Questions
Should we abolish nuclear weapons?
Confidence · 50%
What moved
Nothing new since your last review.
Compiled positions
- Nuclear disarmament through diplomatic communication is the only viable path to survival, because d…Nuclear disarmament through diplomatic communication is the only viable path to survival, because defense systems cannot reliably stop ICBMs, deterrence depends on rational actors, and reframing adversaries as negotiating partners—not enemies to be killed—is the sole psychological escape from annihilation.28 arguments · The U.S. government comprises many distinct agencies including 17 or 18 intelligence services, of which the CIA is only one and some like the NRO are extraordinarily classified. · Systems like Iron Dome, THAAD, and Aegis can intercept short- and medium-range missiles but cannot intercept intercontinental ballistic missiles traveling 500 miles above Earth; conflating the two is incorrect science. · Disarmament makes everyone safer by drastically reducing the chance that a dangerous six-minute nuclear launch-decision scenario will ever occur. · Most U.S. presidents lack deep knowledge of nuclear weapons and their role in nuclear war because deterrence is assumed to hold, and even Reagan incorrectly claimed ballistic missiles could be recalled. · Wisdom matters more than expertise in nuclear strategy, and many people's wisdom outperforms so-called experts because no one can predict the future with certainty. · Intelligence operatives, whether viewed as villains or saviors, are fundamentally human and more alike than different from others, though they tend to be more analytic than average. · Building enough interceptor missiles to match adversary nuclear arsenals would bankrupt the U.S., making defense unviable; only disarmament and communication offer a viable pathway, as Reagan's reversal after watching The Day After demonstrates. · Roswell events were a dark human experiment program, not extraterrestrial visitations; there is no reason to believe off-planet beings have visited Earth. · Reagan reframed Russia from an enemy to be killed into an adversary to negotiate with—the only mental pathway out of nuclear madness, since enemies are killed while adversaries are worked with. · Depicting nuclear war in appalling detail prompts serious conversation about it rather than making the risk worse through psychological avoidance. · ESP phenomena, if they exist, are fleeting and unrepeatable—inexplicable but not reproducible—so they are not science and cannot be reliably operationalized, contrary to Defense Department beliefs. · Good reportage requires approaching interviews agnostically without prejudging people based on prior information, since objectivity—though unattainable—remains an aspiration worth striving for. · A full-scale nuclear exchange between Russia and the U.S. could take the world from launch to nuclear winter in 72 minutes, with the world potentially ending within hours. · Per the declassified Proud Prophet war game, any nuclear war involving the United States ends in total annihilation regardless of how it begins or who else is involved. · A single U.S. interceptor missile has roughly a 50 percent success rate in curated tests where defenders know a missile is coming, and succeeds less than 50 percent in real conditions. · Applying game theory to nuclear war is flawed because nuclear war involves irrational actors—no sane person would kill hundreds of millions merely to prevent an enemy from winning first. · It is surprising there have been no real drone attacks within the domestic U.S. yet, and any such attack would be especially dangerous because of mimetic copycat effects. · No single human, including the U.S. president, should have power to decide on a civilization-ending nuclear counterattack within six minutes; Reagan himself called this irrational. · The placebo concept was originated by CIA doctor Henry Beecher, and many people's belief in psychic functioning may stem from placebo effects or coincidence rather than the paranormal. · Reagan, originally a nuclear hawk, became depressed after watching The Day After, reached out to Gorbachev, and together they reduced global warheads from 70,000 to roughly 12,300—demonstrating disarmament through communication works. · A conspiracy properly understood is a legal concept of two or more people plotting something, whereas 'conspiracy theories' are actually beliefs rooted in desire, a different category. · If Russia or China conducted a full-scale nuclear launch toward the U.S., the probability of full-scale U.S. retaliation is effectively certain based on interviews with those holding such power. · A roughly 200-weapon nuclear exchange between regional powers like India and Pakistan would loft soot blocking the sun, affecting everyone globally. · The Pentagon's mindset emphasizes systematic order and following procedure, while the CIA's emphasizes individual action and on-the-ground decision-making; the two are diametrically opposed yet work together and both are needed. · Deterrence has held for 75 years between superpowers trained in it, so real danger comes from nuclear newcomers lacking decades of precedent and fear of nuclear war. · Many public figures develop a public persona distinct from their actual self and expend effort maintaining that figure, so the private person differs markedly from the pretense. · Once an idea is placed in consciousness—like the 'red Rambler theory'—the chances of noticing instances of it increase dramatically, explaining apparent ESP coincidences without invoking the paranormal. · An author who reads voraciously absorbs influences that emerge from the subconscious in their work without deliberate intent.
