. The article argues that focusing solely on making AI systems resource-efficient or sustainable misses the bigger picture of structural and emergent risks. Instead of efficiency, AI safety should focus on understanding how simple, common elements can combine into powerful and potentially dangerous systems. The core takeaway is that we must anticipate unexpected failure modes and design for robustness, not just optimization.