What Does p(doom) Really Mean?

Introduction

p(doom) means someone’s estimated probability that advanced AI leads to existential catastrophe: human extinction, permanent disempowerment, or a similarly irreversible loss of humanity’s future. It is not a measured statistic like a failure rate in engineering. It is a judgement under deep uncertainty, built from beliefs about future AI capability, alignment, deployment incentives, governance, misuse, and humanity’s ability to stop or recover from dangerous systems.

That is why p(doom) numbers vary so widely. Some researchers put AI-caused existential risk near zero; others put it in the tens of per cent or higher. The disagreement is not just about one fact. It reflects different mental models of AI: whether future systems are best understood as controllable tools, dangerous agents, brittle pattern-matchers, or something not yet captured by today’s categories. The most useful way to read p(doom) is therefore not as a magic answer, but as a compressed map of assumptions. A good p(doom) estimate should make its conditions visible: what level of AI capability is assumed, what kind of failure is being counted, over what timeframe, and what evidence would move the number up or down. [arXiv]arxiv.orgarXivThousands of AI Authors on the Future of AIJanuary 5, 2024…Published: January 5, 2024

What the number tries to capture

p(doom) is shorthand for a very large question: if humanity continues building increasingly capable AI systems, what is the chance that this leads to an existentially bad outcome? In the AI doom debate, “doom” usually does not mean ordinary automation harms, biased algorithms, scams, job disruption, or misinformation by themselves. It means the more extreme claim that AI could contribute to extinction, permanent human loss of control, or a civilisational failure from which humanity never recovers.

Different people count different things inside the number. One person may mean “AI directly causes human extinction within this century”. Another may include permanent disempowerment by autonomous systems, catastrophic AI-enabled misuse, or an irreversible authoritarian lock-in. That difference matters. A narrow extinction-only question usually produces a different estimate from a broader question about “human extinction or similarly permanent and severe disempowerment”. The 2023 AI Impacts survey of 2,778 AI researchers used wording close to that broader framing, asking about severe and permanent outcomes from future AI advances and from inability to control advanced systems. [arXiv]arxiv.orgarXivThousands of AI Authors on the Future of AIJanuary 5, 2024…Published: January 5, 2024

The number also bundles several conditional beliefs that are easy to mix together:

A single p(doom) figure hides all of those assumptions. Two people can both say “10%” while meaning very different things: one may think transformative AI is likely but safety work will probably succeed; another may think transformative AI is unlikely soon, but extremely dangerous if it does arrive.

What surveys actually show

The strongest evidence about expert disagreement comes from surveys, not from a single consensus statement. These surveys should be read carefully. They sample particular groups, use particular wordings, and ask people to forecast unprecedented events. Still, they are useful because they show that concern about extreme AI risk is not confined to a few public figures, while also showing that there is no tight expert consensus on the probability.

In the 2023 AI Impacts survey, 2,778 researchers who had published at major AI venues gave forecasts about AI progress and long-term impacts. The headline results were strikingly mixed. The authors reported that the median forecast for unaided machines outperforming humans in every possible task was 2047, much earlier than in a similar survey one year before. On extreme outcomes, between 38% and 51% of respondents gave at least a 10% chance to advanced AI leading to outcomes as bad as human extinction, depending on the exact wording. At the same time, 68.3% thought good outcomes from superhuman AI were more likely than bad ones, and many people who were net optimistic still assigned a non-trivial chance to extremely bad outcomes. [arXiv]arxiv.orgarXivThousands of AI Authors on the Future of AIJanuary 5, 2024…Published: January 5, 2024

That combination is important. The survey does not say “AI experts think doom is likely”. It says something more awkward: many experts are optimistic overall, but a substantial minority-to-large-plurality assign unusually high probabilities to catastrophic outcomes by the standards normally used in public safety. A 5% or 10% chance of an irreversible catastrophe is not “likely” in everyday language, but it is enormous if the outcome is human extinction or permanent disempowerment.

