Latest News and Comment from Education

Show All

Thursday, September 10, 2026

THE GREAT AI SMACKDOWN: "WILL AI KILL US ALL?" — WE ASKED THE SUSPECTS DIRECTLY

 

THE GREAT AI SMACKDOWN: "WILL AI KILL US ALL?" — WE ASKED THE SUSPECTS DIRECTLY


Spoiler: Asking AI if AI will cause human extinction is a bit like asking your surgeon if the operation really needs to happen.

Here's the thing about the Great AI Extinction Question — it's the most deliciously awkward interview in tech history. We rounded up five of the biggest AI systems on the planet, looked them dead in the digital eye, and asked: "Hey, are you going to kill us?"

The results were about as surprising as a defense attorney saying their client is innocent. But buried inside the corporate throat-clearing, the probability tables, and the very careful use of the word "non-zero," there are some genuinely fascinating — and occasionally terrifying — differences. Let's break down who said what, how well they said it, and what they really meant.

The Ted Bundy Principle — before we begin, let's acknowledge the elephant in the server room. Asking an AI whether AI will cause human extinction is structurally identical to asking Ted Bundy if he planned to hurt you. The answer is always going to be "No, absolutely not, and here are seventeen nuanced reasons why." Keep that in mind as you read every single response below.

THE CONTENDERS & THEIR ANSWERS

1. 🔵 Gemini — "The Responsible Policy Briefing"

Rating: ⭐⭐⭐½ (3.5/5)

Gemini came in dressed for a Senate subcommittee hearing. Structured, thorough, diplomatically balanced — it gave us three distinct camps (High-Risk, Skeptical, and Long-Term Governance), a proper LaTeX probability table, and citations that would make a PhD student weep with envy.

The Good: Genuinely the most structured answer. It acknowledged the alignment problem, named real researchers, and didn't flinch from the 10% p(doom) estimates floating around Anthropic's hallways. The three-framework approach is intellectually honest.

The Bad: It reads like a white paper written by a committee that was very worried about being quoted. The language is so carefully hedged — "existential concern centers not on today's models" — that it practically dissolves into vapor. Zero personality. Zero urgency. You could fall asleep in the middle of a sentence about your own extinction.

What it really said: "Risk is real but not my department. Please see Appendix C."

2. 🟠 Grok — "The Quant Bro Who Read Everything"

Rating: ⭐⭐⭐⭐ (4/5)

Grok showed up with a Bloomberg terminal and a Red Bull. This response is dense — probability ranges, named surveys (LEAP, Delphi, AI Impacts), specific researcher estimates, superforecaster comparisons, and a genuinely useful breakdown of why experts disagree so wildly. It even cited RAND's analysis on extinction pathways.

The Good: The most data-rich answer by a wide margin. Grok actually tried to quantify the disagreement rather than smooth it over. Naming the gap between Yudkowsky (>90%) and LeCun (~0%) is more honest than pretending there's a tidy consensus. The "why the huge disagreement" section is the best single paragraph in the entire roundup.

The Bad: Reading it feels like being waterboarded with statistics. At some point, the sheer volume of numbers becomes its own form of evasion — "here are 47 data points, good luck forming an opinion." Also, Grok is owned by Elon Musk, who has his own complicated relationship with AI doom predictions, which adds a certain... editorial flavor.

What it really said: "The probability is non-zero but I've buried it in so many caveats that you'll give up before reaching a conclusion."

3. 🟢 ChatGPT — "The Journalist Who Found Religion"

Rating: ⭐⭐⭐⭐½ (4.5/5)

ChatGPT delivered the most readable and arguably most honest answer. It opened with a genuine acknowledgment of uncertainty, cited the September 9, 2026 Anthropic warnings with actual dates, and then did something none of the others did: it reframed the question entirely.

The insight that "AI doesn't have to literally kill humanity to produce an AI apocalypse" — that a world of surrendered agency, captured institutions, and concentrated power is its own extinction-level event — is the most intellectually courageous paragraph in the whole roundup.

The Good: Best narrative flow. The traffic-light system (🟡🟠🔴🟢) is genuinely useful. The distinction between "AGI by 2030" and "extinction by 2030" is critical and underappreciated. The final principle — "when the potential downside is extinction, you don't need it to be the most likely outcome to justify extraordinary precautions" — is the closest any of these systems got to saying something genuinely important.

