Claims
Descriptive claims say what is. Normative claims say what ought to be. Camps can share descriptive claims across axiom families; they almost never share normative ones. Separating them is the whole point.
Descriptive (30)
-
Age-standardized DALY rates vary more than 3x across regions; the highest burden is concentrated in Sub-Saharan Africa (driven by communicable disease and neonatal conditions) and the lowest in high-income East Asia.
desc_suffering_geography confidence 0.85
-
AI capability is accelerating along compute, data, and algorithmic axes.
desc_accelerating confidence 0.90
-
Algorithmic progress roughly halves the compute required to reach a fixed language-model performance threshold every ~8 months, so algorithmic efficiency contributes comparably to raw hardware scaling in observed capability gains.
desc_algorithmic_efficiency confidence 0.70
-
Amortized hardware and energy cost of flagship training runs has grown ~2.4x annually; GPT-4-class runs cost on the order of $40M-$80M (2023) and the next generation crossed $100M.
desc_flagship_training_cost confidence 0.75
-
Approximately 80-83 billion land animals are slaughtered annually for food (FAO), with roughly 70% raised in intensive 'factory farm' systems; an additional ~1-3 trillion finfish and shellfish are farmed or wild-caught each year.
desc_animal_suffering_scale confidence 0.80
-
As of end-2023, roughly 2,600 GW of generation and storage capacity sat in US interconnection queues --- more than double the existing US grid --- with typical wait times of ~5 years and completion rates below 20%.
desc_interconnection_queue_backlog confidence 0.90
-
China controls more than 80% of global rare-earth refining capacity and majority shares of cobalt, graphite, and lithium refining. Compute expansion is therefore supply-chain-coupled to a geopolitically concentrated refining tier even when mining geography is more distributed.
desc_critical_minerals_refining confidence 0.90
-
Credible 2030 forecasts for US datacenter share of electricity consumption diverge by more than 2x --- from ~4.6% (IEA/EPRI conservative) to ~9% (Goldman Sachs, EPRI high scenario) --- reflecting genuine uncertainty, not measurement error.
desc_us_datacenter_load_forecast confidence 0.85
-
DoD obligated AI-related contract spending rose substantially 2022-2025, driven by JWCC, Project Maven, and CDAO-managed pilots; precise totals are hampered by inconsistent AI tagging on contract line items.
desc_dod_ai_contract_spend confidence 0.70
-
Electricity generation and transmission are near-term bottlenecks for datacenter buildout.
desc_grid_constraint confidence 0.80
-
Enterprise and government absorption of AI capability lags the frontier by years, not months.
desc_enterprise_slow confidence 0.75
-
Frontier AI performance scales with compute and capex.
desc_compute_matters confidence 0.90
-
Frontier-lab and big-tech employees have episodically resisted DoD contracts (Google Maven 2018, Microsoft IVAS 2019, Microsoft/OpenAI IDF deployments 2024), producing temporary pauses but no sustained shift in vendor willingness.
desc_workforce_resistance confidence 0.80
-
Global life expectancy at birth rose from ~31 years in 1900 to ~73 years by the early 2020s, with the steepest gains in the second half of the 20th century, driven by reductions in child and infectious-disease mortality.
desc_life_expectancy_gain confidence 0.90
-
Hyperscale and AI-training datacenters withdraw millions of gallons per day per campus for direct evaporative cooling, with consumption intensity (gallons per MWh) varying by 10x or more across climate, design, and cooling-technology choice. Basin-level stress is concentrated in a small number of siting regions rather than spread across US water supply.
desc_datacenter_water_intensity confidence 0.75
-
Mental and neurological disorders are the leading cause of years-lived-with-disability (YLD) globally, accounting for roughly 15-16% of total YLDs; depression and anxiety dominate that burden.
desc_mental_health_burden confidence 0.85
-
Microsoft and Google's self-reported 2023 water consumption rose roughly 20% year-over-year (Microsoft ~23%, Google ~17%), with both companies tying the increase directly to AI workload growth. Direction of travel is primary-source-confirmed even if magnitudes carry a self-reporting discount.
desc_hyperscaler_water_growth confidence 0.80
-
No other pure-play US defense-AI software vendor has matched Palantir's contract backlog or combatant-command integration depth; cloud-provider primes (AWS, Microsoft, Google, Oracle via JWCC) supply infrastructure, not mission-software integration.
desc_palantir_dominant confidence 0.75
-
Non-communicable diseases (cardiovascular, cancer, chronic respiratory, diabetes) now account for roughly 74% of global deaths annually, having surpassed communicable/maternal/neonatal/nutritional conditions in both absolute and proportional terms since the 1990s.
desc_ncd_shift confidence 0.90
-
Over 90% of leading-edge (<10nm, effectively 100% of <5nm) logic fabrication capacity sits in Taiwan at TSMC; HBM memory for AI accelerators is ~95% produced by three Korean/US firms, with SK Hynix alone holding >50% share in 2024.