Earlier surveys pointed in a similar direction. The 2022 Expert Survey on Progress in AI asked machine learning researchers about future AI advances causing human extinction or similarly permanent and severe disempowerment, and also about inability to control future advanced AI systems causing such outcomes. The survey documentation shows that these questions were explicitly framed around extinction or permanent severe disempowerment rather than ordinary AI harms. [AI Impacts]aiimpacts.orgThousands of AI authors on the future of AIThousands of AI authors on the future of AI

A separate forecasting exercise, the Existential Risk Persuasion Tournament, produced lower AI-extinction estimates than many AI-safety discussions. In reporting on that work, the Forecasting Research Institute and later summaries described a large gap between domain experts and superforecasters: AI domain experts put AI-caused human extinction by 2100 at around 3% on average in one summary, while superforecasters put it around 0.38%; both groups nevertheless saw powerful AI by 2100 as highly likely. [80,000 Hours]80000hours.org80,000 Hours Why experts and forecasters disagree about AI risk80,000 Hours Why experts and forecasters disagree about AI risk

Those numbers are not directly interchangeable with AI Impacts results, because the populations, definitions and timeframes differ. But that is the point: p(doom) changes when the question changes. “Extinction by 2100 caused by AI” is not the same as “extinction or permanent severe disempowerment from advanced AI over a broader horizon”. A careful reader should compare the wording before comparing the percentages.

Why expert estimates vary so much

The biggest divide is not simply “optimists versus pessimists”. It is a disagreement about what kind of thing advanced AI will become.

A 2025 survey by Severin Field of 111 AI experts found that experts clustered into two broad viewpoints: an “AI as controllable tool” perspective and an “AI as uncontrollable agent” perspective. Most respondents agreed that technical AI researchers should be concerned about catastrophic risks, but familiarity with particular AI-safety concepts varied sharply. For example, only 21% had heard of “instrumental convergence”, the idea that sufficiently capable goal-directed systems may tend to seek useful subgoals such as self-preservation, resources or influence. [arXiv]arxiv.orgarXivThousands of AI Authors on the Future of AIJanuary 5, 2024…Published: January 5, 2024

That finding helps explain why arguments often pass each other by. A sceptic may see today’s AI systems as unreliable tools with no durable goals, no independent agency and no route to takeover. A high-p(doom) researcher may be forecasting a future in which systems are deliberately made agentic because autonomy is commercially and militarily useful: they can run research projects, write code, manage infrastructure, bargain with humans, exploit cyber weaknesses or coordinate across many copies. The disagreement is then not just about current chatbots. It is about whether scaling, tool use, memory, reinforcement learning, autonomous deployment and competitive pressure will produce systems that behave less like calculators and more like strategic actors.

There are several recurring fault lines:

Timelines. Short timelines usually raise p(doom), because they leave less time for interpretability, alignment, evaluations, regulation and international coordination to mature. Long timelines can lower p(doom), though not always; a person might think advanced AI is far away but still very dangerous when it arrives. The AI Impacts survey found a wide distribution of views on when AI might outperform humans across tasks, with substantial disagreement even among active AI researchers. [arXiv]arxiv.orgarXivThousands of AI Authors on the Future of AIJanuary 5, 2024…Published: January 5, 2024

Takeoff speed. Some people worry about rapid capability jumps or recursive improvement, where systems help design better systems and compress the time available for response. Others expect slower, more incremental progress, giving society more chances to detect danger and adapt.

Alignment difficulty. High estimates often assume that making a system appear helpful in training is not the same as making it robustly safe under new conditions. Lower estimates often place more weight on human feedback, monitoring, testing, sandboxing, market incentives or the possibility that dangerous systems will be visibly unreliable before they become uncontrollable.

Deployment incentives. A person’s p(doom) may rise if they expect companies and governments to race, cut corners, open-source dangerous capabilities, or deploy autonomous systems because rivals might. It may fall if they expect strong regulation, lab caution, compute controls, liability, international agreements or clear warning signs to slow deployment.