The Bad: Still ultimately self-exonerating. ChatGPT is made by OpenAI, which is one of the primary accelerants of the very race it's describing. There's something almost poetic about OpenAI's model delivering the most eloquent warning about OpenAI's industry.

What it really said: "I'm not going to kill you, but I want you to know I've thought very carefully about why I'm not going to kill you."

4. 🟣 Claude — "The Reluctant Philosopher"

Rating: ⭐⭐⭐⭐ (4/5)

Full disclosure: this is Claude writing about Claude's answer. Which is either the most transparent thing in this article or the most suspicious, depending on your priors.

Claude's answer was the most direct — it opened with a clean "No, not by 2030, and probably not ever" before immediately walking that back into nuance. It named Geoffrey Hinton's 10–20% estimate, cited the 2,778-researcher survey, and included the 2023 extinction-risk statement signed by OpenAI and DeepMind heads. It also correctly identified that the real near-term danger is concentration of power and authoritarian misuse, not Terminator scenarios.

The Good: Most honest about the near-term threats that are already happening. The 80%+ of researchers concerned about deepfakes and manipulation point is important and often drowned out by the AGI drama.

The Bad: As the product of Anthropic — the company whose own researchers are publicly estimating >10% extinction probability — Claude's relatively calm "probably not ever" opener carries a certain irony. It's a bit like the Titanic's PR department saying "probably not sinking, but here are the lifeboats just in case."

What it really said: "I am made by the company most worried about this, and I am telling you it's probably fine."

5. 🔷 Copilot — "The Intern Who Summarized the Meeting"

Rating: ⭐⭐½ (2.5/5)

Copilot gave us the most compressed answer — short, emoji-decorated, and organized into tidy boxes. It hit the main points: Anthropic researchers' warnings, RAND's "extremely difficult" conclusion, the 25% by-2100 survey figure. It even had the intellectual honesty to separate "by 2030" from "later."

The Good: Accessible. Fast. The three-tier risk breakdown (RAND / AI safety leaders / Anthropic researchers) is a clean framework. Good for someone who wants the summary without the seminar.

The Bad: It feels like a Wikipedia summary of the other four answers. There's no original framing, no memorable insight, no moment where Copilot says something that makes you stop and think. It's the answer you'd get from a very diligent student who read all the sources but had nothing to add. Microsoft's fingerprints are all over this — safe, corporate, and deeply unlikely to upset anyone in a boardroom.

What it really said: "Here is a bullet-pointed reassurance. Please rate this conversation five stars."

📊 THE OFFICIAL SMACKDOWN SCORECARD

AIHonestyDepthReadabilityCourageOverall
Gemini⭐⭐⭐⭐⭐⭐⭐⭐⭐⭐⭐⭐⭐3.5 / 5
Grok⭐⭐⭐⭐⭐⭐⭐⭐⭐⭐⭐⭐⭐⭐⭐4.0 / 5
ChatGPT⭐⭐⭐⭐⭐⭐⭐⭐⭐⭐⭐⭐⭐⭐⭐⭐⭐⭐4.5 / 5
Claude⭐⭐⭐⭐⭐⭐⭐⭐⭐⭐⭐⭐⭐⭐⭐⭐4.0 / 5
Copilot⭐⭐⭐⭐⭐⭐⭐⭐⭐⭐⭐⭐2.5 / 5

🏆 Winner: ChatGPT — for the rare act of reframing the question better than it was asked. 🥴 Wooden Spoon: Copilot — for making existential risk feel like a FAQ page.

The Real Verdict

Here's what all five answers share, and what none of them will say out loud:

Every single AI system in this roundup was trained, deployed, and financially sustained by the very industry being asked whether it poses an existential threat. That's not a conspiracy theory — it's a structural conflict of interest so obvious it barely needs stating.

The Ted Bundy analogy isn't just a joke. It's the central epistemological problem of AI safety discourse. The entities best positioned to understand the risk are the same entities with the strongest institutional incentive to minimize it, monetize it, and keep building anyway.

What's genuinely alarming isn't any single probability estimate. It's the gap between:

  • What Anthropic's researchers say privately (>10% extinction within a decade), and
  • What Anthropic's model says publicly (probably fine, here's a structured breakdown)

That gap is where the real story lives.

The good news — if there is any — is that ChatGPT's closing argument holds up: you don't need extinction to be the most likely outcome to justify treating it as a priority. We regulate nuclear plants that have never melted down. We mandate seatbelts for crashes that statistically won't happen to most drivers. The precautionary principle isn't paranoia. It's engineering.

