Keyboard shortcuts

Press or to navigate between chapters

Press S or / to search in the book

Press ? to show this help

Press Esc to hide this help

Chapter 15: Probability Distributions

The Shape of Uncertainty

Probability distributions reveal not just what’s likely, but the full range of possibilities and their relative chances. This chapter visualizes and interprets the probability landscape emerging from our 1.3 billion simulations.

Understanding the Distributions

What Probability Distributions Tell Us

Beyond simple percentages, distributions reveal:

  • Central tendency: Most likely outcomes
  • Spread: Range of uncertainty
  • Skewness: Asymmetric risks
  • Tails: Extreme possibilities
  • Modality: Single vs multiple peaks

Each distribution tells a story about the future’s shape.

Hypothesis Probability Distributions

H1: AI Progress Trajectory

Distribution Characteristics:

  • Type: Strongly right-skewed
  • Mean: 91.1%
  • Median: 93.2%
  • Standard Deviation: 5.7%
  • 95% CI: [75.1%, 98.1%]

Interpretation: The distribution shows overwhelming probability mass toward continued progress. The long left tail represents low-probability barriers, but the peak is sharp and decisive. This is as close to certainty as forecasting allows.

H2: AGI Achievement

Distribution Characteristics:

  • Type: Nearly uniform
  • Mean: 44.3%
  • Median: 44.1%
  • Standard Deviation: 16.9%
  • 95% CI: [14.6%, 80.1%]

Interpretation: This remarkably flat distribution represents true uncertainty. Evidence is so balanced that probability mass spreads almost evenly across the range. We genuinely don’t know.

H3: Employment Impact

Distribution Characteristics:

  • Type: Left-skewed (toward displacement)
  • Mean: 74.9% displacement
  • Median: 77.3%
  • Standard Deviation: 9.9%
  • 95% CI: [54.9%, 90.8%]

Interpretation: The distribution leans heavily toward job displacement with a sharp peak around 75-80%. The tail toward complementarity exists but carries little probability mass.

H4: Safety Outcomes

Distribution Characteristics:

  • Type: Bimodal
  • Primary Peak: 62% (safe)
  • Secondary Peak: 35% (risky)
  • Standard Deviation: 13.3%
  • 95% CI: [36.6%, 86.3%]

Interpretation: Unusual bimodal structure suggests two distinct equilibria. Either we achieve safety through coordination or fail dramatically—middle ground is unstable.

H5: Development Model

Distribution Characteristics:

  • Type: Strongly left-skewed (toward centralization)
  • Mean: 77.9% centralized
  • Median: 81.2%
  • Standard Deviation: 12.7%
  • 95% CI: [48.9%, 95.8%]

Interpretation: Economic forces create strong pull toward centralization. The distribution’s shape suggests this is nearly inevitable without intervention.

H6: Governance Evolution

Distribution Characteristics:

  • Type: Left-skewed (toward authoritarianism)
  • Mean: 63.9% authoritarian
  • Median: 66.4%
  • Standard Deviation: 13.3%
  • 95% CI: [39.0%, 87.8%]

Interpretation: Democracy faces headwinds. The distribution’s leftward lean shows authoritarian drift as the path of least resistance.

Scenario Probability Distributions

Top Scenarios Show Sharp Peaks

ABBABB (Rank 1: 11.59%):

  • Sharp peak, narrow distribution
  • Standard deviation: 1.2%
  • Robust across models
  • High confidence prediction

AABABB (Rank 2: 9.21%):

  • Clear peak, moderate spread
  • Standard deviation: 1.8%
  • Some model sensitivity
  • Good confidence

Mid-Tier Scenarios Show Broader Distributions

Scenarios ranking 20-40 show:

  • Wider distributions
  • Multiple smaller peaks
  • Higher uncertainty
  • Model dependence

Low Probability Scenarios Are Flat

Bottom tier scenarios show:

  • Nearly uniform distributions
  • No clear peaks
  • High uncertainty
  • Extreme model sensitivity

Temporal Evolution of Distributions

Uncertainty Funnel

Plotting distributions over time reveals a funnel pattern:

2025: Wide Distributions

  • High uncertainty
  • Multiple peaks
  • Overlapping scenarios
  • Intervention possible

2030: Separating Peaks

  • Distributions sharpen
  • Peaks diverge
  • Overlap decreases
  • Paths crystallizing

2035: Distinct Distributions

  • Sharp peaks
  • Clear separation
  • Little overlap
  • Paths locked

2040-2050: Narrow Peaks

  • Very sharp distributions
  • No overlap
  • Fixed outcomes
  • Change difficult

Joint Probability Distributions

Correlation Structures

Examining joint distributions reveals dependencies:

H1-H2 Joint Distribution:

  • Strong positive correlation
  • Progress drives AGI probability
  • Conditional probability shifts

H5-H6 Joint Distribution:

  • Very strong correlation
  • Centralization enables authoritarianism
  • Near-linear relationship

H3-H6 Joint Distribution:

  • Moderate negative correlation
  • Displacement threatens democracy
  • Threshold effects visible

Distribution Clustering

Three Meta-Distributions Emerge

When we overlay all 64 scenario distributions, three distinct clusters appear:

Cluster 1: Adaptive Integration

  • Center: 42% probability
  • Spread: ±8%
  • Shape: Normal-like
  • Stability: High

Cluster 2: Fragmented Disruption

  • Center: 31% probability
  • Spread: ±11%
  • Shape: Left-skewed
  • Stability: Medium

Cluster 3: Constrained Evolution

  • Center: 27% probability
  • Spread: ±10%
  • Shape: Right-skewed
  • Stability: Medium

Extreme Value Analysis

Tail Risks

Right Tail (Best Cases):

  • Probability of utopian outcomes: <0.1%
  • Maximum achievable success: ~85% positive across all dimensions
  • Requires perfect coordination

Left Tail (Worst Cases):

  • Probability of dystopian collapse: 2.3%
  • Maximum failure: ~95% negative across all dimensions
  • Results from cascade failures

Black Swan Potential

Fat tails in several distributions suggest:

  • Extreme events more likely than normal distribution predicts
  • Cascade effects can amplify
  • Prepare for 1-in-100 events

Sensitivity of Distributions

To Evidence Quality

Higher quality evidence → Sharper distributions Lower quality evidence → Flatter distributions

To Causal Model

Conservative model → More uniform distributions Aggressive model → More extreme peaks

To Time Horizon

Near term → Broader distributions Long term → Sharper peaks (path dependency)

Confidence Calibration

How Sure Are We?

Comparing predicted distributions to confidence levels:

Overconfident on:

  • None identified

Well-calibrated on:

  • AI progress
  • Employment impact
  • Centralization

Underconfident on:

  • AGI timing
  • Governance outcomes

Key Distribution Insights

1. Certainty Is Rare

Only H1 (AI progress) shows overwhelming directional certainty. Everything else has meaningful uncertainty.

2. Bimodality Is Common

Several distributions show multiple peaks, suggesting discrete equilibria rather than continuous outcomes.

3. Skewness Reveals Bias

Most distributions lean toward negative outcomes, suggesting positive futures require active effort.

4. Tails Matter

Fat tails mean extreme scenarios, while unlikely individually, collectively represent significant probability.

5. Time Sharpens

Distributions narrow over time as path dependencies lock in, making early intervention critical.

Using Distributions for Decision-Making

For Risk Management

  • Focus on full distribution, not just means
  • Prepare for tail events
  • Understand skewness direction

For Strategic Planning

  • Plan for distribution peaks
  • Prepare for full range
  • Monitor distribution shifts

For Investment

  • Diversify across distribution
  • Hedge tail risks
  • Position for multiple peaks

The Distributions’ Message

The probability distributions tell us:

  1. Few things are certain (except AI progress)
  2. Multiple equilibria exist (bimodal distributions)
  3. Negative outcomes are “downhill” (skewness)
  4. Time reduces options (narrowing distributions)
  5. Extremes are possible (fat tails)

Understanding these distributions means understanding not just what’s likely, but the full landscape of possibility—and that’s essential for navigating uncertainty.


Next: Temporal Evolution →
Previous: Monte Carlo Results ←