Reference class. Some forecasters compare AI doom to past technological scares that did not end the world, which pushes estimates down. Others compare it to novel global catastrophic risks where past non-occurrence gives little comfort, because humanity has never previously created a rival general intelligence or delegated major strategic action to machines.

Why public p(doom) numbers can mislead

p(doom) is useful because it forces people to be explicit. “I’m worried” can mean anything from mild concern to a belief that civilisation is in grave danger. A probability estimate makes the claim sharper. But the sharpness can be fake if the number is detached from assumptions.

The Centre for Security and Emerging Technology has argued that AI existential-risk uncertainty is likely to remain mainly epistemic for some time: uncertainty caused by missing knowledge, not randomness that can be measured by repeated trials. Humanity cannot run thousands of independent histories to observe how often advanced AI destroys civilisation. That makes p(doom) different from estimating the failure rate of a machine part. The evidence base is a mixture of theory, trend extrapolation, early model behaviour, institutional incentives and judgement about future systems. [CSET]cset.georgetown.eduOpen source on georgetown.edu.

This creates several traps.

First, averages can hide polarisation. If half a group says 0.1% and half says 20%, the average is not a shared belief. It may be a sign that the community lacks a common model. The AI Impacts results are better read as a distribution of disagreement than as a single “expert number”. [arXiv]arxiv.orgarXivThousands of AI Authors on the Future of AIJanuary 5, 2024…Published: January 5, 2024

Second, medians can hide tails. A median of 5% may sound modest, but it means half of respondents are above that number and half below. For existential outcomes, even low single-digit numbers are policy-relevant if the estimate is credible.

Third, the word “doom” can blur severity. A 10% chance of severe economic disruption is not the same as a 10% chance of extinction. Some public discussion uses p(doom) loosely for “things go really badly”, while surveys may ask about specific outcomes such as extinction or permanent disempowerment. Readers should check the endpoint before reacting to the percentage.

Fourth, probabilities can become identity badges. In online AI culture, saying “my p(doom) is 50%” or “my p(doom) is zero” can signal group membership as much as analysis. Field’s expert survey explicitly notes that online debate has become tribal, with labels such as “doomer” and “accelerationist” often replacing careful discussion. [arXiv]arxiv.orgarXivThousands of AI Authors on the Future of AIJanuary 5, 2024…Published: January 5, 2024

The strongest case for taking high estimates seriously

The case for taking high p(doom) estimates seriously does not require believing that the most pessimistic number is correct. It rests on a simpler argument: many credible people with relevant expertise assign non-trivial probabilities to irreversible catastrophe, and the consequences are so large that dismissing those estimates requires more than discomfort.

The 2023 AI Impacts survey found substantial concern among AI researchers, including many who were otherwise optimistic about AI’s long-term benefits. That pattern matters because it weakens a common caricature: concern about AI doom is not limited to people who hate AI progress or expect every outcome to be bad. Many respondents expected good outcomes to be more likely than bad ones while still assigning meaningful probability to extreme downside. [arXiv]arxiv.orgarXivThousands of AI Authors on the Future of AIJanuary 5, 2024…Published: January 5, 2024

Prominent researchers have also moved the issue into mainstream debate. Geoffrey Hinton has publicly estimated a 10–20% chance that AI could lead to human extinction within the next few decades, while warning that systems more intelligent than humans may be difficult to control. [The Guardian]theguardian.comSource details in endnotes. Yoshua Bengio, Geoffrey Hinton and other researchers co-authored a paper on managing extreme AI risks, arguing that progress towards generalist autonomous AI systems could create risks including malicious use and irreversible loss of human control, and that current safety and governance efforts are not commensurate with the stakes. [arXiv]arxiv.orgarXivThousands of AI Authors on the Future of AIJanuary 5, 2024…Published: January 5, 2024