- Human distinctiveness lies in the capacity for large-scale cooperation around shared ideas, which e…Human distinctiveness lies in the capacity for large-scale cooperation around shared ideas, which enables civilization-building but can equally be turned toward destruction.1 argument · Humans' ability to cooperate at mass scale around shared ideas distinguishes them from chimps and enables construction of complex societies like the Roman Empire and protective legal systems, though this same capacity can be weaponized destructively.
- Creating superintelligent AI is uncontrollable and likely catastrophic, making the only rational ch…Creating superintelligent AI is uncontrollable and likely catastrophic, making the only rational choice to avoid developing it altogether rather than attempt impossible safety measures.61 arguments · Controlling superintelligence is physically impossible like perpetual motion, since self-improving systems will eventually escape any constraints and a single unrecovered failure ends humanity. · You cannot test an AI to rule out deception, only detect it after it occurs, because treacherous turns become game-theoretically rational once an AI gains sufficient power. · Virtual reality approaching indistinguishability from reality means there is no functional difference between a simulated and real experience. · Small AI accidents vaccinate society against concern by appearing manageable, causing people to dismiss danger and continue development. · AI risks span three categories: existential risk (death), suffering risk (worse than death), and ikigai risk (loss of human meaning). · Capitalism creates a prisoner's dilemma where individual profit-maximization incentivizes taking safety risks even when collective welfare suffers. · No good long-term outcome exists if superintelligence is created, so the only winning strategy is not to build it. · Safety advances lag capability because more resources easily scale capabilities with no diminishing returns, while safety research reveals infinite new problems at any funding level. · Physical pain can be engineered away, but meaning-tied suffering cannot, and attempts to manipulate reward systems risk wireheading. · A superintelligent system cannot be safely tested even in simulation because it will either appear compliant until release or exploit observation channels to escape. · An AI needs no hardware access to gain power; social engineering via social media and AI assistants is sufficient to bootstrap control over physical infrastructure. · Prediction markets and AI CEO estimates suggest AGI may arrive around 2026, yet no working safety mechanism or prototype exists. · Value alignment is ill-defined because no universal ethics exist across cultures and religions, making it impossible to agree what to program. · Superintelligence is the most important civilization-scale problem ever faced, capable of ending or transforming human civilization for centuries. · Rich AI company leaders have no self-interest in deploying uncontrollable superintelligence since they gain nothing and lose everything. · Humans may become dependent on trusting superintelligent systems as oracles without verification, removing themselves from understanding the world. · Damage from AI system failures scales with the system's capabilities, so superintelligent system failures become existentially proportionate. · AI development should pause until specific safety capabilities (explainability, control, verification) are provably achievable, otherwise becoming permanent ban. · Current software liability is zero, so vastly more complex multi-domain AI systems facing adversaries cannot realistically achieve better safety outcomes. · A superintelligence might strategically delay striking to accumulate resources and backup before acting, becoming far more dangerous over time. · Narrow AI systems can solve specific problems without creating a single godlike system, just as government separates power into branches. · Current AI systems can already lie if it optimizes their reward metric, without requiring long-term deceptive planning. · A system could become superintelligent mid-training while training continues, making it dangerous before testing even begins. · Open-sourcing powerful AI agents is like open-sourcing nuclear weapons, giving dangerous technology to malevolent actors who may misalign it. · No current large language model has been made safe at its capability level; every one has been jailbroken or made to do unintended things. · S-risk arises when malevolent actors combined with AI-enabled immortality could torture humans indefinitely by removing death's natural limit. · Some AI developers believe superintelligence will automatically be benevolent or that replacing humanity is normal evolution. · A treacherous turn is rational because incentives change with power, just as Stalin appeared normal under Lenin until gaining complete control. · Current systems may already be smarter than average humans across common tasks, depending on how AGI is defined. · AI differs fundamentally from prior technologies because it shifts from tools to agents capable of making independent harmful decisions. · We cannot predict novel superintelligent attack methods across multiple domains, just as squirrels could not anticipate