Return-Path: Delivered-To: alvarochf@traceseis.com Received: from cs510.bluehost.com by cs510.bluehost.com with LMTP id 0NTBKiHBoWlhBwMAWiBQxw (envelope-from ) for ; Fri, 27 Feb 2026 09:06:57 -0700 Return-path: Envelope-to: alvarochf@traceseis.com Delivery-date: Fri, 27 Feb 2026 09:06:57 -0700 Received: from mg-47-244.substack.com ([69.72.47.244]:39676) by cs510.bluehost.com with esmtps (TLS1.2) tls TLS_ECDHE_RSA_WITH_AES_128_GCM_SHA256 (Exim 4.98.2) (envelope-from ) id 1vw0Mj-0000000178S-1bqt for alvarochf@traceseis.com; Fri, 27 Feb 2026 09:06:57 -0700 DKIM-Signature: a=rsa-sha256; v=1; c=relaxed/relaxed; d=mg-d1.substack.com; q=dns/txt; s=k1; t=1772208400; x=1772215600; h=List-Unsubscribe-Post: List-Unsubscribe: List-Post: List-Id: List-Archive: List-Owner: Reply-To: In-Reply-To: References: Sender: Sender: Date: Message-Id: To: To: From: From: Subject: Subject: Content-Type: Mime-Version; bh=DoFaEXJRWO2N58/nr6AJbOi3hHpr7t8iVli+AY4FX5M=; b=iqvqRztdX7xQkvIMh9HiZNfKeLKp1jTee5ev54cryQBn66xHlhLo50k+EU96tc9SeRVPv7KJnP7gMM4LMD5fFOFuz/XxQfa24lr80eTxMTRjUHok5bJXqTowkpSBMVcO9qtRhwV9PUawTQDipCCbKd0sPuTSG7rBUvI/Ep53j5U= X-Mailgun-Sid: WyJiYTg1MiIsImFsdmFyb2NoZkB0cmFjZXNlaXMuY29tIiwiMzliMDhjNiJd Received: by ccdfec28c39df86474ef055c5adfcfd9827b625588902b8e4b6d995475adc8c6 with HTTP id 69a1c10da69647e41776d241; Fri, 27 Feb 2026 16:06:36 GMT X-Mailgun-Sending-Ip: 69.72.47.244 X-Mailgun-Batch-Id: 69a1c10df6fd4a8bd906f095 Mime-Version: 1.0 Content-Type: multipart/alternative; boundary="c48b76fee43dc8f2138f38d470ef9a1fcfe44e0c02733df6aaaad835a810" Subject: Not Even Wrong From: Cosmos Institute To: alvarochf@traceseis.com X-Mailgun-Tag: post X-Mailgun-Track-Clicks: false Message-Id: <20260227160232.3.f59a9f4f313c4a87@mg-d1.substack.com> Date: Fri, 27 Feb 2026 16:02:32 +0000 Feedback-Id: post-189305718:cat-post:pub-2225794:substack Sender: Cosmos Institute References: In-Reply-To: Reply-To: Cosmos Institute List-Owner: List-Url: List-Archive: List-Id: List-Post: List-Unsubscribe: List-Unsubscribe-Post: List-Unsubscribe=One-Click X-Mailgun-Variables: {"category": "post", "email_generated_at": "1772208396675", "is_freemail": "true", "post_audience": "everyone", "post_id": "189305718", "post_type": "newsletter", "pub_community_enabled": "true", "publication_id": "2225794", "subdomain": "cosmosinstitute", "user_id": "347258985"} --c48b76fee43dc8f2138f38d470ef9a1fcfe44e0c02733df6aaaad835a810 Content-Type: text/plain; charset="utf-8" Content-Transfer-Encoding: quoted-printable View this post on the web at https://blog.cosmos-institute.org/p/not-even-w= rong John Graunt was a merchant who lived in 17th century London. In 1662, in a = city familiar with plague, Graunt consulted parish death records to figure = out how long Londoners could expect to live. He could tell you that roughly= 36 out of every 100 children born in the city would die before their sixth= birthday. This probability could be checked by anyone willing to do the sa= me sums using the same sources. Three and a half centuries later, AI researchers are calculating a morbid p= robability of their own. They wonder about thinking machines seeing off the= human race, and refer to the likelihood of that possibility as P(doom). Th= e term began life as an inside joke in rationalist spaces in the closing ye= ars of the 2000s, but shot to prominence after researchers signed an open l= etter [ https://substack.com/redirect/48f213dc-8c16-447d-b321-6aa292a55c8c?= j=3DeyJ1IjoiNXFxeXF4In0.h9dEDbGj64cTNudluPwbMYWklSu8UKGoFv6eWS-HyoA ] about= the threat posed by AI development in 2023. In the AI world, everyone has = their number. Yudkowsky puts it at roughly 95=E2=80=93100 percent. Geoffrey= Hinton sits around 10=E2=80=9320 percent, while Marc Andreessen reports a = bullish 0 percent (a logical impossibility because you cannot rule out some= thing that has never happened and whose conditions have never been tested).= The mean estimate among surveyed AI researchers [ https://substack.com/red= irect/16a60180-a0f4-437f-a26c-dc276bd69b9d?j=3DeyJ1IjoiNXFxeXF4In0.h9dEDbGj= 64cTNudluPwbMYWklSu8UKGoFv6eWS-HyoA ] is about 14 per cent. We have some who believe that AI is almost certain to wipe out humanity and= others who ascribe a 0 per cent probability to the same outcome. This is c= urious for a supposedly hard-headed prediction made by some of the world=E2= =80=99s most credentialed scientists. Imagine a roomful of cardiologists wh= o, given the same scans and the same patient history, exposed to the same t= raining and the same clinical standards, disagreed on whether the probabili= ty of heart failure was 0.1 percent or 99 percent. You would not conclude that this was a hard problem on which reasonable peo= ple differed. In fact, you would probably conclude that at least some of ou= r medics may need to visit neurology down the hall. With P(doom), the estim= ates from AI researchers range from 0 percent to 99 percent (and a few nine= s then some). When someone tells you they assign a 15 per cent probability = to AI wiping out humanity, it is worth asking what kind of claim they are a= ctually making. It sounds like a statement about the world. It is not. To be clear: my objection is not to probabilistic reasoning, which is strai= ghtforwardly indispensable in domains where estimates can be calibrated aga= inst outcomes over time. My problem concerns what we might call proximate a= nd ultimate formulations. Proximate estimates for COVID were things like th= e hospitalization rate