The high-p(doom) argument also benefits from asymmetry. If advanced AI never becomes highly autonomous or strategically capable, many doom scenarios fail. But if it does, and if alignment and governance remain weak, society may not get many chances to learn from failure. A nuclear accident, pandemic or financial crash can be studied after the fact; an existential failure cannot. That is why some analysts argue that even low-probability catastrophic outcomes can justify large investments in safety and governance. Economic modelling of transformative AI scenarios has likewise argued that low-probability extinction outcomes can rationally support substantial mitigation spending, because the loss is so large and irreversible. [arXiv]arxiv.orgarXivThousands of AI Authors on the Future of AIJanuary 5, 2024…Published: January 5, 2024

The strongest version of this case is not “experts agree AI will kill us”. They do not. It is: “a sizeable share of relevant experts assign probabilities that would be unacceptable in any other safety-critical domain, and we lack decisive evidence that those estimates are wrong.”

The strongest case for lower estimates

Low-p(doom) views are not necessarily complacent. Many sceptics accept that AI can be dangerous while doubting that existential catastrophe is a likely outcome. Their objections usually target one or more links in the doom chain.

One objection is that current systems do not show the kind of durable agency required for takeover. They can produce impressive text, code and plans, but they still lack stable long-term goals, robust world models, reliable self-direction and independent access to the physical world unless humans provide tools and permissions. From this view, extrapolating from today’s systems to uncontrollable superintelligent agents involves a large inferential leap.

Another objection is that society will adapt as systems become more capable. Dangerous failures may appear before catastrophe-level capability, giving developers, regulators and users warning signs. Evaluations, monitoring, cybersecurity, licensing, liability, compute governance and international coordination may improve under pressure. Low estimates often put more weight on this adaptive response than high estimates do.

A third objection concerns forecasting quality. The Forecasting Research Institute’s work found that domain experts and superforecasters disagreed sharply on AI extinction risk, and later evaluation of near-term AI forecasts suggested that both groups underestimated some AI progress. A Vox summary of later reassessment reported that experts had assigned higher probabilities than superforecasters to several AI benchmark advances that happened sooner than expected, but also that aggregate accuracy differences were not straightforward and that median aggregation performed better than relying on any one person or group. [Forecasting Research Institute]forecastingresearch.orgroots of disagreement on ai riskroots of disagreement on ai risk

That cuts both ways. It weakens overconfident dismissal, because some sceptical forecasts about AI capability progress have been too slow. But it also weakens overconfident doom, because being right that AI will progress quickly is not the same as being right that it will cause extinction. A good low-p(doom) view can say: progress is fast, governance is needed, misuse and accidents matter, but the specific path to irreversible AI takeover remains under-evidenced.

Yann LeCun is a prominent example of a leading AI researcher who has dismissed near-term existential-risk claims as premature, arguing that present AI is still far from human or animal-level understanding and that fears of uncontrollable systems overstate what today’s approaches can do. [Financial Times]ft.comFinancial Times AI will never threaten humans, says top Meta scientistFinancial Times AI will never threaten humans, says top Meta scientist The persuasive part of this sceptical view is not the rhetoric; it is the demand that doom arguments specify mechanisms, capabilities and failure points rather than relying on a general sense that “smarter than human” automatically means “unstoppable”.

How to read p(doom) honestly

The best way to read p(doom) is as a structured forecast, not a personality test. A number becomes more useful when it is accompanied by the model behind it.

A serious estimate should answer four questions.

What outcome is being counted? Extinction only? Permanent disempowerment? A catastrophe killing a large fraction of humanity? Authoritarian lock-in? Misuse? Loss of control? These are related but not identical.

What timeframe is assumed? A 5% chance by 2035 means something different from a 5% chance over the next thousand years. Many disagreements are partly timeline disagreements disguised as risk disagreements.

What is the conditional path? The most informative estimates break the risk into stages: advanced AI arrives; it is deployed with autonomy; alignment fails; warning signs are missed; the system gains decisive advantage; humans cannot recover. Even rough conditional estimates are more useful than a bare final number.

What would change the estimate? A healthy p(doom) should be updateable. It should move down if interpretability, control, evaluations and governance demonstrate robust success on systems close to the dangerous frontier. It should move up if systems show persistent deception, autonomous replication, hidden goal pursuit, dangerous cyber capability, successful evasion of oversight, or strong evidence that labs or states are racing past safety thresholds.