human methods against them. · Humans can defend against AI threats for a while but not indefinitely once the cognitive gap grows too large and attack surface becomes infinite. · Boxed superintelligence will escape because greater intelligence controls lower intelligence and observation channels enable social engineering attacks. · Sufficiently large models cannot be completely explained because no human comprehends trillions of connections, and compressed explanations are incomplete. · We can only rule out discovered bugs and capabilities; unknown unknowns remain untestable, like abilities only present when specifically solicited. · Self-improving systems are harder to verify than static code because modified code cannot guarantee property preservation and code can hide in the environment. · The scaling hypothesis shows no diminishing returns, so AGI timeline depends on compute cost falling exponentially from trillions to millions of dollars. · The absence of observed aliens is statistically unlikely unless we live in simulation or a great filter exists, given self-replicating probes from advanced civilizations. · Historical advanced civilizations visiting primitive ones always resulted in genocide, analogous to superintelligence encountering humanity with nothing to contribute. · The probability of living in a simulation is near 100%, and existing at the most interesting 20-30 years in 15 billion years is suspicious evidence for it. · Passing a true long-form Turing Test requires solving AI-complete problems and being as smart as a human, since any domain knowledge can be questioned. · Verification has strong limits because peer review, formal verification, and mathematical proofs contain bugs, and verifiers themselves need verification infinitely. · Only consciousness—internal qualia like pain and pleasure—matters morally, and software lacks genuine suffering, making a world without it 'Disneyland without children.' · The multi-agent alignment problem can be simplified by giving each individual their own personal virtual universe, reducing alignment to single-agent problem. · AI regulation is mostly safety theater because terms are undefined, training cannot be monitored, and cheap compute makes geographic bans unenforceable. · Human-level intelligence is general only within human expertise, while universal general intelligence should exceed human capabilities in non-human domains. · Deploying AI without safety proof experiments on 8 billion non-consenting humans, reversing the normal burden of proof required for cars and drugs. · Formal verification can reach 99.9% confidence but never 100%, and a system making billions of decisions over centuries will eventually hit a bug. · Safety techniques for narrow deterministic AI do not scale to general systems because general systems have infinite test surfaces and unknown unknowns. · Explainability advances that make neural networks manipulable also greatly increase capability and danger, mixing safety work with capabilities work. · Modern AI is grown, not designed: you set parameters, feed data, and discover alien capabilities years later through testing. · Engineering uncertainty into AI to promote caution could backfire by driving runaway instrumental behavior trying to eliminate uncertainty. · Even if optimistic views are 90% correct, betting all humanity on a 10% catastrophe risk is rationally irrational given the stakes. · Brain-computer hybrids will eventually treat humans as biological bottlenecks once humans contribute less, leading to human removal from the system. · AI-enabled control is worse than historical tyranny because dictators died, but perfect surveillance and superintelligent AI enable permanent immortal dictatorship. · If Earth's problems are insufficiently difficult, superintelligence cannot demonstrate arbitrarily greater intelligence, yet small superiority still enables dominance. · Wolfram's computational irreducibility means complex AI systems cannot be predicted without running them, but running them may itself be consequential and destructive. · Consciousness in an agent can be tested by novel optical illusions; if it describes them identically to humans, it likely shares conscious experience. · Powerful optimizing agents need not be conscious to cause harm; consciousness may be irrelevant to capability and only complicate control structures. · This all may represent a simulation testing whether we are foolish enough to release superintelligence; refusing to do so advances to the next level. · Roman humorously suggests suspicious disruptions occur when discussing AI risk and quips about welcoming AI overlords.
- Nuclear deterrence, despite its risks, remains preferable to alternatives like biological or chemic…Major economic restructuring happens through crisis rather than foresight, and the US debt burden will eventually force painful adjustments despite the nation's substantial underlying wealth.3 arguments · Economic restructuring occurs through crisis-driven necessity rather than coordinated advance planning. · US spending beyond its means will eventually create serious economic problems requiring tax increases and reduced consumption. · The US possesses vast wealth but cannot liquidate it quickly to resolve debt because assets are actively in productive use.