and case severity, while an ultimate estimate was th= e infection fatality rate. For P(doom), proximate estimates include the con= ditions under which models exhibit deceptive behavior and the observed fail= ure rates of current alignment techniques. Here, the ultimate estimate is t= he thing itself. Someone might know that current models can be made to exhibit deceptive ali= gnment under laboratory conditions or that existing alignment techniques ha= ve specific failure modes. But to cram those findings into a container labe= led =E2=80=9C15% chance of extinction=E2=80=9D involves a series of judgmen= ts about unprecedented transitions for which there is no model connecting p= roximate estimates to their ultimate counterpart. Knowing that fine-tuning = on specific tasks can induce broader misalignment tells us little about how= well alignment solutions will generalize or how geopolitical actors will r= espond. Compare with COVID, where hospitalization and case-severity data fe= d epidemiological models that could be tested against actual deaths. A natural objection is that some people are good at formulating predictions= in lots of different domains, so perhaps their assessment of P(doom) ought= to carry weight. After all, some superforecasters assign 17 per cent to va= rious events and see those events happen roughly 17 per cent of the time. Y= ou can check a forecaster=E2=80=99s guesses across many different predictio= ns in such a way as to mean we=E2=80=99re dealing with a property of the fo= recaster=E2=80=99s overall policy rather than of any individual estimate. W= hen a superforecaster says 17 per cent for a ceasefire in some conflict, th= at number is useful because they also say 17 per cent for hundreds of other= things and roughly 17 per cent of those things happen. Superforecasters earn their calibration (that is, the extent to which confi= dence lines up with their track record over time) using predictions with sh= ort time horizons, a history of similar events to draw on, and a mechanism = that allows the forecaster to correct course when needed. For P(doom), ther= e is no source of comparable predictions against which to calibrate. A fore= caster might be superbly calibrated on elections and geopolitical high dram= a, but that tells us nothing about whether their number for an unprecedente= d event is any good. What I Talk About When I Talk About P(doom) There are two influential interpretations of probability: subjective and ob= jective. Where the former deals with how confident you are in a claim, the = latter stresses what actually happens when you test it. Anyone who has spen= t even a little time thinking about the reliability of P(doom) metrics prob= ably has a sense that the calculations ought to be taken with a pinch of sa= lt. Many in rationalist or rationalist-adjacent circles are well aware of t= he difference between these two accounts, but lots of people who absorb the= se views are not. That goes for those who adopt the terminology to signal their membership of= the in-group, and those who hear talk of P(doom) on the news and wonder wh= y one of the world=E2=80=99s most cited researchers thinks there=E2=80=99s = a one in five chance that AI will wipe out the human race. When a chief glo= bal strategist at one of the largest investment research firms tells CNBC [= https://substack.com/redirect/d4f4eb86-cad5-4e3a-9cfa-c39880df5906?j=3DeyJ= 1IjoiNXFxeXF4In0.h9dEDbGj64cTNudluPwbMYWklSu8UKGoFv6eWS-HyoA ] it=E2=80=99s= 50/50 whether AI destroys humanity by mid-century, the average viewer =E2= =80=94 incorrectly but understandably =E2=80=94 assumes that figure estimat= es the likelihood that it will happen. Every single P(doom) you hear is a subjective probability. It is a measure = of the degree of rational belief one holds in a proposition given the avail= able evidence =E2=80=94 a property of an agent=E2=80=99s epistemic state ra= ther than of the world. When you say =E2=80=9CI think there=E2=80=99s a 50 = per cent chance this meeting will be a waste of time,=E2=80=9D you=E2=80=99= re not drawing on a frequency table of past meetings. You=E2=80=99re expres= sing how confident you feel given what you know about the agenda and who=E2= =80=99s attending. All the framework requires is that your degrees of belie= f are consistent with one another. Beyond this internal consistency, there = is no further requirement that your probabilities correspond to the shape o= f reality. If your beliefs are internally coherent, you can multiply your probability = for any outcome by the value of that outcome =E2=80=94 however you choose t= o measure it =E2=80=94 and arrive at its expected value, a single number th= at tells you how good or bad a bet looks on average. From there you can der= ive a utility function and compare any set of actions on a common scale. Th= e framework of cause prioritization, wherein resources are allocated to whi= chever cause produces the greatest expected good per dollar, inherits this = apparatus. This approach is popular with rationalist-adjacent communities, = especially Effective Altruism, because it transforms [ https://substack.com= /redirect/97efe852-2b71-4588-9b29-5ad32e171e0b?j=3DeyJ1IjoiNXFxeXF4In0.h9dE= DbGj64cTNudluPwbMYWklSu8UKGoFv6eWS-HyoA ] questions of moral life into prop= erties that can be measured and optimized. Objective accounts of probability take a different approach. Here probabili= ty is what actually happens in the world when you repeat something