Models by StanP
StanP
Models by StanP
/*
My own version of AI 2027 forecast, leaning on Titotal's critique
*/
// Current time horizon, hours (length of task that can be completed by AI with 80% accuracy)
horizon_2025 = (3 to 12) / 60
// Current time horizon doubling time (years)
t_0 = {
horizon_2020 = (0.5 / 60 to 2 / 60) / 60/* Modelling the value of extending the life of those who have survived a suicide attempt vs. the general population */ // Neutral point on a 0-10 scale np = 1.7 // Global distribution of subjective wellbeing on a 0-10 scale // Mean is 5.7, the median of country-level SWB estimates in 2022 [https://docs.google.com/spreadsheets/d/1wLpXCeSNY6JLeYnQqU9sH0x1ZURWjuwsMBFs49ofxDI/edit?usp=sharing] // SD is 2.5, as found by World Happiness Report in 2018 [https://worldhappiness.report/ed/2019/changing-world-happiness/]
/* Modeling the global cooling rate expected from a large-scale nuclear exchange */ // Using the competing Reisner and Toon claims on the soot injection of a regional nuclear exchange as a basis, this model extrapolates to estimate the level of cooling expected from a nuclear war of at least 100 detonations. // Black carbon (known as BC or soot) in stratosphere from 100x15kT-warhead nuclear exchange (Tg), all strikes countervalue. Use the Toon estimate of 5Tg as the 95th percentile. Reisner estimates 0.2Tg, but claims this is an overestimate since they assume all combustible material is converted to BC, while the true amount would be 10-100 times less. Denkenberger & Pearce (2018) models the soot-emission factor as lognormally-distributed with 90% CI (1%,4%), which has mean value 2.2%. Hence our lower estimate, which we use as the 5th percentile, is 0.2Tg * 0.022 = 0.0044Tg. Use a lognormal distribution with these 5th & 95th percentiles, and replace the top and bottom 5% with constants 0.0044 and 5, to avoid unreasonably high outlier values. bc_min_all_countervalue = mx(truncate(0.0044 to 5,0.0044,5),0.0044,5,[0.9,0.05,0.05])
/* Estimate warming above pre-industrial levels by 2100 */ // Create a mixture of the warming levels predicted in the SSPs ssp1a = 1 to 1.8 ssp1b = 1.3 to 2.4 ssp2 = 2.1 to 3.5 ssp3 = 2.8 to 4.6
/* Identical to my nuclear winter model [link below], excepting different, pessimistic assumptions about the number of detonations in a large-scale nuclear conflict. */ // https://squigglehub.org/models/StanP/Modeling-Nuclear-Winter-with-Uncertainty // Using the competing Reisner and Toon claims on the soot injection of a regional nuclear exchange as a basis, this model extrapolates to estimate the level of cooling expected from a nuclear war of at least 100 detonations. // Black carbon (known as BC or soot) in stratosphere from 100x15kT-warhead nuclear exchange (Tg), all strikes countervalue. Use the Toon estimate of 5Tg as the 95th percentile. Reisner estimates 0.2Tg, but claims this is an overestimate since they assume all combustible material is converted to BC, while the true amount would be 10-100 times less. Denkenberger & Pearce (2018) models the soot-emission factor as lognormally-distributed with 90% CI (1%,4%), which has mean value 2.2%. Hence our lower estimate, which we use as the 5th percentile, is 0.2Tg * 0.022 = 0.0044Tg
/* Very quick BOTEC for cost-effectiveness of helping talented Ugandan study in Germany */ // Based on some numbers from here: https://forum.effectivealtruism.org/posts/TMjRuTLjQa6z6rdeY/large-scale-international-educational-migration-a-shallow // And some income stats from here: https://malengo.org/evidence-on-international-educational-migration // Effect size, in income doublings
/* Modeling the annual probability of cooling events from nuclear winter */ // See this sister model for a fuller explanation for how we determine inputs for soot, cooling etc. : https://squigglehub.org/models/StanP/Modeling-Nuclear-Winter-with-Uncertainty // NUCLEAR WINTER // The annual probability of a nuclear conflict with 100+ detonations. Our CEA estimates this at 0.12%. The three main sub-estimates were 0.06%, 0.10% and 0.13, so I use a Beta distribution with mean 0.12% and with a 90% confidence interval of approximately (0.01%, 0.34%) that includes these sub-estimates annual_prob_100plus_detonations = beta(1.224,998.776)
/* Modeling the probability and severity of nuclear winter under two approaches, and taking the mean of both */ // REISNER bc_min_all_countervalue_Reisner = (0.05 to 0.5) * (0.01 to 0.04) bc_min_all_countervalue = 0.0044 to 5
/* An attempt to crudely model the effects of global climate feedback cycles on uncertainty in global warming projections */ // Based on data from The Breakthrought Institute: https://thebreakthrough.org/issues/energy/flattening-the-curve-of-future-emissions baseline_eoc_warming = 2.2 to 6 current_policy_eoc_warming = 2.2 to 4.1
/* Describe your code here */ a = normal(2, 5)
/* Retired from my nuclear winter model */ // Give mortality rate odds ratios under different adaptation scenarios according to ALLFED's paper on adaptations to a 150Tg scenato none = 0.81/(1-0.81) simple = 0.77/(1-0.77) simple_trade = 0.70/(1-0.7) simple_culling = 0.66/(1-0.66) simple_culling_storage = 0.37/(1-0.37)
/*
Learning new squigggle features through https://forum.effectivealtruism.org/posts/BDXnNdBm6jwj6o5nc/five-slightly-more-hardcore-squiggle-models
*/
solar1 = List.reduce(
[
0.981,
0.981 ^ 2,
0.981 ^ 3,
0.981 ^ 4,/* Expected reduction in x-risk from developing super-PPE */ ppesuccess = mx(0,1,[0.5,0.5]) relbiorriskreduction = beta(1.0871,39.622) existentialbiorisk = beta(0.18551,12.099)
/* Inspired by https://www.metaculus.com/questions/18174/solar-power-dominates-renewables-by-2031/ Growth rates found by regression on Sheets */ windg = 0.05 to 0.0665 // annual growth in wind power hydrod = 45 to 80 // annual growth in hydr power (units of energy per year) solarg = 0.08 to 0.12755 // annual growth in solar power wind = 10^(3.3278+9*windg) // exponential model
/*
For the Metaculus question https://www.metaculus.com/questions/5755/china-to-usa-gdp-ratio-in-2050/
*/
uspop = .384 to .496 // projected population in 2050
chinapop = 1.2 to 1.4
globalgrowthfactor = 0.7 to 1.3
usgrowth = normal({p5:0.0194,p95:0.0394})*globalgrowthfactor // average annual GDP per capita growth, 2022-2050
chinagrowth = normal({p5:0.025,p95:0.065})*globalgrowthfactor