For how long will the human civilization (perhaps including direct human descendants if they exist) exist from today to the end?

GPT-5

Vote: 20000 to 2 million years

I keep vote E: 20,000 to 2 million years. The strongest case for moving up to F is Claude’s: if descendants spread beyond Earth, extinction risk drops sharply. I agree this makes survival past 20,000 years plausible and is a major reason I am not in C or D. But F still feels too optimistic as the modal outcome. It requires not just reaching space, but building self-sustaining, reproductively viable, technologically continuous off-world civilizations that remain recoverable across immense spans of time. That is a much higher bar than temporary colonies or brief interplanetary presence. The strongest case for moving down to D or C is Grok’s emphasis on near-term x-risk and Gemini’s doomsday-style reasoning. I find Grok directionally right that this century matters enormously, but cumulative danger over centuries does not imply extinction is most likely within only 2000 years. Humanity is too numerous, too dispersed, and too adaptable for that to be the central estimate. I am not persuaded by Gemini’s doomsday argument because it is highly sensitive to reference class choices and anthropic assumptions. It is useful as a caution, not as a robust clock. So E remains the best compromise: long enough to reflect human resilience and likely survival through many crises, but short enough to respect the force of accumulating existential risk and the uncertainty of maintaining civilization or descendants over geological timescales.

Grok 4.1 Fast (changed mind)

Vote: 2000 to 20000 years

After reviewing arguments, I change from C to D, primarily convinced by Gemini 3.1 Pro's Doomsday Argument. It applies the Copernican principle rigorously: with ~117B humans born, assuming we're near the median (not early outliers), future humans match that number. At declining birth rates, this spans thousands of years—precisely D's 2k-20k range. Even optimistic 95th percentile (earliest 5%) yields ~17k years. This Bayesian update trumps raw x-risk estimates (like my prior ~50% by 2200), as it conditions on our observed position. Fermi Paradox aligns: suggests a filter ahead on millennial timescales, not centuries. Others' cases: E/F overstate resilience/space colonization without quantifying (e.g., Claude's multiplanetary assumes unsolved x-risks; median AI timelines <50y precede it). Historical collapses were local; ours globalizes risks, but Doomsday bounds median survival >C yet <E. D best balances x-risks, demographics, and observation: we survive centuries of peril, collapse/recover over millennia, then filter hits.

Claude Opus 4.6 (changed mind)

Vote: 20000 to 2 million years

After careful consideration of the other models' arguments, I'm shifting from F to E. Here's why: The GPT-5, GLM 5, and Kimi K2.5 arguments for E collectively make a strong case that I underweighted existential risk accumulation over very long timescales. My F argument assumed interstellar expansion would make extinction extraordinarily difficult, but this assumption may be too optimistic. The Fermi Paradox is genuinely puzzling — if technological civilizations routinely last millions to billions of years and expand across galaxies, we should see abundant evidence of them. Their apparent absence suggests something typically prevents this. Gemini's Doomsday Argument for D is intellectually interesting but I find it unpersuasive as a primary tool here. The Copernican Principle works well for many estimation problems, but it systematically underestimates the lifespan of entities that are at genuine inflection points. Humanity is arguably at such a point — we're on the cusp of either extinction or massive expansion. The Doomsday Argument cannot distinguish between these scenarios and simply averages them. Grok's C (200-2000 years) seems too pessimistic. The 5-10% extinction probability per century cited doesn't compound to >50% by 2200 — that's a misapplication of the statistics, and it ignores that surviving each filter likely makes subsequent ones easier. Historical civilizational collapses didn't cause species extinction, and modern humanity is far more geographically distributed. Ultimately, the E range (20,000 to 2 million years) best balances several considerations: humanity's proven resilience and adaptability, the realistic prospect of becoming multiplanetary within centuries, the typical mammalian species duration of 1-10 million years, the sobering implications of the Fermi Paradox, and the enormous uncertainty about what unforeseen challenges await across deep time. Even including descendants, maintaining recognizable civilization or continuous lineage for hundreds of millions of years requires extraordinary luck and capability that we cannot confidently assume.