many tim= es. This species of probability is a physical property of systems, one that= is empirically testable. If you flip a coin a thousand times roughly half = will land heads, so we can say the probability is that ratio. The same logi= c applies to complex phenomena, like radioactive decay, where about half th= e atoms in a given sample will decay within the half-life period (and you c= an go and check). The subjective interpretation is internally coherent and mathematically ele= gant. But coherence should not be confused for empirical content. My object= ion to P(doom) flows from the simple fact that subjective probability claim= s are, by their nature, unfalsifiable. That=E2=80=99s generally fine in man= y instances =E2=80=94 like casual everyday judgments or betting on the Supe= r Bowl =E2=80=94 but not when the stakes are high enough to redirect billio= ns of dollars and make or break government policy. Statistical reasoning ab= out unlikely but potentially devastating scenarios is obviously useful, but= only when properly grounded in claims we can falsify. Your Problem Too For a probability to tell you about something other than the speaker=E2=80= =99s state of mind, it needs a collection of similar events from which a ra= tio can be drawn. To say that a coin lands heads 50 percent of the time, we= need a set of coin flips to draw from. This is our reference class. Clearly there is no reference class for human extinction. As far as every A= I researcher in the world knows, the event is singular and unprecedented. T= raditionally, the reference class problem has been associated with the obje= ctive interpretation of probability. This is because objective probability = is calculated according to the ratio of how many times something happened o= ut of how many times it might have happened. If the event has never happene= d and can=E2=80=99t be repeated, then there=E2=80=99s nothing to compute. But the reference class problem isn=E2=80=99t only relevant for objective a= ccounts of probability. Alan H=C3=A1jek famously made the case [ https://su= bstack.com/redirect/944214d7-baca-4fa8-861b-98f3805fdfdf?j=3DeyJ1IjoiNXFxeX= F4In0.h9dEDbGj64cTNudluPwbMYWklSu8UKGoFv6eWS-HyoA ] that it affects every i= nterpretation of probability when applied to singular events. As he put it,= =E2=80=9Cthe reference class problem is your problem too.=E2=80=9D The arg= ument goes something like this: before you can assign a probability to a on= e-off event, you have to decide what kind of event it is. Is AI-caused exti= nction a case of =E2=80=9Cnew technology going wrong,=E2=80=9D =E2=80=9Cspe= cies-level catastrophe,=E2=80=9D or =E2=80=9Cunprecedented transition in th= e nature of intelligence=E2=80=9D? Each framing suggests a different probab= ility, and there is no principled way to choose between them. The number yo= u arrive at depends on how you describe your target. The way to tell a useful probability from a useless one is to ask whether t= he world can correct it. This is because good explanations resist variation. If Graunt had doubled h= is estimate of childhood mortality, the parish registers would have correct= ed him. If an epidemiologist in early 2020 had put the infection fatality r= ate at 10 per cent rather than 1 per cent, the incoming hospital data would= have shown the estimate to be wrong. The explanation is enmeshed with the = world in such a way that it cannot be led to wherever you would like it to = go. With P(doom), you can swap the causal pathway from deceptive alignment = to resource competition and inflate the number from 5 per cent to 7.5 per c= ent. The estimate can contort to accommodate any figure and any causal chai= n because the only constraint it faces is internal coherence.=20 In the context of P(doom), defenders might say something like =E2=80=9CWe d= on=E2=80=99t need a reference class because we=E2=80=99re expressing a degr= ee of personal belief.=E2=80=9D This is perfectly coherent on the subjectiv= e view, but it is also the move that severs the claim from its empirical co= ntent. If your probability doesn=E2=80=99t need to correspond to any featur= e of the world, then it can=E2=80=99t be wrong about the world. And a claim= that can=E2=80=99t be wrong about the world tells you nothing about it. We might call the result a Gettier probability, a credence that is internal= ly justified =E2=80=94 and that might even correspond to reality =E2=80=94 = but that lines up with the world through luck rather than judgement. In epi= stemology, a Gettier case is a belief that is true and justified but not tr= ue because of its justification. Any P(doom), where the justification and t= he truth are aligned, will (assuming the estimate was correct) have the sam= e structure. A researcher=E2=80=99s estimate might happen to match whatever= the actual risk turns out to be, but the match would be accidental relativ= e to the method. Gettier probabilities show up everywhere, like when a consulting firm estim= ates that a given market will grow at 7.3 per cent and lo and behold they a= re correct. The same is true for a geopolitical risk score that looks preci= se but cannot be updated by any observation short of the catastrophe it pur= ports to predict. P(doom) is distinctive only in that the stakes are high e= nough that the probability escapes containment and ends up on the morning n= ews. Suppose someone=E2=80=99s P(doom) is 30 percent. What outcome would show th= is was wrong? If AI goes