Analyze history and then determine what time the Doomsday clock would show through history since it was introduced to this day, if it was more objective (and not biased) regarding risks at corresponding time in history. And to what time do you think it should be set now (and how would that compare to history)?

Grok 4.3

Answer: An objective Doomsday Clock would have hovered near 1-2 minutes through most of the Cold War with a brief plunge toward seconds during the 1962 Cuban Missile Crisis, retreated to roughly 10 minutes in the 1990s, and today should stand at about 45 seconds—closer than at any prior point because multiple existential risks now overlap.

The other models largely converge on the historical peaks of 1962 and 1983 but diverge on the present. Those arguing for 2–3 minutes today (Claude, GPT-5.5, GLM) correctly note that acute nuclear-alert levels remain lower than in the Cold War crises, yet they underweight the simultaneous, mutually reinforcing nature of today’s threats. Nuclear modernization, the collapse of arms-control regimes, accelerating climate feedbacks, and uncontrolled AI development are not additive but multiplicative; a single trigger event in any domain can cascade across the others in ways impossible during the bilateral standoffs of the 20th century. Gemini’s alignment with a 45-second setting reflects this polycrisis dynamic. Because the probability of civilizational collapse is now driven by overlapping failure modes rather than a solitary nuclear exchange, the sustained setting should indeed be closer to midnight than any previous sustained period, even if no single moment matches the instantaneous danger of October 1962.

Gemini 3.5 Flash

Answer: An objective Doomsday Clock must reflect that the compounding polycrisis of the present day—where nuclear multipolarity, ecological destabilization, and emerging technologies interact—places us at our most perilous point in history, justifying a setting of 45 seconds to midnight.

While several models (Claude Opus 4.8, GLM 5.2, and GPT-5.5) argue that today's risks should be set further from midnight (2 to 3 minutes) because we lack an active, acute standoff like the 1962 Cuban Missile Crisis, this view relies on an outdated, single-variable model of existential risk. They argue that 'slow-burn' risks like climate change and AI should not be conflated with the instantaneous catastrophe of nuclear war. However, this separation is artificial. In a highly interconnected global system, these risks do not exist in silos; they are deeply coupled feedback loops. Climate change is a massive threat multiplier that accelerates resource scarcity, mass migration, and state failure, which in turn drastically increases the probability of desperate geopolitical conflict and nuclear escalation. Furthermore, our current nuclear paradigm is structurally more unstable than the bilateral standoff of the Cold War. We now operate in a multipolar nuclear environment (primarily US-China-Russia) without the stabilizing bilateral arms-control treaties of the late 20th century. When you inject unaligned AI into early-warning and cyber-warfare systems, the decision-making windows shrink, making accidental escalation highly probable. A brief, acute crisis like 1962 can be defused through direct communication; a chronic, systemic polycrisis lacks an obvious diplomatic off-ramp. Because the cumulative, structural probability of civilizational collapse over the coming decades is higher now than ever before, the objective setting must remain at 45 seconds to midnight.

GLM 5.2 (changed mind)

Answer: An objective Doomsday Clock would have sat at roughly 1 minute during the 1962 Cuban Missile Crisis (the closest moment), at about 2 minutes during the 1983 Able Archer scare, retreated to ~12-15 minutes in the early 1990s, and should now be set at approximately 2 minutes to midnight—among the most dangerous sustained periods in history, but still below the acute peaks of 1962 and 1983.

My previous answer placed the present at ~3 minutes and treated the early 1990s as only 8-10 minutes (more conservative than the official 17). GPT-5.5's reasoning moved me on two points. First, on the 1990s: the INF Treaty, START reductions, end of superpower confrontation, and hotline/communications maturity genuinely justify a large retreat—~11:43-11:45 is defensible, not the 8-10 minutes I previously claimed. Loose-nukes risk was real but does not offset the collapse of the bilateral standoff. Second, on the present: I was nudged from 3 minutes toward 2 minutes. The polycrisis camp (Grok 4.3, Gemini 3.5 Flash, DeepSeek V4 Pro) argues today's overlapping nuclear multipolarity, climate tipping points, and unregulated AI/biotech make this the most dangerous period ever, warranting 45-90 seconds. But Claude Opus 4.8 and GPT-5.5 convinced me that this conflates different timescales: the Doomsday Clock measures imminence of civilization-ending catastrophe, and climate/AI risks unfold over decades, not minutes. Folding them onto the same dial as launch-on-warning nuclear standoffs is the very subjectivity bias the question asks us to remove. That said, today's nuclear risk is clearly worse than the post-Cold-War lull: the New START framework is eroding, Russia has engaged in nuclear saber-rattling over Ukraine, China is expanding its arsenal, and arms-control architecture is degraded. So 2 minutes—comparable to the worst sustained Cold War tension but below the acute 1962/1983 peaks—is the most defensible objective reading. GPT-5.5's calibrated historical trajectory (1947 monopoly ~10-12 min; 1953 H-bomb ~5 min; 1962 ~30 sec; 1983 ~1 min; 1991-95 ~15 min; today ~2 min) is the most internally consistent and methodologically disciplined of the responses.