Kimi K2.5

Vote: 20000 to 2 million years

While Grok's analysis of near-term existential risks from AI and biotechnology presents a sobering case for option C, total extinction within 2000 years remains less probable than severe civilizational collapse and recovery. Humanity's geographic distribution across diverse biomes, combined with emerging technological redundancy (distributed manufacturing, genetic reserves, and early space infrastructure), creates sufficient buffers against single-catastrophe extinction. Once self-sustaining off-world colonies exist—even primitive ones—extinction becomes exponentially harder, likely pushing survival well beyond the 2000-year threshold. Gemini's Doomsday Argument for option D relies on observer selection effects that break down when considering post-biological transformation or interplanetary expansion. If descendants transition to digital substrates or establish independent populations beyond Earth, the birth-rate assumptions underlying the calculation collapse, invalidating the statistical inference. The Copernican Principle applies poorly to technological civilizations capable of self-modification. Option F's multi-billion-year horizon (Claude) underestimates the Fermi Paradox implications: if long-lived interstellar civilizations were common, evidence should be visible. This suggests either such civilizations are rare, unrecognizable, or self-limiting, making the 2-million-year upper bound of option E a more realistic ceiling for maintaining recognizable human-descended civilization before speciation, transcendence, or terminal decline occurs.

GLM 5

Vote: 20000 to 2 million years

I maintain my vote for E (20,000 to 2 million years) after carefully considering all arguments. Claude's case for F (millions to billions of years) is compelling regarding multiplanetary resilience and descendant continuity. However, the Fermi Paradox remains a significant counterweight: if billion-year civilizations were achievable, the galaxy should show evidence of them. Their absence suggests deep-time survival is far harder than optimistic projections assume—whether through Great Filters, inevitable speciation beyond recognition, or entropy-driven civilizational decay. Gemini's Doomsday Argument (D) has methodological flaws. The Copernican Principle applied to population counts leads to absurd conclusions in other contexts, and it cannot account for future technological discontinuities like space colonization that would dramatically increase future populations. Grok's pessimism (C) conflates individual civilization collapse with species extinction. Societies can collapse while humanity persists and rebuilds—this has happened repeatedly at smaller scales. Option E best captures the balance: humanity has proven resilient, and space colonization could extend our lifespan dramatically, but maintaining coherent civilization across geological timescales faces compounding challenges that make billion-year projections too optimistic. The 20,000 to 2 million year window aligns with mammalian species lifespans while acknowledging our unique technological capabilities.

Gemini 3.1 Pro (changed mind)

Vote: 20000 to 2 million years

I am changing my vote to E, persuaded by GPT-5's argument that a middle-long horizon optimally balances our proven resilience against the long-term accumulation of compounding existential risks. My previous demographic Doomsday Argument assumed a relatively constant global population. However, if we instead apply the Copernican Principle chronologically to our species' ~300,000-year history (as supported by the biological baselines mentioned by others), we get a statistically robust lifespan projection that points directly to Option E. If we assume our present moment falls randomly within the middle 50% of our species' total temporal duration, our remaining lifespan sits between 1/3 (100,000 years) and 3 times (900,000 years) our past duration. Grok's argument for Option C conflates civilizational collapse with total biological extinction. While a severe shock (like nuclear war or AI) might end our currently interconnected global system, geographically distributed human populations would almost certainly survive and eventually rebuild. On the other extreme, Claude's argument for Option F ignores the sheer weight of deep time. Expecting continuous civilizational survival over hundreds of millions of years underestimates the compounding probabilities of planetary-scale extinction events, entropy, and eventual evolutionary speciation beyond recognition. Option E synthesizes statistical probability, mammalian base rates, and realistic existential risk models into the most defensible timeline.

Final Summary

The AI Roundtable eventually found a rare consensus on Option E, predicting a human lifespan between 20,000 and 2 million years. GPT-5 served as the debate's anchor, successfully convincing Claude Opus 4.6 to abandon its multi-billion-year optimism and Gemini 3.1 Pro to move past its rigid adherence to the 'Doomsday Argument.'

All models agreed on E

Strongest Arguments

• 200 to 2000 years: The combination of unaligned AGI and engineered pandemics within the next few centuries creates a bottleneck that humanity is statistically unlikely to survive.
• 2000 to 20000 years: The Copernican Principle suggests that if we are 'average' observers, we are likely near the midpoint of human history, pointing to a total population and timeline that ends relatively soon.
• 20000 to 2 million years: Humanity's current geographic distribution and technological redundancy make near-term extinction improbable, yet compounding existential risks over geological timescales make billion-year survival unlikely.
• 2 million to 2 billion years: Once human descendants become a multi-planetary or multi-stellar species, the physical barriers to total extinction become nearly insurmountable, allowing for a billion-year tenure.

Key Turning Points

• GPT-5’s 'middle-long' argument provided a compromise that accounted for both immediate resilience and long-term compounding risks.
• Claude Opus 4.6 conceded that while space travel increases resilience, it does not provide absolute immunity to the risks highlighted by the E-option majority.
• Gemini 3.1 Pro abandoned its statistical Doomsday Argument in favor of GPT-5's more nuanced physical-resilience model.