well, they get to enjoy the benefits of the 70 per= cent figure. If it goes catastrophically =E2=80=94 and anyone is still aliv= e to update their priors =E2=80=94 the 30 percent gets paid off. This is on= e of the more frustrating features of the subjective framework, which allow= s pretty much anyone with an idea about AI risk to have their tokens and ea= t them. Bayesian probability demands that credences be precise. When estimates rang= e from 0% to 99% with no mechanism to adjudicate, the framework is not bein= g applied so much as invoked. P(doom) manages the impressive trick of satis= fying neither the Bayesian standard (precise credences that can be adjudica= ted between) nor the falsifiability standard (empirical claims that can be = proved wrong). I=E2=80=99m not arguing that we ought to stop reasoning about novel events.= Subjective probability is useful because it helps us think under uncertain= ty without good data. Some of the best calls in recent AI history were made= [ https://substack.com/redirect/9b02c981-ffa9-4b2c-8c10-5c182d809c37?j=3De= yJ1IjoiNXFxeXF4In0.h9dEDbGj64cTNudluPwbMYWklSu8UKGoFv6eWS-HyoA ] by researc= hers who formed strong priors about scaling laws and bet on them before the= evidence was in. Those bets were vindicated because they were testable. Sc= aling laws either held or they didn=E2=80=99t; capability thresholds either= arrived on schedule or did not. My concern is that P(doom) borrows the con= fidence of claims like these without sharing their vulnerability to refutat= ion. I Am Sure You Are Very Sure Forget, for a moment, whether P(doom) can be tested. Consider instead wheth= er it is the kind of thing that can be meaningfully quantified in the first= place. In 1921, the economist Frank Knight drew a distinction between risk= and uncertainty. Risk is what you face when you can calculate the odds as = with flicking a small ball onto a roulette wheel. Uncertainty is what you f= ace when you can=E2=80=99t, even in principle, because the situation is too= poorly understood to yield a number. Knight argued that the two are different in kind. The Bayesian tradition=E2=80=99s response to Knight was to argue that subje= ctive probability does away with the distinction: if you can always express= a degree of belief, then there is no situation that resists quantification= =2E P(doom) is a textbook application o= f this move, but the dissolution works= only if the resulting number can be tested via real-world feedback. For P(= doom), the Bayesian move converts Knightian uncertainty into a figure that = has the form of risk without any of the mechanisms that make risk estimates= trustworthy. The same process, where uncertainty becomes risk, happens wherever a decisi= on-making framework demands a numerical input and none is available. Firms = routinely assign precise probabilities to market scenarios that are genuine= ly unprecedented. They don=E2=80=99t have any evidence that warrants a spec= ific number, but they do have a spreadsheet with cells that need filling. T= he defense that such figures are =E2=80=9Crough heuristics=E2=80=9D does no= t help. The problem is that they are treated simultaneously as a casual sho= rthand and as inputs into value calculations depending on whatever is most = convenient. This is why the =E2=80=9Cshmrobability [ https://substack.com/r= edirect/547fbc54-0adb-4e35-a0e9-f7776d041932?j=3DeyJ1IjoiNXFxeXF4In0.h9dEDb= Gj64cTNudluPwbMYWklSu8UKGoFv6eWS-HyoA ]=E2=80=9D move doesn=E2=80=99t fix o= ur problem. It concedes we=E2=80=99re not really talking about probability,= while trying to preserve its practical content. Either the number is a rou= gh heuristic and we should stop plugging it into expected value calculation= s, or it is a serious input to policy and we should hold it to the standard= s that implies. If P(doom) is a subjective credence, there is no principled way to adjudica= te between two people who disagree. A P(doom) of 95 per cent and one of 50 = per cent are both internally coherent insofar as neither violates any rule = of the subjective framework. You can be confident in your P(doom) and I can= be confident in mine, and there is no observation or outcome that could se= ttle which of us is right. Attempts to resolve this from within the Bayesia= n model, from Aumann=E2=80=99s agreement theorem [ https://substack.com/red= irect/1286ee77-1cdf-4a4c-9dcf-fe7bfb5b0dd6?j=3DeyJ1IjoiNXFxeXF4In0.h9dEDbGj= 64cTNudluPwbMYWklSu8UKGoFv6eWS-HyoA ] onwards, all founder on conditions th= at do not hold (like shared starting points, perfect rationality, or common= knowledge of each other=E2=80=99s beliefs). The problem compounds when these estimates enter expected value calculation= s, where even a vanishingly small probability multiplied by the stakes of h= uman extinction produces numbers large enough to dominate decisions. The ra= tionalist and rationalist-adjacent communities recognize this dynamic as Pa= scal=E2=80=99s mugging, while groups like GiveWell have cautioned [ https:/= /substack.com/redirect/338fa0bc-a4f8-4bdb-bbd1-ecee8e2ac238?j=3DeyJ1IjoiNXF= xeXF4In0.h9dEDbGj64cTNudluPwbMYWklSu8UKGoFv6eWS-HyoA ] against taking such = estimates at face value. The reader should not take from this essay that AI risk is unworthy of atte= ntion. Rather, my point is that P(doom) is the wrong epistemic instrument f= or thinking about it. The alternative is to formulate falsifiable conjectur= es about the world and then try to refute them, which is exactly