Avalanche Dynamics

In the AI Story

The physics of avalanche propagation depends critically on the pile's state. In a subcritical pile, where the slope is below the critical angle, most grains land and settle without triggering any cascade. When a small avalanche does occur — a few grains shifting — it's absorbed locally by the friction and interlocking of the surrounding grains. The system is in the 'frozen' regime that Christopher Langton identified: stable, predictable, incapable of producing interesting complexity. At the critical angle, the picture changes completely. The grains are arranged such that each is on the verge of shifting. The local friction exactly balances the local stress. A single grain's shift removes the support from its neighbors, which shift and remove support from their neighbors, and the cascade propagates. How far it propagates depends on the specific configuration of grains along every potential path — a configuration that's unique to each moment and unknowable in advance.

Edo Segal's Trivandrum experiment produced a small avalanche: twenty engineers, a local reorganization, effects contained to one team and one company. The SaaS Death Cross was a large avalanche: a trillion dollars redistributed, thousands of companies affected, investor confidence reshaped across an industry. Both follow the same power-law distribution. Both are manifestations of the same critical system. The magnitude difference is not a difference in kind — it's a difference in how far the cascade propagated before encountering configurations that absorbed it. The Trivandrum cascade encountered absorbing boundaries quickly (the team was small, the company wasn't publicly traded, the effects didn't propagate into financial markets). The Death Cross cascade propagated through densely connected investor networks, automated trading algorithms, and correlated repricing mechanisms across the entire software sector.

The aftershock pattern following large avalanches is itself a consequence of criticality. A major earthquake doesn't leave the fault in a relaxed state — it leaves it in a reorganized critical state with new stress concentrations. Secondary ruptures follow, tertiary ruptures follow those, in a diminishing sequence that follows its own power law. Similarly, the December 2025 capability threshold triggered cascading aftershocks: the market repricing in January–February 2026, the hiring freezes and restructurings in March, the educational panic in April. Each was a genuine avalanche, not merely an echo. Each was triggered by the reorganized configuration the previous avalanche produced. The system didn't absorb the disruption and return to stability. It absorbed the disruption and returned to criticality — ready for the next cascade.

Understanding avalanche dynamics reveals why individual resilience is necessary but insufficient. A well-positioned grain — deeply connected to its neighbors, part of a dense local network, embedded in structures that provide friction — can survive an avalanche that displaces less-connected neighbors. This is the physics behind Segal's prescription to build connections, judgment, and institutional trust. But no individual grain's connections can protect it from a sufficiently large avalanche. When the correlation length spans the system, even the most robustly connected grains are subject to cascades propagating from distant parts of the pile. Individual resilience must be complemented by structural interventions — dams, dissipative structures, institutional mechanisms — that channel avalanches at the system level rather than merely protecting individual positions.

Origin

The study of avalanches in granular materials predates Bak by decades, rooted in engineering needs for silo design and mining safety. What Bak contributed was the recognition that avalanche size distributions in critical systems follow universal mathematical laws independent of material properties. Whether the avalanche is made of sand grains, tectonic plates, or professional disruptions, the same power-law exponent governs the relationship between frequency and magnitude. This universality allowed Bak to predict that any system exhibiting power-law event distributions had self-organized to criticality — a prediction that could be tested across domains and has been progressively validated.

Key Ideas

Trigger versus cause distinction. The grain that triggers an avalanche is not meaningfully its cause — the pile's global critical state determines magnitude, making the specific trigger almost irrelevant.

Power-law size distribution. Avalanches at criticality follow f(s) ~ s^(-α), where s is size and α the exponent — many small, rare large, no characteristic scale separating normal from catastrophic.

Aftershock sequences. Large avalanches produce reorganized critical configurations that trigger secondary cascades, following their own power-law distribution of diminishing average magnitude.

Correlation-length dependence. How far a cascade propagates depends on the system's correlation length — at criticality this diverges, making system-wide avalanches possible from arbitrarily small triggers.

Unpredictability in principle. The specific path, timing, and magnitude of individual avalanches are fundamentally unknowable, not due to insufficient data but due to the chaotic sensitivity of critical systems to initial conditions.