График, после которого вы посмотрите на мир иначе [Veritasium]

Scientific concepts, discoveries, and nature phenomena

1) Normal distributions vs. power laws

• Normal distribution: Most values cluster near an average; extremes are very rare (bell curve).
• Power laws: Heavy tails mean extreme events occur much more often than predicted by normal distributions.
• Consequence: In power-law systems, ideas like a characteristic scale (or stable variance) can break down; averages may behave unexpectedly.

2) Pareto distribution and income inequality (power-law tail)

• Vilfredo Pareto (late 19th century): proposed hidden patterns in income distributions.
• Method/fit:
  • Collect income tax data across multiple countries.
  • Plot the distribution of people earning above a threshold.
  • Use a logarithmic scale: the curve becomes approximately a straight line.

• Key result:

  • The survival function (number of people with income ≥ X) follows a power law: [ P(\text{income} \ge X)\propto X^{-\alpha} ]

  • Tail behavior implies some people earn 10x, 100x, etc. more than others—unlike a normal (bell-curve) model.

3) Toy “casino games” showing how distribution changes with multiplicative processes

• Game 1 (coin flips add winnings) → connects to normal distribution / expected value intuition.
• Game 2 (coin flips multiply wealth):
  • Multiplicative randomness → lognormal distribution (bell-shaped only on a log-scale).
• Game 3 (early stopping / doubling until first head):
  • Produces the St. Petersburg paradox:
    • Infinite theoretical expected value due to a tail that allows arbitrarily large payoffs with nonzero probability.
  • On log-log axes, payoff probability follows a power law (slope ≈ -1 in the described example).
• Key point: for power laws, standard deviation can diverge (no meaningful spread by usual measures).

4) Self-organized criticality and fractals in disasters

• Critical state: systems “hover” near a point where small changes can trigger cascades.
• Fractals + power laws:
  • Branching structures repeat across scales (self-similarity).
  • Their size distributions can follow power laws.

Example: Earthquake mechanics (magnitude vs. energy)

• Earthquake “frequency” decreases roughly exponentially with magnitude.
• Seismic energy increases exponentially with magnitude.
• Together, these trends yield a power law for probability vs. measurable outcomes.

Example: Forest fires (Yellowstone 1988)

• Fire size distribution is simulated on a grid model:
  • Cells can be tree present / absent.
  • Lightning probability triggers fires.
  • Feedback loop:
    • Small fires burn local areas → prevent total buildup.
    • If major fires are absent, forest density increases → later larger fires occur.
    • After a large fire, regrowth returns the system toward criticality.
• Result: simulations show power-law behavior and fractal-like patterns of burned/remaining regions.
• Historical intervention:
  • The “10 AM rule” (1935, U.S. Forest Service): suppress by next morning to prevent escalation, but in critical systems this can worsen outcomes.

5) Critical points in magnets (physics of phase transitions)

• Heating a magnet destroys magnetism by disrupting aligned domains.
• At the critical point (phase transition):
  • Domain behavior becomes scale-free: similar patterns appear for small and large domains.
  • Domain-size distribution follows a power law.
• Near criticality, the system lacks a characteristic scale and shows diverging influence (a local change can propagate widely).
• Predictability decreases; the “next event” becomes effectively unpredictable even though statistical laws (power laws) describe outcomes.

6) Universality (why different systems behave similarly near criticality)

• Universality: many systems at critical transitions share the same large-scale statistical behavior, regardless of microscopic details.
• Implication:
  • If one system class shows a certain scaling law, others in the same class can share it.
  • Even crude models can still capture universal exponents and distributions.

7) Sandpile model (avalanche statistics) as a minimal critical model

• Per Bak and colleagues proposed a sandpile model showing how avalanches of many sizes can arise spontaneously.
• Core dynamics:
  • Add one grain at a time to a cell.
  • If local height exceeds a threshold, the pile topples, distributing grains to neighbors.
  • Avalanche sizes follow power-law statistics.
• Connection to earthquakes:
  • Avalanche-size curves resemble observed earthquake energy/release distributions.
• Controversy:
  • Whether the model captures real sand behavior.
  • Bak’s response (paraphrased): the model explains the universal emergence of power laws.

8) Broader prevalence of power laws in society and nature

Subtitled examples said to follow power laws include:

• DNA sequence patterns
• Species distributions in ecosystems
• Mass extinctions
• Distribution of city populations
• Stock-price fluctuations / crash sizes
• Citation index of scientific articles
• Number of casualties in wars

9) Networks and the “preferential attachment” model (Barabási–Albert)

• Albert-László Barabási + Réka Albert:
  • Internet/web links follow a power law (few nodes have enormous degree).
• Hypothesis: new nodes link preferentially to already well-connected nodes (rich-get-richer).
• Simulation methodology:
  • Start with a small network.
  • Add new nodes over time.
  • Attach new nodes to existing ones with probability proportional to current degree.
• Result: degree distribution approaches a power law (slope ≈ -2 in the example).
• Interpretation: similar “power-law” outcomes can emerge for people/companies—more fame/riches attract more of both.

10) Practical framing: why insurance and risk models fail under power laws

• For power-law event sizes:
  • Rare large events dominate risk.
  • “Normal-ish” assumptions underestimate tail risk.
• Example: Paradise, California wildfire (2018)
  • An insurer wasn’t prepared for tail behavior and went bankrupt.

11) Final overarching claim

In systems governed by power laws and/or near criticality:

• Extremes are not “outliers” in the usual sense—they are structurally expected.
• “Success” often requires tolerating many small failures to have a chance at a huge payoff.

---

Researchers and sources featured (as named in the subtitles)

• Vilfredo Pareto
• Abraham de Moivre
• Per Bak (and “his colleagues” / team)
• Albert-László Barabási
• Réka Albert
• Kasper (named for preparing magnet visualization; first name only as given)
• Bill Gates and Elon Musk (used as illustrative examples in the narrative)
• U.S. Forest Service (source of the “10 AM rule,” 1935)
• Veritasium / Veretasium (channel/host branding)
• Y Combinator (mentioned as a source/statistic for startup profits)