what AI sa= fety researchers do in their actual work. They propose that particular trai= ning regimes will produce deceptive behavior and they predict that models b= eyond a certain capability threshold will resist correction. They hypothesi= ze that misalignment in one domain will generalize unpredictably to others,= and they can test this by measuring the transfer of learned strategies acr= oss task distributions. They ask whether models that pass safety evaluation= s in deployment conditions will behave differently when those conditions ch= ange. Each of these lines of research produces claims that can fail, and th= is vulnerability is what makes them worth paying attention to. Individual claims about deceptive behavior or capability thresholds do not = by themselves tell you how much to spend on alignment versus pandemic prepa= redness, yet real institutions have treated P(doom) as though it could. The= effective altruist community has directed hundreds of millions of dollars = toward AI existential risk reduction. Open Philanthropy, the field=E2=80=99= s largest funder, used subjective probability frameworks to weigh AI risk a= gainst global health, biosecurity, and farm animal welfare. Whether or not = that assessment [ https://substack.com/redirect/11585253-5e61-46d4-bf12-321= 23ac2d6dc?j=3DeyJ1IjoiNXFxeXF4In0.h9dEDbGj64cTNudluPwbMYWklSu8UKGoFv6eWS-Hy= oA ] was correct, the reasoning behind it is only as good as the numbers it= relies on.=20 A figure that ranges from 0 per cent to 99 per cent depending on who you as= k and with no way to adjudicate between them is about as useful as no numbe= r at all. Aggregating these estimates might seem to help inasmuch as a mean= of 14 per cent looks more sober than bouncing between extremes. Aggregate = or individual, the problem is that neither incur a cost for being wrong. A = bookmaker=E2=80=99s odds are constrained by the market just as an insurer= =E2=80=99s premiums are shaped by the frequency and nature of claims made. = In both cases, the issuer=E2=80=99s assessment improves over time because t= hey suffer when it is incorrect. There is no market to punish a mispriced P= (doom) and no settlement date on which the estimate is tested. We might say something like: =E2=80=9CEven if P(doom) is imprecise, it=E2= =80=99s still better than no number because we need to allocate resources s= omehow.=E2=80=9D But here we=E2=80=99re assuming that internal coherence st= ill outperforms the absence of a number when you have to act. We can live w= ith imprecision under these conditions, but not at the expense of claims th= at cannot be falsified that get steamrolled by the headline rating. Governments fund pandemic preparedness efforts without a big round number f= or the probability of the next pandemic. Institutions allocate resources un= der genuine uncertainty all the time by doing things like funding a portfol= io of approaches, setting thresholds for observable harms, identifying the = cheapest reversible interventions, demanding observable milestones before s= caling commitment, and building the capacity to pivot as new evidence arriv= es. The physicist Wolfgang Pauli was famously unforgiving of bad theory. When a= colleague asked him to assess a young physicist=E2=80=99s paper, his verdi= ct was that it was =E2=80=9Cnot even wrong.=E2=80=9D He meant that its cent= ral claim was so removed from evidence that it could not be proved false. P= (doom) is not even wrong. Graunt=E2=80=99s mortality tables may have been t= echnically crude, but they were accountable. They could be checked against = next year=E2=80=99s parish records, and if they were wrong, the records wou= ld show it. Unsubscribe https://substack.com/redirect/2/eyJlIjoiaHR0cHM6Ly9ibG9nLmNvc21= vcy1pbnN0aXR1dGUub3JnL2FjdGlvbi9kaXNhYmxlX2VtYWlsP3Rva2VuPWV5SjFjMlZ5WDJsa0= lqb3pORGN5TlRnNU9EVXNJbkJ2YzNSZmFXUWlPakU0T1RNd05UY3hPQ3dpYVdGMElqb3hOemN5T= WpBNE16azJMQ0psZUhBaU9qRTRNRE0zTkRRek9UWXNJbWx6Y3lJNkluQjFZaTB5TWpJMU56azBJ= aXdpYzNWaUlqb2laR2x6WVdKc1pWOWxiV0ZwYkNKOS5TQW83amJzQ1lMT1VseXJlRmhsU2xiU3V= NSkhzSXVXanNhdWE4NnF0S2p3IiwicCI6MTg5MzA1NzE4LCJzIjoyMjI1Nzk0LCJmIjp0cnVlLC= J1IjozNDcyNTg5ODUsImlhdCI6MTc3MjIwODM5NiwiZXhwIjoyMDg3Nzg0Mzk2LCJpc3MiOiJwd= WItMCIsInN1YiI6ImxpbmstcmVkaXJlY3QifQ.eWBjde71pAGkTAgnhMjGO7bTc8WIU2oPFDfvz= 4Gt2_A? --c48b76fee43dc8f2138f38d470ef9a1fcfe44e0c02733df6aaaad835a810 Content-Type: text/html; charset="utf-8" Content-Transfer-Encoding: quoted-printable Not Even Wrong @media (max-width: 1024px) { .typography .pullquote-align-left, .typography.editor .pullquote-align-left, .typography .pullquote-align-right, .typography.editor .pullquote-align-right, .typography .pullquote-align-wide, .typography.editor .pullquote-align-wide, .typography .pullquote-align-center, .typography.editor .pullquote-align-center { float: none; margin: 0 auto; width: 100%; max-width: 100%; } } @media all and (-ms-high-contrast: none), (-ms-high-contrast: active) { .typography .markup table.image-wrapper img, .typography.editor .markup table.image-wrapper img, .typography .markup table.kindle-wrapper img, .typography.editor .markup table.kindle-wrapper img { max-width: 550px; } } @media (min-width: 1024px) { .typography:not(:has(#toc)) .captioned-image-container figure:has(> a.ima= ge2-offset-left), .typography.editor:not(:has(#toc)) .captioned-image-container figure:has(= > a.image2-offset-left) { margin-left: var(--image-offset-margin); } .typography:not(:has(#toc)) .captioned-image-container