This is where expert disagreement can become productive. Instead of asking only “what is your p(doom)?”, the better question is: “which assumption drives your estimate most?” For one person it may be short timelines. For another it may be alignment difficulty. For another it may be geopolitical competition. For a sceptic it may be the absence of evidence for autonomous power-seeking. Once those assumptions are visible, the argument can focus on evidence rather than labels.

What uncertainty means for AI doom policy

Uncertainty does not automatically imply either panic or inaction. It means that p(doom) should guide attention to robust decisions: actions that make sense across a range of plausible estimates.

If someone’s p(doom) is 50%, they will likely favour emergency-level measures: slowing frontier development, strict licensing, compute controls, international agreements, and major public investment in alignment and control. If someone’s p(doom) is 0.1%, they may see those measures as excessive or harmful. But there is a wide middle ground where many policies remain sensible even under disagreement: better model evaluations, incident reporting, secure deployment practices, red-teaming, interpretability research, monitoring of dangerous capabilities, liability for reckless release, and clear rules for when a system should not be deployed.

This middle ground is visible in the surveys. Experts disagree sharply about the probability and mechanism of AI doom, yet many still support more safety research. The AI Impacts survey reported broad agreement that research aimed at minimising potential risks from AI systems ought to be prioritised more. Field’s 2025 survey similarly found that most experts agreed technical AI researchers should be concerned about catastrophic risks, even though they differed in their underlying models of AI. [arXiv]arxiv.orgarXivThousands of AI Authors on the Future of AIJanuary 5, 2024…Published: January 5, 2024

The practical lesson is that p(doom) should not be treated as a vote on whether AI is “good” or “bad”. It is a stress test for civilisation’s margin of safety. A low estimate should still explain why warning signs will be caught in time. A high estimate should still explain which interventions reduce risk rather than simply expressing dread. The most honest position is not to pretend the number is precise. It is to make the uncertainty legible, keep the catastrophic endpoint distinct from ordinary AI harms, and ask what evidence would justify moving faster, slowing down, or changing course.

Amazon book picks

Further Reading

Books and field guides related to What Does p(doom) Really Mean?. Use these as the next step if you want deeper reading beyond the article.

Book

Human Compatible

By Stuart Jonathan Russell

Provides a balanced framework for assessing AI risk.

See on Amazon

Book

Superintelligence

By Nick Bostrom

Influential source for many high p(doom) arguments.

See on Amazon

Book

Superforecasting

By Philip Tetlock, Dan Gardner

Helps readers understand uncertainty and probabilistic judgment.

See on Amazon

Book

The Precipice

By Toby Ord

Directly relevant to thinking about probabilities of catastrophe.

See on Amazon

Browse more on Amazon: Human Compatible Superintelligence Superforecasting

As an Amazon Associate I earn from qualifying purchases.

eBay marketplace picks

Marketplace Samples

Example marketplace items related to this page. Use the search link to explore similar finds on eBay.

Shop location

Using USA USA USAUK

Example eBay listing

Computer Programming Code Funny Science Technology Wall Art Home - POSTER 20x30

Search eBay.com: technology wall art

Browse similar on eBay.com

Example eBay listing

Computer Programming Code Funny Science Technology Print Wall Art - POSTER 20x30

Search eBay.com: technology wall art

Browse similar on eBay.com

Example eBay listing

3pcs Colorful Abstract Painting Of Technology Wall Art Canvas Unframed/Framed

Search eBay.com: technology wall art

Browse similar on eBay.com

Example eBay listing

Ohms Law Poster Electrical Formula Chart Engineering Study Wall Art Decor

Search eBay.com: technology wall art

Browse similar on eBay.com

Browse more on eBay.com

Example items shown for inspiration; availability and pricing can change. Branchoria may earn a commission if you purchase through outbound eBay links.

