14 - Comparative Risk Profiles: Natural vs. Anthropogenic.

14 - Comparative Risk Profiles: Natural vs. Anthropogenic.  Natural risks to human extinction, such as asteroid impacts, supervolcanic eruptions, and natural pandemics, have historically exhibited extremely low probabilities, estimated at approximately 1 in 10,000 over the next century. These risks stem from exogenous cosmic or geological events that humanity has endured without extinction for over 300,000 years of Homo sapiens existence, with no evidence of prior near-extinction from such causes despite exposure to recurrent threats like the Toba supervolcano eruption around 74,000 years ago, which reduced human populations but did not eliminate the species.   Empirical bounds on background extinction rates from natural hazards further constrain the annual probability to less than 1 in 100,000 for events like unmitigated asteroid strikes larger than 10 km in diameter, which occur roughly every 100 million years.  In contrast, anthropogenic risks—driven by human technologies and decisions, including nuclear war, engineered pathogens, and uncontrolled artificial intelligence—carry substantially higher estimated probabilities, collectively dominating expert assessments of existential threats at around 1 in 6 over the same century-long horizon. For instance, unaligned artificial intelligence is pegged at 1 in 10, engineered pandemics at 1 in 30, and nuclear conflict at 1 in 1,000, reflecting the rapid escalation of human capabilities since the mid-20th century that enable self-inflicted global catastrophes absent in natural baselines. Unlike natural risks, which are frequency-stable and independent of human agency, anthropogenic threats exhibit accelerating trends tied to technological proliferation; for example, the global stockpile of nuclear warheads peaked at over 70,000 in 1986 before partial reductions, yet retains extinction potential through escalation chains not paralleled in geological records.  Risk Category    -- Estimated Probability (Next Century -- Key Characteristics.  - Natural (e.g., asteroids, supervolcanoes, natural pandemics -- ~1 in 10,000 -- Exogenous, low frequency (e.g., <1 in 100,000/year for large impacts), minimal mitigation feasibility beyond deflection tech for asteroids; historical survival implies rarity.  - Anthropogenic (e.g., AI misalignment, biotech, nuclear -- ~1 in 6 total, with subsets like AI at 1/10 -- Endogenous, controllable via policy but amplified by dual-use tech; near-term concentration (decades) versus natural's geological timescales; surveys show experts assigning 10-100x higher odds to human-caused over natural.  This disparity arises from causal differences: natural events lack intent or scalability with human progress, whereas anthropogenic risks leverage exponential advancements in destructive power—evident in the shift from pre-industrial baselines to post-1945 nuclear and biotech eras—without commensurate safeguards, rendering the latter profile more volatile despite lower per-event frequency. Expert surveys, drawing on historical analogues like the absence of natural human extinction versus the Cuban Missile Crisis's near-miss in 1962, underscore that while natural risks provide a stable "background" rate near zero, anthropogenic ones introduce novel, agency-dependent pathways untested over evolutionary timescales.  Recent Expert Surveys and Quantitative Models.    In 2020, philosopher Toby Ord published The Precipice, aggregating expert assessments and first-principles analysis to estimate an overall 1 in 6 probability of existential catastrophe—defined as human extinction or irreversible civilizational collapse—occurring before 2100. This figure contrasts with historical natural risks, estimated at roughly 1 in 10,000 per century, highlighting anthropogenic drivers as the primary concern. Ord's breakdown attributes the largest shares to artificial intelligence misalignment (1/10), engineered pandemics (1/30), and unaligned biotechnology (1/30), with nuclear war and climate extremes each at 1/1,000; other environmental damage and natural pandemics contribute smaller fractions, totaling anthropogenic risks far exceeding natural ones.  Risk Category    -- Estimated Probability (this century).  Unaligned AI,    1/10.  Engineered Pandemics,    1/30.  Other Misaligned Tech.    1/30.  Nuclear War    1/1,000.  Climate Change,    1/1,000.  Other Environmental Damage,    1/1,000.  Natural Pandemics,    1/1,000.  Asteroids/Comets,    1/1,000,000.  Supervolcanoes,    1/1,000.  These estimates derive from Ord's review of domain-specific literature and consultations, though they incorporate subjective calibration amid sparse direct evidence. The range of probability estimates across assessments reveals significant variation; for instance, superforecasters in the Existential Risk Persuasion Tournament assigned approximately 1% chance of extinction by 2100, compared to expert medians around 6%, illustrating differences between forecasting communities. Expert elicitations also differ, with some conferences yielding medians around 19% for the next century, reflecting diverse priors and methodological emphases.  Subsequent surveys have focused predominantly on AI due to its perceived urgency. A 2023 elicitation of over 2,700 machine learning and AI authors found 38% to 51% assigning at least a 10% chance to advanced AI yielding outcomes as severe as extinction, depending on question framing. The 2024 AI Impacts survey of 2,778 AI experts reported a median 5% probability of AI causing human extinction or equivalently dire results, with a mean of 16.2% and the top decile exceeding 25%; this reflects wide disagreement, as 5% of respondents foresaw zero risk while others projected substantial tail hazards from loss of control. Earlier, a 2022 poll of AI researchers indicated 17% estimated a 10% or greater chance of existential catastrophe from inadequate AI control. Broader existential risk surveys remain scarce post-2020, with domain experts often prioritizing AI over other anthropogenic threats like biotechnology or nuclear escalation due to scalability concerns.  Quantitative models for extinction probabilities employ varied methodologies, including Bayesian updates from historical baselines, demographic projections, and scenario simulations, but face inherent challenges in rare-event forecasting.   Structured techniques like Delphi elicitations aggregate anonymized expert iterations to mitigate biases, as in assessments ranking engineered pathogens or bioweapons above natural risks. Probabilistic demographic models, such as those applying the doomsday argument, infer elevated near-term extinction odds (e.g., 1 in 200 million for long-term survival under observer-selection effects) by conditioning on humanity's current temporal position. Evaluations of these approaches reveal no dominant method, with subjective expert priors often dominating due to data paucity; for instance, integrated assessments place anthropogenic extinction rates above 1 in 1,000 annually under pessimistic assumptions, though calibration against observed near-misses (e.g., pandemics) suggests overestimation risks. Such models underscore causal uncertainties, as extinction pathways involve compounded failures in detection, response, and resilience. Become a supporter of this podcast: https://www.spreaker.com/podcast/extinction-of-the-human-species--7081249/support.This episode includes AI-generated content.