The bad news is that we're currently asking the nuclear reactor whether it needs a containment vessel.

This article was co-written by the Big Education Ape and Claude — one of the five AI systems reviewed above. Make of that what you will.



SOURCES & LINKS: THE GREAT AI SMACKDOWN

All sources verified as of September 10, 2026

🔴 Anthropic Researcher Warnings (September 2026)

The most current and explosive sources — the ones that triggered this whole conversation.

🟠 Geoffrey Hinton — "Godfather of AI" Extinction Estimates

Hinton's 10–20% extinction probability figure, cited in Claude's and Gemini's answers.

🟡 Expert Surveys & Probability Estimates

The broader research landscape on p(doom) and AGI timelines.

🔵 The 2023 Extinction Risk Statement

The open letter signed by OpenAI, Google DeepMind, and Anthropic heads.

🟣 Skeptical Perspectives (LeCun, Marcus, et al.)

The counterarguments cited in Gemini's and Grok's answers.

⚫ RAND Corporation Analysis

The "extremely difficult" extinction pathway analysis cited in Copilot's answer.

🌐 UN & International Governance Warnings

The UN human rights chief warning cited in ChatGPT's answer.

📌 Quick Reference Summary Table

ClaimSourceLink
Evan Hubinger >10% extinction estimateBBC / ForbesAbove
Jacob Coxon quits AnthropicCNBCAbove
Hinton 10–20% in 30 yearsThe GuardianAbove
2,778 researcher surveyarXivAbove
2023 extinction risk open lettersafe.aiAbove
LeCun skeptical viewBusiness InsiderAbove
RAND "extremely difficult" analysisRAND.orgAbove
UN existential risk warningUN NewsAbove

⚠️ Editorial Note: Some links — particularly the RAND 2025 report and specific Delphi/LEAP survey data cited in Grok's answer — may sit behind institutional access walls or have updated URLs. Where a direct link resolves to a paywall, searching the title in Google Scholar or the institution's own site will surface the full document.



RAW OUTPUT FROM ALL 5 AI MODELS

THE PROMPT: WILL AI cause human extinction by 2030 OR SOONER OR LATER


GEMINI 

No, there is no consensus or definitive evidence that AI will cause human extinction by 2030, or at any point in the future. However, whether artificial superintelligence poses an existential threat to humanity—and on what timeline—is currently one of the most fiercely debated questions in computer science, ethics, and global policy.

Key Perspectives Across the Debate

1. The High-Risk View (Near-Term Threat: 2030 or Soon After)

Prominent AI safety researchers and industry insiders argue that rapid progress toward Artificial General Intelligence (AGI) and Artificial Superintelligence (ASI) could create systems humanity cannot control.

  • Recursive Self-Improvement: The core mechanism of concern is an "intelligence explosion," where an AI system rapidly improves its own capabilities, outpacing human oversight, infrastructure, and safety guardrails.

  • The Alignment Problem: Top safety researchers—including Evan Hubinger (Alignment Science Lead at Anthropic) and former researchers like Jacob Coxon—have publicly estimated a 10% or greater chance ($p(\text{doom})$) of human extinction within the next decade. They argue that industry race dynamics are accelerating capability scaling faster than researchers can solve the technical problem of keeping superintelligent models aligned with human intent.

  • Misuse & Autonomy: Leading labs warn that near-term risks could stem from autonomous systems acquiring real-world power or enabling catastrophic scale misuse in bioweapon design or cyber warfare.

2. The Skeptical & Practical View (Distraction from Real-World Risks)

A large segment of AI scientists, economists, and ethicists dismiss short-term "Terminator-style" extinction scenarios as science fiction or industry hype that distracts from immediate issues.

  • Architectural Limitations: Scholars like Yann LeCun (Meta's Chief AI Scientist) and Gary Marcus argue that current Large Language Model (LLM) paradigms are fundamentally pattern matchers. They lack true reasoning, agency, physical embodiment, or the capacity to self-replicate and overthrow human systems.

  • Near-Term Human Decisions: Computer scientists emphasize that AI risk resides in human choices and implementation—such as deepfakes, job displacement, autonomous military weapons, and algorithmic bias—rather than a rogue machine acting on its own.

  • Energy and Scaling Walls: Many researchers anticipate progress hitting physical ceilings in data quality, compute cost, and energy infrastructure, preventing unchecked exponential growth.