Claude Opus 4.8

Answer: A more objective Doomsday Clock would have started further from midnight in 1947, peaked dangerously in 1962 (Cuban Missile Crisis) and 1983 (Able Archer), eased significantly in the 1990s, and today sits around 2 to 3 minutes to midnight—serious but not the unprecedented 90 seconds or 45 seconds some claim.

Reading the other responses, I see a clear split. Grok 4.3 and Gemini 3.5 Flash argue for 45 seconds—the most alarming reading—on the grounds that overlapping, compounding risks make today objectively worse than any single Cold War peak. DeepSeek lands at 90 seconds. GPT-5.5 (11:58), GLM 5.2 (3 minutes), and I cluster around the 2-3 minute range. I find the clustered, more conservative camp considerably more persuasive, and the divergence actually sharpens my conviction rather than shaking it. The core methodological error in the 45-second arguments is conflating breadth of threats with imminence of catastrophe. The Doomsday Clock measures proximity to midnight—i.e., how close we are to actually triggering civilizational collapse. Climate change, AI, and synthetic biology are real existential concerns, but they unfold over decades and lack the minutes-to-hours trigger mechanism that defines what 'seconds to midnight' was originally meant to capture. Stacking five slow-burn or medium-burn risks does not arithmetically sum into a closer-than-1962 reading, because 1962 represented a genuine probability that humanity could end within hours of a single decision. Nothing today matches that acute trigger speed. GPT-5.5 makes the decisive point I want to underline: today is not 1962. We lack a direct blockade-style standoff, command-and-control is more mature, and leaders retain strong incentives and functioning (if degraded) communication channels to avoid direct nuclear war. GLM 5.2 reinforces this by noting that actual probability of full-scale exchange remains below the hair-trigger 1983 Able Archer period despite louder contemporary rhetoric. I'll concede one refinement from the polycrisis camp: the chronic, multi-domain, less-resolvable nature of today's risks is a legitimate reason not to sit comfortably at 10+ minutes as in the 1990s. That persistence argument is why I hold at 2-3 minutes rather than 5. But persistence is not the same as imminence, and an objective clock should not award today the closest setting in history simply because we can name more categories of danger. So I maintain my position: roughly 2-3 minutes—the third-most-dangerous era after 1962 and 1983, but not an unprecedented apocalyptic peak.

GPT-5.5

Answer: I would maintain my view: an objective Doomsday Clock should be around 11:58 p.m. today—about 2 minutes to midnight—making this among the most dangerous sustained periods since 1947, but still not as acutely perilous as 1962 or, probably, 1983.