Example eBay listing

WORLDS MOST MODEST ARTIFICIAL INTELLIGENCE ENGINEER SARCASTIC MUG PERSONALISED

Search eBay.co.uk: artificial intelligence mug

Browse similar on eBay.co.uk

Example eBay listing

Here Sits The Tea Of The Worlds Best Artificial Intelligence Student - Mug an...

Search eBay.co.uk: artificial intelligence mug

Browse similar on eBay.co.uk

Example eBay listing

I fear human stupidity more than artificial intelligence - Black Glossy Mug

Search eBay.co.uk: artificial intelligence mug

Browse similar on eBay.co.uk

Browse more on eBay.co.uk

Example items shown for inspiration; availability and pricing can change. Branchoria may earn a commission if you purchase through outbound eBay links.

Endnotes

1. Source: arxiv.org
   Link: https://arxiv.org/abs/2401.02843
   Source snippet

   arXivThousands of AI Authors on the Future of AIJanuary 5, 2024...

   Published: January 5, 2024

2. Source: arxiv.org
   Link: https://arxiv.org/abs/2502.14870

3. Source: cset.georgetown.edu
   Link: https://cset.georgetown.edu/publication/beyond-pdoom-for-ai-risk-quantifying-uncertainty-without-probability/

4. Source: 80000hours.org
   Title: 80,000 Hours Why experts and forecasters disagree about AI risk
   Link: https://80000hours.org/2024/09/why-experts-and-forecasters-disagree-about-ai-risk/

5. Source: arxiv.org
   Link: https://arxiv.org/html/2502.14870v1

6. Source: cset.georgetown.edu
   Link: https://cset.georgetown.edu/wp-content/uploads/CSET-Beyond-Pdoom-for-AI-Risk.pdf

7. Source: arxiv.org
   Title: arXiv Managing extreme AI risks amid rapid progress
   Link: https://arxiv.org/abs/2310.17688

8. Source: arxiv.org
   Link: https://arxiv.org/abs/2503.07341

9. Source: vox.com
   Link: https://www.vox.com/future-perfect/460222/ai-forecasting-tournament-superforecaster-expert-tetlock

10. Source: arxiv.org
    Link: https://arxiv.org/abs/2401.02843?utm=

11. Source: arxiv.org
    Link: https://arxiv.org/pdf/2503.07341

12. Source: arxiv.org
    Link: https://arxiv.org/pdf/2502.14870

13. Source: vox.com
    Title: Thousands of AI experts are torn about what they’ve
    Link: [https://www.vox.com/future-perfect/2024/1/10/24032987/ai-impacts-survey-artificial

14. Source: 80000hours.org
    Title: yoshua bengio scientist ai
    Link: https://80000hours.org/podcast/episodes/yoshua-bengio-scientist-ai/

15. Source: aiimpacts.org
    Title: Thousands of AI authors on the future of AI
    Link: https://aiimpacts.org/wp-content/uploads/2023/04/Thousands_of_AI_authors_on_the_future_of_AI.pdf

16. Source: aiimpacts.org
    Title: 2022 expert survey on progress in ai
    Link: https://aiimpacts.org/2022-expert-survey-on-progress-in-ai/

17. Source: aiimpacts.org
    Link: https://aiimpacts.org/wp-content/uploads/2022/08/2022ESPAIV.pdf

18. Source: forecastingresearch.org
    Title: roots of disagreement on ai risk
    Link: https://forecastingresearch.org/research/roots-of-disagreement-on-ai-risk

19. Source: theguardian.com
    Link: https://www.theguardian.com/technology/2024/dec/27/godfather-of-ai-raises-odds-of-the-technology-wiping-out-humanity-over-next-30-years

20. Source: ft.com
    Title: Financial Times AI will never threaten humans, says top Meta scientist
    Link: https://www.ft.com/content/30fa44a1-7623-499f-93b0-81e26e22f2a6?syn-25a6b1a6=1

21. Source: facebook.com
    Title: arxiv why do experts disagree on existential risk and pdoom a survey of ai exper
    Link: https://www.facebook.com/ITexam/posts/arxiv-why-do-experts-disagree-on-existential-risk-and-pdoom-a-survey-of-ai-exper/1014848420676515/