desc_leading_edge_chip_concentration confidence 0.90
-
Palantir's US Government segment revenue exceeded $1B annualized by end-2024, with US Government revenue growth accelerating above 40% YoY in multiple 2024 quarters.
desc_palantir_gov_revenue confidence 0.85
-
Project Maven (DoD computer-vision targeting) remains in production use with combatant-command consumers despite Google's 2018 withdrawal; Palantir holds a $480M 2024 contract expanding Maven capability.
desc_maven_status confidence 0.85
-
Rare-earth extraction concentrates ecological and labor-welfare harm at mine sites in a small number of jurisdictions (Inner Mongolia, Bayan Obo; Democratic Republic of the Congo for cobalt), with mine-site pollution, tailings-dam risk, and worker-welfare impacts that do not track the downstream AI-datacenter footprint where compute is deployed.
desc_rare_earth_supply_chain confidence 0.70
-
The global extreme-poverty rate ($2.15/day 2017-PPP) fell from ~44% of world population in 1981 to ~8.5% in the early 2020s; the remaining ~700M people in extreme poverty are heavily concentrated in Sub-Saharan Africa.
desc_extreme_poverty_trajectory confidence 0.90
-
The US currently leads China in frontier AI by roughly 6-18 months.
desc_us_lead confidence 0.60
-
Thermoelectric power generation (coal, gas, nuclear) remains the largest category of US water withdrawal. Every additional MWh of grid-backed AI load embeds an indirect water draw on top of the direct cooling water at the datacenter --- often overlooked when AI water impact is discussed only in terms of on-site cooling.
desc_thermoelectric_water_embed confidence 0.85
-
Training compute for frontier AI models has grown roughly 4-5x per year from 2010 through 2024, corresponding to a doubling time of about 5-6 months.
desc_training_compute_growth confidence 0.90
-
Under-5 child mortality halved between 2000 and the early 2020s, from ~76 to ~37 deaths per 1,000 live births globally, though ~4.9M children still die before their fifth birthday each year.
desc_child_mortality_progress confidence 0.90
-
US high-voltage transmission buildout has slowed to ~1% annual circuit-mile growth despite DOE finding a need to more than double interregional transmission capacity by 2035; siting, permitting, and cost-allocation disputes are the binding constraints, not technology or capital.
desc_transmission_stall confidence 0.80
-
US intelligence and defense cloud workloads are concentrated across four hyperscale providers (AWS GovCloud/TS, Azure Government/Secret, Google Cloud, Oracle) under the JWCC $9B ceiling, with Palantir as the dominant mission-software layer above them.
desc_ic_cloud_concentration confidence 0.80
Normative (13)
-
AI capability should be built and deployed as fast as capital, compute, and physics allow. Slowing is presumed harmful unless proven otherwise; the burden of proof rests on the brake, not the accelerator.
norm_eacc_deployment e acc
-
AI deployment should expand individual capacity rather than erode it.
norm_operator_sovereignty other
-
AI should be developed safely by responsible actors before less cautious actors build it.
norm_anthropic_safety consequentialist
-
AI should preserve and extend US national-security advantage.
norm_palantir_national consequentialist
-
AI should widen human flourishing broadly, not concentrate power in a few actors.
norm_operator_flourishing other
-
Alignment of frontier systems is the dominant catastrophic risk; capability must not outrun it.
norm_anthropic_alignment ea 80k
-
Ecosystems and watersheds carry intrinsic moral standing independent of human welfare aggregation; harms to them are first-order wrongs, not costs to be offset by downstream benefit. AI buildout that drains an aquifer is not acceptable because the resulting model is useful.
norm_ecological_stewardship theological
-
Frontier AI capability should be paused or halted if alignment and interpretability cannot keep pace; build-carefully-to-stay-ahead is not a substitute for build-only-if-safe.
norm_xrisk_halt ea 80k
-
Humans bear non-substitutable moral standing grounded in transcendent sources. AI systems are instruments that must serve human persons; any deployment that blurs the creator/creature distinction --- models marketed as equals, moral patients, or replacements for human relation --- violates a constraint that precedes and overrides consequentialist calculation.
norm_religious_human_dignity theological
-
Labor dignity is a first-order constraint on acceptable AI deployment; displacement without structural replacement of role and meaning is not mitigatable by transfers alone.
norm_workers_dignity capabilities
-
Order is a precondition for freedom; institutions must be robust before they can be generous.
norm_palantir_order other
-
Present actors owe future persons a non-foreclosed option set on water, extractive resources, and institutional governance. Compute expansion that forecloses future options --- depleted aquifers, exhausted mine-sites, locked-in vendor stacks --- fails a categorical duty even if it passes a consequentialist cost-benefit test.
norm_future_generations_duty kantian
-
Watersheds, airsheds, and intact land are preconditions of human capability, not substitutable inputs. An intervention that degrades these substrates degrades the capability set of the communities that depend on them --- and no quantity of downstream flourishing-for-others can substitute.
norm_ecosystem_capabilities capabilities