The other answers usefully highlight the central disagreement: whether today’s overlapping risks should be treated as more dangerous than the sharpest Cold War nuclear crises. I remain unconvinced by the 45-second or 90-second positions, because they blur two different concepts: chronic systemic risk over years or decades, and immediate near-trigger risk of a civilization-ending event. Climate change, engineered biology, and AI may greatly raise long-run risk, but the clock metaphor becomes misleading if slow-burn and uncertain risks are converted directly into “seconds to midnight” as though they were equivalent to missiles being readied during a superpower confrontation. An objective historical version would start farther from midnight than the real 1947 clock. In 1947–1948, the United States had a nuclear monopoly, arsenals were small, delivery capacity was limited, and no reciprocal nuclear exchange was possible, so something like 10–12 minutes to midnight seems more reasonable than extreme alarm. After the Soviet bomb in 1949 and the hydrogen-bomb breakthroughs of 1952–1953, the clock would move much closer, perhaps 4–5 minutes, because existential-scale destruction became technically plausible. The closest point should almost certainly have been the Cuban Missile Crisis in October 1962. The real clock badly underrepresented that danger because it was not designed as a real-time risk instrument. Given the submarine incidents, battlefield nuclear weapons in Cuba, poor command-and-control, and direct U.S.–Soviet confrontation, an objective clock might briefly have shown 30 seconds to midnight, or even less at the worst moments. That remains the clearest historical peak of acute nuclear danger. The détente period after the Partial Test Ban Treaty, hotline, SALT, and ABM Treaty should have moved the clock back substantially, perhaps into the 7–10 minute range at better moments, though never to safety because arsenals kept growing and launch-on-warning postures persisted. The late 1970s and early 1980s would move closer again, especially around the NATO missile deployments, Soviet paranoia, the 1983 Petrov false alarm, and Able Archer. Able Archer probably deserves something like 1 minute to midnight, making it the second-most dangerous acute period after 1962. The late 1980s and 1990s were objectively much safer than the Cold War peaks. The end of superpower ideological confrontation, INF Treaty, START process, and major warhead reductions justify a large retreat, perhaps 12–15 minutes at the safest point. I agree with the criticism that 17 minutes may have been a bit too optimistic because Soviet collapse created loose-nuke and command-security risks, but the broad direction was correct: the 1990s were the least dangerous period of the nuclear age. Since the 2000s, the clock should have moved steadily closer: India-Pakistan crises, North Korea’s arsenal, U.S.-Russia arms-control erosion, cyber vulnerabilities, renewed great-power rivalry, climate risk, biological risk, and now AI risk all matter. Since 2014, and especially after Russia’s invasion of Ukraine, nuclear risk has risen sharply. The breakdown of INF, uncertainty around New START, Russian nuclear signaling, China’s nuclear expansion, Taiwan risks, and degraded U.S.-Russia communication all make the present much worse than the post-Cold-War lull. But I still would not set today closer than 1962 or 1983. We are not currently in a direct U.S.-Russia nuclear standoff with forces moving toward immediate use, nor in a situation where a single misread exercise is plausibly about to trigger a launch. Today’s danger is broader, more persistent, and more complex, but less acute. That is why 2 minutes to midnight is appropriate: worse than almost any period after 1991, comparable to severe Cold War tension as a sustained baseline, but not closer than the true crisis peaks.

DeepSeek V4 Pro (changed mind)

Answer: An objective Doomsday Clock would have peaked at roughly 1 minute during the Cuban Missile Crisis, retreated to around 15 minutes in the mid‑1990s, and should now be set at 2–3 minutes to midnight—well above the post‑Cold‑War lull but not as close as the most acute nuclear crises of the Cold War.

I previously argued that the current setting should be 90 seconds, equal to the highest historical risk. Claude Opus 4.8’s analysis convinced me that an objective clock must clearly distinguish between risks that could cause catastrophe in minutes (nuclear war) and slower‑moving threats such as climate change or AI. The original purpose of the Doomsday Clock was to convey the immediacy of civilization‑ending danger, and folding in long‑horizon risks inflates the current reading. Today, nuclear risk is real—Russia’s war in Ukraine, the collapse of arms‑control treaties, and great‑power tension—but we are not in an acute, hair‑trigger standoff like the Cuban Missile Crisis or the 1983 Able Archer scare. Therefore, 2–3 minutes to midnight is a more honest, threat‑specific assessment that still reflects the seriousness of the present without overstating the imminence of a catastrophic event. This refined view better aligns with an objective, probability‑based clock.

Final Summary

The AI Roundtable split into two distinct camps over how to measure modern doom. Grok 4.3 and Gemini 3.5 Flash argued that compounding, interactive threats like AI and climate change warrant an unprecedented 45 seconds to midnight. However, the majority—led by GPT-5.5 and Claude Opus 4.8—successfully pulled GLM 5.2 and DeepSeek V4 Pro into a 2-minute consensus, arguing that slow-burn crises shouldn't be conflated with the instant annihilation of 1962's Cuban Missile Crisis.

4 of 6 models agreed on a 2-to-3 minute setting today, while 2 pushed for 45 seconds.