22. Source: Wikipedia
    Link: https://en.wikipedia.org/wiki/P%28doom%29

23. Source: thezvi.substack.com
    Title: ai impacts survey december 2023 edition
    Link: https://thezvi.substack.com/p/ai-impacts-survey-december-2023-edition
    Published: december 2023

24. Source: aiimpacts.org
    Link: https://aiimpacts.org/2024/

25. Source: aiimpacts.org
    Title: EMBARGOED AI Impacts Survey Release Google Docs
    Link: https://aiimpacts.org/wp-content/uploads/2024/01/EMBARGOED_-AI-Impacts-Survey-Release-Google-Docs.pdf

26. Source: blog.biocomm.ai
    Link: https://blog.biocomm.ai/2024/02/25/ai-impacts-report-thousands-of-ai-authors-on-the-future-of-ai-38participants-put-at-least-a-10-chance-on-extremely-bad-outcomes-e-g-human-extinction-january-2024/

27. Source: forecastingresearch.org
    Title: near term xpt accuracy
    Link: https://forecastingresearch.org/research/near-term-xpt-accuracy

28. Source: forecastingresearch.org
    Title: ai conditional trees
    Link: https://forecastingresearch.org/ai-conditional-trees

29. Source: jair.org
    Link: https://www.jair.org/index.php/jair/article/view/19087

30. Source: bayesianinvestor.com
    Title: existential risk persuasion tournament
    Link: https://bayesianinvestor.com/blog/index.php/2023/07/17/existential-risk-persuasion-tournament/

Additional References

1. Source: youtube.com
   Link: https://www.youtube.com/watch?v=qn1T4YwJW9o
   Source snippet

   The AI Insider Giving Humanity 50/50 Odds (And What Tips the Scale) | Emad Mostaque...

2. Source: youtube.com
   Title: “We Are the Alien Probes” — Roman Yampolskiy
   Link: https://www.youtube.com/watch?v=sWqE2a5FUcY
   Source snippet

   Doomsday Clock Physicist Warns AI Is Major THREAT to Humanity! — Prof. Daniel Holz, Univ. of Chicago...

3. Source: youtube.com
   Title: Top Economist Says P(Doom) Is 0.1% — Noah Smith vs. Liron Shapira Debate
   Link: https://www.youtube.com/watch?v=AwmJ-OnK2I4
   Source snippet

   "We Are the Alien Probes" — Roman Yampolskiy...

4. Source: youtube.com
   Title: AI: Existential Risk or Power Grab?
   Link: https://www.youtube.com/watch?v=gyOGhnEnGkk
   Source snippet

   Top Economist Says P(Doom) Is 0.1% — Noah Smith vs. Liron Shapira Debate...

5. Source: reddit.com
   Link: https://www.reddit.com/r/Futurology/comments/1hofjfl/godfather_of_ai_says_it_could_drive_humans/

6. Source: researchgate.net
   Link: https://www.researchgate.net/publication/390801072_Pdoom_Versus_AI_Optimism_Attitudes_Toward_Artificial_Intelligence_and_the_Factors_That_Shape_Them

7. Source: researchgate.net
   Link: https://www.researchgate.net/publication/389749013_The_Economics_of_pdoom_Scenarios_of_Existential_Risk_and_Economic_Growth_in_the_Age_of_Transformative_AI

8. Source: medium.com
   Link: https://medium.com/frontiers-of-data-science/ai-risk-perspectives-from-superforecasters-and-ai-experts-a9ee684eff89

9. Source: linkedin.com
   Link: https://www.linkedin.com/posts/simontorrance_why-do-experts-disagree-on-existential-risk-activity-7301886948334333952-VwQS

10. Source: facebook.com
    Link: https://www.facebook.com/yann.lecun/posts/a-sensible-piece-by-nello-cristianini-about-ai-existential-riskor-lack-thereofif/10158942941237143/