Module 3 / Lesson 6.4
Tools to Estimate Existential Risks
How do we predict the probability of an event that has never happened? This lesson explores the scientific tools used to model rare risks, including prospective modeling, correcting for survivorship bias, and aggregation systems.
6.4.1 Estimating Risks with Historical Data vs. Models
- Learning: Watch the Professor's Brief and read thoroughly key concepts mentioned in the text.
- Reflection: Risk estimation requires choosing between Historical Data, which relies on the frequency of past events like crater records, and Prospective Modeling, which uses physical laws and current observations to predict future likelihoods.
Watch this video: Professor's Brief: Historical Data vs. Modeling
- Key Concept: There are two distinct ways to estimate probability.
| Approach | 1. Historical Data | 2. Modeling (Prospective) |
|---|---|---|
| Description | Looking at past frequency. "How often has this happened before?" | Analyzing the mechanism. "Based on the physical properties, what is likely to happen?" |
| Coin Flip Example | You flip the coin 1,000,000 times; you get heads 499,612 times. You conclude the chance is 50.0%. | You examine the coin. It is symmetrical with two sides. You reason the chance is 50%. |
| Asteroid Example | Average Century Risk: We look at craters on Earth to see how often we have been hit historically. | Next Century Risk: We scan the sky with telescopes and use Newton's laws to predict if any known objects are on a collision course. |
Asteroid Risk Table (from The Precipice, p. 71)
| Size (Diameter) | Estimated Total | Found (Known) | Average Century Risk | Next Century Risk |
|---|---|---|---|---|
| 1km – 10km | ~900 | ~95% | 1 in 6,000 | 1 in 120,000 |
| 10km+ (Mass Extinction) | ~4 | >99% | 1 in 1,500,000 | <1 in 150 million |
Source: Toby Ord, The Precipice. Note the massive reduction in "Next Century Risk" for large objects we have already found and tracked.
6.4.2 Survivorship Bias
- Learning:
- Watch this video to learn how survivorship bias works.
- Read The Precipice, page 84 (Box on Survivorship Bias).
- Reflection: If a risk had wiped us out in the past, we wouldn't be here to observe it. Therefore, our historical data is biased towards "survival", showing fewer extinction events in past years (zero, in fact) than we should have expected and should expect going forward. This is known as an observation selection effect.
Your progress stays in this browser.