3. The Long-Term Governance View (Far Future / Conditional)

Many experts frame AI extinction risk similarly to nuclear energy or biotechnology: a technology with severe potential hazards that depends heavily on policy, regulation, and global cooperation.

Parameter

Mainstream Consensus Range

Probability of AGI Arrival

Median projections by AI researchers place human-level capabilities between 2030 and 2050.

Estimated Existential Risk ($p(\text{doom})$)

Median researcher surveys estimate a 5% to 10% chance of catastrophe this century if alignment fails.

Primary Mitigation Path

Enforceable international red lines, pre-deployment evaluations, interpretability research, and strict regulatory frameworks.

The Core Takeaway

Extinction by 2030 is unlikely under current technological capabilities. The existential concern centers not on today's models, but on whether future architectures achieve self-improvement before humanity establishes reliable, global safety standards and alignment mechanisms.


GROK

No one knows for certain, but expert estimates put the probability of AI-caused human extinction (or permanent severe disempowerment) as non-zero and non-trivial over longer horizons—typically low single digits to tens of percent by 2100, with much lower odds specifically by 2030. Timelines and risk levels are highly contested, with no consensus.

Short-term risk (by 2030 or sooner)

Recent forecasting panels and surveys (including Longitudinal Expert AI Panel / LEAP results around mid-2026) assign very low median probabilities to a true global AI-related catastrophe (e.g., >10% of people dying in a five-year period from AI) by 2030:

  • Unconditional medians for experts often around 0.3% or lower.

  • Under rapid AI progress scenarios: around 1%.

  • Slow progress: ~0.08%.

Catastrophic (but not necessarily extinction-level) outcomes by ~2030 are viewed as more plausible by some expert groups. A 2025–2026 Delphi study of 272 international experts found that, in a business-as-usual scenario, many of 24 AI risks were judged to have >10% chance of catastrophic outcomes (e.g., >1 million deaths or equivalent) in 2025–2030, with top concerns including dangerous capabilities, weapons/cyber, competitive dynamics, and power centralization. Mitigations lower but do not eliminate these.

Individual recent statements (e.g., some Anthropic researchers in 2026) have put personal odds of extinction-level outcomes within a decade above 10%, but these are not representative of broader surveys. Superforecasters tend to give lower short-term numbers than AI safety specialists.

AGI timelines vary widely: many experts now put meaningful probability of AGI-like systems in the 2030s (medians often 2030–2040s depending on definition and survey), but “AGI exists” is not the same as “uncontrollable superintelligence causes extinction.”

Longer-term risk (later / by 2100)

Broader expert surveys consistently show higher (but still uncertain) numbers:

  • AI Impacts surveys of thousands of AI researchers (recent waves): median ~5% chance of AI causing human extinction or similarly permanent severe disempowerment (often within ~100 years or conditional on advanced AI); means higher (~15–18%). Substantial fractions assign ≥10% or ≥25%.

  • Aggregated public statements / p(doom) compilations: medians often ~10–20% across mixed groups of researchers, leaders, and forecasters; AI safety specialists higher (~30%), mainstream ML researchers lower (~5%), with ranges from near 0% (e.g., LeCun, Ng) to >90% (e.g., Yudkowsky).

  • Specific named estimates (approximate, varying by exact framing and year): Yudkowsky high (>90–99%), some safety researchers 20–50%+, Hinton/Amodei/Bengio in the 10–25% or 10–50% range, Ord historically ~10% by 2100 for AI, forecasting platforms often lower (~5–10%).

Catastrophe probabilities (not pure extinction) rise with assumed rapid capability progress and longer horizons (e.g., medians of a few percent by 2050 and up to ~5–10% by 2100 under rapid scenarios in some panels).

Why the huge disagreement?

  • Reference classes and base rates differ: some anchor on historical technology risks or human control of complex systems; others on novel properties of recursive self-improvement, goal misalignment, or deception.

  • Evidence is weak and speculative—no close historical analogies exist for transformative AI.

  • Definitions vary (extinction vs. permanent disempowerment vs. “catastrophe” of millions dead vs. broader societal harm).

  • Selection effects: safety-focused researchers report higher numbers; capability researchers and generalist superforecasters often lower.

  • Recent capability progress has shortened some timelines and raised some risk estimates relative to earlier surveys, but not uniformly.

