Talk:P(doom)

I came up with the "p(doom)" term - in 2009

As for "originating as an inside joke among AI researchers" - I came up with the term - as far as anyone can tell. I posted about it many times on LessWrong - which still has the posts archived. I was not an AI researcher. For the story, see the Roose, Kevin (2023-12-06) NYT article already linked as a reference in the article. TylerTim (talk) 02:54, 11 January 2025 (UTC)

Here's a thread where I tell my story on Twitter/X: https://x.com/tim_tyler/status/1665571547111649286 TylerTim (talk) 02:56, 11 January 2025 (UTC)

Here's a link to a relevant page on my <cough> machine intelligence web site: https://matchingpennies.com/pdoom/ TylerTim (talk) 03:00, 11 January 2025 (UTC)

Yup, seems correct. LessWrong search returns your comments as being 8 years ago, I reckon that's due to the revival code having it indexed the comments then. Not sure that WP citing guidelines can point to LW comments since they're not WP:RS, but the original article mentions you. Niplav (talk) 15:33, 13 October 2025 (UTC)

Thanks. I am not sure exactly when it happened, but I now appear to be recognized as the first user of the term in the main Wikipedia article on the topic. It mentions 2009 - which is when I first used the term in public. Yay! TylerTim (talk) 23:17, 30 October 2026 (UTC)

Nate Silver's p(doom) is 5-10%

Nate Silver estimates p(doom) as 5-10% here: https://www.natesilver.net/p/its-time-to-come-to-grips-with-ai

Perhaps a regular contributor can add this to the list of prominent figure's p(doom) estimates. 71.161.118.179 (talk) 02:30, 28 January 2025 (UTC)

Geoffrey Hinton

In a recent interview with CBS he stated that he agrees with Elon Musk in that there's a 10-20% chance.

https://www.youtube.com/watch?v=mvAeKSBTD8c?t=138 (at 2:18 if the timestanp doesn't work) — Preceding unsigned comment added by 80.5.201.152 (talk) 11:06, 25 May 2025 (UTC)

No, Hinton's personal probability was somewhere above 50%.

He estimated that the average probability of AI researchers was somewhere in the realm of 10%. FavourNSpice (talk) 23:32, 5 June 2025 (UTC)

Possible addition to popular culture

Hi all, I noticed the article doesn’t currently include fictional or cultural portrayals of the P(doom) concept. Would it be appropriate to mention the 2025 speculative novel P(doom): Scales of Survival by Marco Sors? The book explores a world after an AI-driven takeover where humanity undergoes extreme adaptive transformations due to technological and environmental collapse, echoing themes of AI alignment failure and existential risk.

Not sure it meets notability thresholds yet, but I wanted to raise it for consideration. Thanks! 2001:B07:A9A:4E3A:5978:D976:7D1A:6153 (talk) 17:29, 24 June 2025 (UTC)

“>0%” = “50%”?

I know this is kinda splitting hairs but I think saying “more than zero percent” is justifiably NOT sorted under a 50% sort value. I think there's real cause to assert that when someone says “more than zero percent” they have a weight curve that WAY more heavily weights 1% than 99%.

I think Hassibis should be sorted at least under something like 25%, instead of 50%.

2600:1702:6B11:7C40:CC77:6B39:A54B:4E38 (talk) 15:45, 31 July 2025 (UTC)

Sutton, Crowley

I don't think Sutton should be in this table. As far as I know, he hasn't explicitly given a P(doom). The "0%" appears to be inferred from his belief that AI replacing humans wouldn't be that bad. But even if we accept this framing and don't count human extinction as "doom", there may still be other kinds of doom that he considers plausible. This entry looks like original synthesis (WP:SYNTH).

Paul Crowley may also not be sufficiently notable to be listed in this table. Alenoach (talk) 13:35, 18 October 2025 (UTC)

Possible expansion of public p(doom) estimates

The article currently mentions a number of individual p(doom) estimates from prominent AI researchers and commentators.

Would it be useful to add a sourced overview of publicly stated p(doom) estimates from notable figures such as Geoffrey Hinton, Eliezer Yudkowsky, Paul Christiano, Yoshua Bengio, Yann LeCun, and others?

A structured summary might help readers understand the range of views while keeping the article neutral and well sourced.

I'm not proposing any particular source here, only wondering whether such a section would improve the article. ~2026-33114-64 (talk) 10:11, 4 June 2026 (UTC)

All of the people you named and more are included in the table already. Is what you are suggesting basically a prose summary of the table? Or, how are you imagining it? -- irn (talk) 01:10, 7 June 2026 (UTC)

Adding dates to pdoom estimates

I think we should add a column listing when the estimate was given. Things are moving fast and it would be helpful to know whether the estimate might be relatively outdated, instead of comparing each row as apples to apples. I know one can click the citation to see the date, but I think it still helps seeing the date more easily in a column Brangston (talk) 15:46, 20 September 2026 (UTC)

Having an image column would be nice too ~2026-51144-01 (talk) 00:42, 23 September 2026 (UTC)

I added some dates, and someone else added some images before me. ISaveNewspapers (talk) 02:52, 21 October 2026 (UTC)

Really appreciate it, great job! Brangston (talk) 20:50, 21 October 2026 (UTC)

Specify what each person means by "P(doom)" + focus on recent estimates

In the table of notable P(doom) values, we must make it clear exactly what each person is actually estimating the probability of. I consider this a high priority for this article. The term "P(doom)" is concerningly vague in multiple dimensions, including time frame, outcomes, and conditions. For more on this, see the Ambiguity section of "Stop talking about P(doom)" by Isaac King, as cited in this article's Criticisms section.

Wikipedia will not stop talking about P(doom): our goal is to give information on notable topics covered in independent, reliable sources. What we can do is expand on that information by specifying what each person means by "P(doom)". By way of footnotes in the P(doom) column, this is currently partially achieved. However, of the 38 entries, only 13 contain a footnote; of those, only 8 contain any degree of specification; and only 3 seem to contain a full specification. These are my impressions from a quick skim, so don't take it literally.

There are a few options for adding specification. One option is to add footnotes to the entries lacking them, and edit the existing footnotes. However, for ease of reading, I suggest taking the info out of the footnotes. Either keep it in the P(doom) column, or move it to at least one new column. For some entries, the relevant person has not given a full specification, so just add as much info as you can.

I am adding this topic as a request for both help and feedback. Meanwhile, I will likely keep working on the article myself.

On another note: Due to the fast pace of AI development, I suggest a focus on recent P(doom) estimates--say, within the past couple of months--when adding to the table. I also suggest looking at some of the people currently in the table and researching whether they have recently updated their estimate. The bulk of entries are currently from years ago. ISaveNewspapers (talk) 05:02, 21 October 2026 (UTC)

Scott Aaronson's p(doom) may be out of date

In a recent video he said his p(doom) was between 5 and 50%. It was in an interview with doom expert Roman. The article here says 2%.