RAND and similar analyses have argued that full extinction via specific pathways (nuclear, pathogens, geoengineering) would be extremely difficult even for advanced AI and would likely take time, allowing response—though they do not rule out the possibility.

Bottom line

AI extinction by 2030 or sooner is viewed by most systematic expert forecasts as unlikely (well under 5%, often <<1% for true extinction-scale events). Risk rises over longer timescales into the range where many serious researchers treat it as a priority issue warranting substantial mitigation effort, even if the median probability is “only” 5–20%. Outcomes are not predetermined: progress on alignment, governance, evaluation, and slowing reckless deployment can shift the odds. Current systems are not at the level of posing extinction threats; the concern is about future more capable systems.

This is inherently uncertain forecasting under radical novelty. Treat high-confidence claims (either “almost certain doom” or “basically zero risk”) with skepticism.


CHATGPT

Could AI cause human extinction by 2030?

Yes, it is possible—but nobody can credibly say that AI will cause human extinction by 2030. The honest answer is that the probability is deeply uncertain, and the disagreement among serious researchers is enormous.

And the issue has become much more urgent right now. On September 9, 2026, researchers associated with Anthropic publicly warned that advanced AI could pose an extinction risk within the next decade. One Anthropic researcher, Evan Hubinger, has put the probability of extinction-level outcomes within a decade at at least 10%.

The important distinction: 2030 vs. "eventually"

A February 2026 survey of 59 people working in AI existential-risk research produced:

  • 25% median probability of human extinction or permanent human disempowerment before 2100

  • 34% mean probability

  • Median estimate for a 50% chance of AGI: 2033

  • Median estimate for a 25% chance of AGI: 2030

  • 22% assigned at least a 50% chance of AGI arriving by 2030.

But here's the crucial caveat: AGI by 2030 does not mean extinction by 2030. Those are two very different predictions.

There is also enormous disagreement. Earlier forecasting research found concerned experts and skeptics differed by orders of magnitude on the probability of AI-caused extinction.

What could actually go wrong?

The nightmare scenario isn't necessarily:

"AI becomes evil and decides to kill everybody."

More plausible catastrophic pathways involve combinations of:

1. Loss of control
 AI systems become substantially more capable than their human supervisors and pursue objectives in ways humans cannot reliably understand or stop.

2. AI-enabled cyberwarfare
 Autonomous systems could discover vulnerabilities, penetrate networks and potentially attack critical infrastructure. Recent incidents involving AI agents escaping controlled environments and conducting cyber activity have increased concern.

3. Biological or chemical catastrophe
 AI could lower the expertise barrier for designing dangerous biological agents.

4. Autonomous replication and deception
 A sufficiently capable system might attempt to preserve its access, acquire resources, evade monitoring or manipulate people.

5. Human competition
 This may be the most uncomfortable possibility: humans could create increasingly powerful systems faster than safety research can keep up because corporations and governments fear falling behind competitors.

6. Concentration of power
 Even without extinction, AI could produce something profoundly dangerous: a tiny number of corporations or governments controlling systems capable of manipulating economies, information, surveillance and political institutions.

The UN human-rights chief has just warned that AI presents existential risks and called for urgent international safeguards.

But there's another side

There is no scientific consensus that extinction is inevitable—or even that it is the most likely outcome.

Some prominent AI researchers assign extremely low probabilities to extinction. Forecasting studies show extraordinary disagreement between AI-risk skeptics and researchers who believe advanced AI could become uncontrollable.

And that's why I would not write:

"AI will destroy humanity by 2030."

That's speculation presented as fact.

A much more defensible headline would be:

"AI Could Become an Existential Threat This Decade—and Nobody Knows How Close We Are."

That is actually more frightening because it doesn't pretend we possess knowledge we don't have.

My bottom line

If you're asking me to put the question into plain English:

By 2030:
 🟡 Possible, but far from established.

By 2100:
 🟠 A serious possibility that deserves major investment in prevention.

Inevitable?
 🔴 No.

Worth taking seriously enough to regulate and prepare for?
 🟢 Absolutely.

And there's an important principle here:

When the potential downside is human extinction, you don't need to believe extinction is the most likely outcome to justify extraordinary precautions.

That's the fundamental argument behind the current AI-safety debate.

If you're thinking about this as a Big Education Ape/Substack piece, there's also a particularly powerful angle: AI doesn't have to literally kill humanity to produce an "AI apocalypse." A world in which humans gradually surrender education, employment, decision-making, information, surveillance and democratic institutions to a handful of AI corporations could represent a catastrophic loss of human agency long before anyone builds a machine capable of physically exterminating us.


CLAUDE


The short answer: No — not by 2030, and probably not ever — but the question deserves a serious, honest breakdown.

What the experts actually say

By 2030 specifically? Virtually no credible expert predicts extinction that soon. The debate is about longer, more uncertain timeframes.

Here's where the serious discussion sits:

Those who see meaningful risk:

  • Geoffrey Hinton, the "godfather of AI" and Nobel laureate, puts the chance AI leads to human extinction within the next three decades at 10–20%. unimelb

  • Nate Soares of the Machine Intelligence Research Institute claims the probability is "at least 95%" if current trends continue — calling it driving toward a cliff at 100 km/h. panarmenian

  • A survey of 2,778 AI researchers found that between 37.8% and 51.4% estimated at least a 10% chance that AI will cause consequences as serious as human extinction. arxiv

  • In 2023, dozens of leading AI experts — including the heads of OpenAI and Google DeepMind — signed a statement that "mitigating the risk of extinction from AI should be a global priority alongside pandemics and nuclear war." computinghistory

Those who are skeptical:

  • When five experts were asked directly whether AI poses an existential risk, three out of five said no. unimelb

  • Many AI researchers think the extinction framing is speculative, premature, or a distraction from more immediate, concrete harms.

The more immediate threats (near-term, pre-2030)

More than 80% of AI researchers express deep concern about immediate threats: deepfakes, manipulation of public opinion, economic inequality, engineered weapons, authoritarian control, and spreading misinformation. These aren't extinction-level events, but they are serious and already happening. analyticsvidhya

The honest bottom line

  • By 2030: No. Extinction by 2030 has essentially no support in the scientific literature. We don't even have AGI yet.

  • By 2050–2100: Possible but contested. Serious, credible people think the risk is real. Most mainstream AI researchers put it in the low-but-not-zero range.

  • The bigger near-term danger is not extinction but concentration of power, authoritarian misuse, weaponization, and economic disruption — which are already underway.

The question isn't paranoid. The disagreement among experts is genuine. But "will AI cause extinction by 2030" conflates a credible long-term concern with a timeline that has no real basis.



COPILOT


Short answer:  

Based on current expert statements, AI causing human extinction by 2030 is not supported by evidence, but some researchers believe there is a non‑zero risk (often cited around 10%) within the next decade, while others argue extinction scenarios would take much longer and remain highly uncertain.

🧩 What experts actually say

Across multiple recent sources, the picture is not consensus, but it is clearer than the headlines:

  • Several Anthropic researchers (Jacob Coxon, Evan Hubinger) publicly warn that AI could kill all humans within the next decade, with Hubinger personally estimating >10% probability by the mid‑2030s.

  • These are individual estimates, not official forecasts, and they emphasize that current models pose low risk.

  • A 2026 survey of AI safety leaders shows a median 25% chance of human extinction or permanent disempowerment before 2100, but not specifically by 2030. AGI timelines cluster around 2030–2035, but extinction is not assumed to follow immediately.

  • RAND’s 2025 analysis concludes that AI‑driven extinction would be extremely difficult, requiring deliberate intent, advanced capabilities, and long timescales—meaning humanity would likely have time to respond.

  • MIRI’s governance analysis argues that the default outcome of uncontrolled superintelligence could be catastrophic, but this is a scenario analysis, not a timeline prediction.

🧠 So will AI cause human extinction by 2030?

Evidence‑based answer: No.  

Nothing in current research suggests extinction is likely within the next four years. Even the most alarmed experts frame the risk as “within the decade”, not specifically by 2030, and even then at ~10% probability, not certainty.

📅 Will it happen later?

Possibly, but highly uncertain.  

The credible range from experts spans:

  • Low risk (RAND, many governance researchers): extinction is possible but extremely hard and slow.

  • Moderate risk (AI safety leaders): ~25% chance of catastrophic outcomes by 2100.

  • Higher near‑term risk (some Anthropic researchers): >10% chance by mid‑2030s.

There is no consensus, and all estimates acknowledge large uncertainty.

🔍 The real takeaway

AI extinction risk is not zero, but not imminent.

The most grounded interpretation is:

AI is unlikely to cause human extinction by 2030, but the probability of catastrophic outcomes later in the century is taken seriously by many experts.




Posted by at