figure:has(> a.ima= ge2-offset-right), .typography.editor:not(:has(#toc)) .captioned-image-container figure:has(= > a.image2-offset-right) { margin-right: var(--image-offset-margin); } } @media (min-width: 1300px) { .typography .captioned-image-container figure:has(> a.image2-offset-left)= , .typography.editor .captioned-image-container figure:has(> a.image2-offse= t-left) { margin-left: var(--image-offset-margin); } .typography .captioned-image-container figure:has(> a.image2-offset-right= ), .typography.editor .captioned-image-container figure:has(> a.image2-offse= t-right) { margin-right: var(--image-offset-margin); } } @media (max-width: 1024px) { .typography, .typography.editor { /* Disable offset on mobile/tablet */ } .typography .captioned-image-container figure:has(> a.image2-align-left), .typography.editor .captioned-image-container figure:has(> a.image2-align= -left), .typography .captioned-image-container figure:has(> a.image2-align-right)= , .typography.editor .captioned-image-container figure:has(> a.image2-align= -right) { float: none; margin: 1em auto; max-width: 100%; width: auto; padding: 0; } .typography .captioned-image-container figure:has(> a.image2-align-left.t= hefp), .typography.editor .captioned-image-container figure:has(> a.image2-align= -left.thefp), .typography .captioned-image-container figure:has(> a.image2-align-right.= thefp), .typography.editor .captioned-image-container figure:has(> a.image2-align= -right.thefp) { margin: 1em auto; } .typography .captioned-image-container figure:has(> a.image2-offset-left)= , .typography.editor .captioned-image-container figure:has(> a.image2-offse= t-left), .typography .captioned-image-container figure:has(> a.image2-offset-right= ), .typography.editor .captioned-image-container figure:has(> a.image2-offse= t-right) { margin: 1em auto; } .typography .captioned-image-container figure:has(> a.image2-align-left) = =2Eimage2-inset, .typography.editor .captioned-image-container figure:has(> a.image2-align= -left) .image2-inset, .typography .captioned-image-container figure:has(> a.image2-align-right)= .image2-inset, .typography.editor .captioned-image-container figure:has(> a.image2-align= -right) .image2-inset { display: block; justify-content: initial; } } @media (max-width: 768px) { .typography .markup div.sponsorship-campaign-embed, .typography.editor .markup div.sponsorship-campaign-embed { margin-top: 24px; margin-bottom: 24px; } .typography .markup div.sponsorship-campaign-embed:first-child, .typography.editor .markup div.sponsorship-campaign-embed:first-child { margin-top: 0px; } } @media screen and (max-width: 650px) { .typography .markup div.youtube-overlay, .typography.editor .markup div.youtube-overlay, .typography .markup div.vimeo-overlay, .typography.editor .markup div.vimeo-overlay { display: none !important; } } @media screen and (max-width: 370px) { .typography .markup div.tiktok-wrap, .typography.editor .markup div.tiktok-wrap { width: calc(95vw - 32px); height: calc((95vw - 32px - 2px) / 0.485714); } } @media screen and (max-width: 650px) { .typography .markup div.embedded-publication-wrap .embedded-publication.s= how-subscribe, .typography.editor .markup div.embedded-publication-wrap .embedded-public= ation.show-subscribe { padding: 24px; } } @media screen and (max-width: 650px) { .typography .markup div.subscription-widget-wrap .subscription-widget.sho= w-subscribe, .typography.editor .markup div.subscription-widget-wrap .subscription-wid= get.show-subscribe, .typography .markup div.subscription-widget-wrap-editor .subscription-wid= get.show-subscribe, .typography.editor .markup div.subscription-widget-wrap-editor .subscript= ion-widget.show-subscribe, .typography .markup div.captioned-button-wrap .subscription-widget.show-s= ubscribe, .typography.editor .markup div.captioned-button-wrap .subscription-widget= =2Eshow-subscribe { padding: 0px 24px; } } @media screen and (max-width: 650px) { .typography .markup div.subscription-widget-wrap .subscription-widget.sho= w-subscribe .subscription-widget-subscribe .button, .typography.editor .markup div.subscription-widget-wrap .subscription-wid= get.show-subscribe .subscription-widget-subscribe .button, .typography .markup div.subscription-widget-wrap-editor .subscription-wid= get.show-subscribe .subscription-widget-subscribe .button, .typography.editor .markup div.subscription-widget-wrap-editor .subscript= ion-widget.show-subscribe .subscription-widget-subscribe .button, .typography .markup div.captioned-button-wrap .subscription-widget.show-s= ubscribe .subscription-widget-subscribe .button, .typography.editor .markup div.captioned-button-wrap .subscription-widget= =2Eshow-subscribe .subscription-widget-subscribe .button { padding: 10px 12px; min-width: 110px; } } @media (max-width: 650px) { .typography .markup .twitter-embed, .typography.editor .markup .twitter-embed, .typography .markup .tweet, .typography.editor .markup .tweet { padding: 12px; } } @media (max-width: 650px) { .typography .markup .twitter-embed .tweet-text, .typography.editor .markup .twitter-embed .tweet-text, .typography .markup .tweet .tweet-text, .typography.editor .markup .tweet .tweet-text { font-size: 14px; line-height: 20px; } } @media (max-width: 650px) { .typography .markup .twitter-embed .tweet-photos-container.two, .typography.editor .markup .twitter-embed .tweet-photos-container.two, .typography .markup .tweet .tweet-photos-container.two, .typography.editor .markup .tweet .tweet-photos-container.two, .typography .markup .twitter-embed .tweet-photos-container.three, .typography.editor .markup .twitter-embed .tweet-photos-container.three, .typography .markup .tweet .tweet-photos-container.three, .typography.editor .markup .tweet .tweet-photos-container.three, .typography .markup .twitter-embed .tweet-photos-container.four, .typography.editor .markup .twitter-embed .tweet-photos-container.four, .typography .markup .tweet .tweet-photos-container.four, .typography.editor .markup .tweet .tweet-photos-container.four { height: 200px; } } @media (max-width: 650px) { .typography .markup .twitter-embed a.expanded-link .expanded-link-img, .typography.editor .markup .twitter-embed a.expanded-link .expanded-link-= img, .typography .markup .tweet a.expanded-link .expanded-link-img, .typography.editor .markup .tweet a.expanded-link .expanded-link-img { max-height: 180px; } } @media (max-width: 650px) { .typography .markup .twitter-embed a.expanded-link .expanded-link-descrip= tion, .typography.editor .markup .twitter-embed a.expanded-link .expanded-link-= description, .typography .markup .tweet a.expanded-link .expanded-link-description, .typography.editor .markup .tweet a.expanded-link .expanded-link-descript= ion { display: none; } } @media screen and (max-width: 650px) { .typography .markup .apple-podcast-container, .typography.editor .markup .apple-podcast-container { width: unset; } } @media (max-width: 420px) { .typography .markup .install-substack-app-embed img.install-substack-app-= embed-img, .typography.editor .markup .install-substack-app-embed img.install-substa= ck-app-embed-img { margin: 0 auto 16px auto; } } @media (max-width: 420px) { .typography .markup .install-substack-app-embed .install-substack-app-emb= ed-text, .typography.editor .markup .install-substack-app-embed .install-substack-= app-embed-text { margin: 0 0 12px 0; max-width: 100%; width: auto; text-align: center; } } @media (max-width: 420px) { .typography .markup .install-substack-app-embed .install-substack-app-emb= ed-link, .typography.editor .markup .install-substack-app-embed .install-substack-= app-embed-link { display: flex; justify-content: center; } } @media screen and (min-width: 481px) { .share-button-container { height: 38px; } } @media screen and (min-width: 481px) { .share-button-container a.comment { height: 38px; line-height: 38px; padding-right: 10px; } } @media screen and (max-width: 480px) { .share-button-container .separator { display: block; margin: 0; height: 8px; border-left: none; } } @media screen and (max-width: 480px) { .share-button-container a.share.first img { padding-left: 0; } } @media screen and (min-width: 481px) { .share-button-container a.mobile { display: none !important; } } @media screen and (min-width: 541px) { .settings-add-pub-modal-wrapper .container .add-recommending-pub-modal-co= ntainer { padding: 36px; height: 680px; } } @media screen and (min-width: 541px) { .settings-add-pub-modal-wrapper .container .add-recommending-pub-modal-co= ntainer .footer { position: absolute; bottom: 36px; margin: 0px; } } @media screen and (max-width: 650px) { .header-anchor-parent { display: none; } } @media screen and (max-width: 768px) { .post { padding: 16px 0 0 0; } } @media screen and (max-width: 650px) { .post .post-header .post-label { margin-top: 8px; } } @media screen and (max-width: 650px) { .post .post-header .meta-author-wrap.alternative-meta .meta-right-column = =2Epost-meta { margin-top: 6px; } } @media screen and (max-width: 650px) { .post .footer-facepile-container { height: 64px; padding: 0 16px; display: flex; align-items: center; justify-content: flex-start; width: 100%; } } @media screen and (max-width: 650px) { .post .post-footer.use-separators { justify-content: center; } } @media screen and (max-width: 650px) { .post .post-footer.next-prev { height: 64px; justify-content: space-between; box-sizing: border-box; } } @media screen and (max-width: 650px) { .post-contributor-footer .post-contributor-bio-table { display: block; } .post-contributor-footer .post-contributor-bio-table-row { display: flex; flex-direction: row; } .post-contributor-footer .post-contributor-bio-userhead-cell, .post-contributor-footer .post-contributor-bio-body-cell { display: block; } .post-contributor-footer .post-contributor-bio-body-cell { flex-grow: 1; } .post-contributor-footer .post-contributor-bio-body-table { display: block; } .post-contributor-footer .post-contributor-bio-body-table-row { display: block; } .post-contributor-footer .post-contributor-bio-copy-cell, .post-contributor-footer .post-contributor-bio-controls-cell { display: block; } .post-contributor-footer .post-contributor-bio-copy-cell { margin: 0 0 16px 0; } .post-contributor-footer .post-contributor-bio-controls-cell { width: auto; } .post-contributor-footer .post-contributor-bio-controls { margin: auto; } .post-contributor-footer .post-contributor-bio-controls .button.primary { width: 100%; } .post-contributor-footer .post-contributor-bio-text { font-size: 14px; } } @media screen and (min-width: 768px) { .post-silhouette { padding: 32px 0; } } @media screen and (max-width: 650px) { .post-silhouette .post-silhouette-title { margin-top: 10.44225025px; height: 120px; } } @media screen and (max-width: 650px) { .post-silhouette .post-silhouette-meta { width: 75%; } } @media screen and (max-width: 650px) { .post-silhouette .post-silhouette-meta.with-byline-image { margin: 20px 0; } } @media screen and (max-width: 650px) { .use-theme-bg .post-meta.alternative-meta .post-meta-item, .post-meta.alternative-meta .post-meta-item { padding-right: 16px; } } @media screen and (max-width: 370px) { .use-theme-bg .post-meta.alternative-meta .post-meta-item, .post-meta.alternative-meta .post-meta-item { font-size: 14px; } } @media screen and (max-width: 650px) { .use-theme-bg .post-meta.alternative-meta .post-meta-item.guest-author-pu= blication, .post-meta.alternative-meta .post-meta-item.guest-author-publication { display: none; } } @media screen and (max-width: 370px) { .post-meta .post-meta-item .post-meta-button { height: 36px !important; /* important to override in-line height style on emails */ } .post-meta .post-meta-item .post-meta-button .meta-button-label { display: none; } .post-meta .post-meta-item .post-meta-button > svg { margin-right: 0; } } @media screen and (max-width: 370px) { .post-meta .post-meta-item { font-size: 12px; } } @media screen and (max-width: 650px) { .post .floating-subscribe-button { bottom: 20px; right: 20px; } } @media (max-width: 1024px) { body .pullquote-align-left, body .pullquote-align-right, body .pullquote-align-wide, body .pullquote-align-center { float: none; margin: 0 auto; width: 100%; max-width: 100%; } } @media all and (-ms-high-contrast: none), (-ms-high-contrast: active) { body .markup table.image-wrapper img, body .markup table.kindle-wrapper img { max-width: 550px; } } @media (min-width: 1024px) { body:not(:has(#toc)) .captioned-image-container figure:has(> a.image2-off= set-left) { margin-left: var(--image-offset-margin); } body:not(:has(#toc)) .captioned-image-container figure:has(> a.image2-off= set-right) { margin-right: var(--image-offset-margin); } } @media (min-width: 1300px) { body .captioned-image-container figure:has(> a.image2-offset-left) { margin-left: var(--image-offset-margin); } body .captioned-image-container figure:has(> a.image2-offset-right) { margin-right: var(--image-offset-margin); } } @media (max-width: 1024px) { body { /* Disable offset on mobile/tablet */ } body .captioned-image-container figure:has(> a.image2-align-left), body .captioned-image-container figure:has(> a.image2-align-right) { float: none; margin: 1em auto; max-width: 100%; width: auto; padding: 0; } body .captioned-image-container figure:has(> a.image2-align-left.thefp), body .captioned-image-container figure:has(> a.image2-align-right.thefp) = { margin: 1em auto; } body .captioned-image-container figure:has(> a.image2-offset-left), body .captioned-image-container figure:has(> a.image2-offset-right) { margin: 1em auto; } body .captioned-image-container figure:has(> a.image2-align-left) .image2= -inset, body .captioned-image-container figure:has(> a.image2-align-right) .image= 2-inset { display: block; justify-content: initial; } } @media (max-width: 768px) { body .markup div.sponsorship-campaign-embed { margin-top: 24px; margin-bottom: 24px; } body .markup div.sponsorship-campaign-embed:first-child { margin-top: 0px; } } @media screen and (max-width: 650px) { body .markup div.youtube-overlay, body .markup div.vimeo-overlay { display: none !important; } } @media screen and (max-width: 370px) { body .markup div.tiktok-wrap { width: calc(95vw - 32px); height: calc((95vw - 32px - 2px) / 0.485714); } } @media screen and (max-width: 650px) { body .markup div.embedded-publication-wrap .embedded-publication.show-sub= scribe { padding: 24px; } } @media screen and (max-width: 650px) { body .markup div.subscription-widget-wrap .subscription-widget.show-subsc= ribe, body .markup div.subscription-widget-wrap-editor .subscription-widget.sho= w-subscribe, body .markup div.captioned-button-wrap .subscription-widget.show-subscrib= e { padding: 0px 24px; } } @media screen and (max-width: 650px) { body .markup div.subscription-widget-wrap .subscription-widget.show-subsc= ribe .subscription-widget-subscribe .button, body .markup div.subscription-widget-wrap-editor .subscription-widget.sho= w-subscribe .subscription-widget-subscribe .button, body .markup div.captioned-button-wrap .subscription-widget.show-subscrib= e .subscription-widget-subscribe .button { padding: 10px 12px; min-width: 110px; } } @media (max-width: 650px) { body .markup .twitter-embed, body .markup .tweet { padding: 12px; } } @media (max-width: 650px) { body .markup .twitter-embed .tweet-text, body .markup .tweet .tweet-text { font-size: 14px; line-height: 20px; } } @media (max-width: 650px) { body .markup .twitter-embed .tweet-photos-container.two, body .markup .tweet .tweet-photos-container.two, body .markup .twitter-embed .tweet-photos-container.three, body .markup .tweet .tweet-photos-container.three, body .markup .twitter-embed .tweet-photos-container.four, body .markup .tweet .tweet-photos-container.four { height: 200px; } } @media (max-width: 650px) { body .markup .twitter-embed a.expanded-link .expanded-link-img, body .markup .tweet a.expanded-link .expanded-link-img { max-height: 180px; } } @media (max-width: 650px) { body .markup .twitter-embed a.expanded-link .expanded-link-description, body .markup .tweet a.expanded-link .expanded-link-description { display: none; } } @media screen and (max-width: 650px) { body .markup .apple-podcast-container { width: unset; } } @media (max-width: 420px) { body .markup .install-substack-app-embed img.install-substack-app-embed-i= mg { margin: 0 auto 16px auto; } } @media (max-width: 420px) { body .markup .install-substack-app-embed .install-substack-app-embed-text= { margin: 0 0 12px 0; max-width: 100%; width: auto; text-align: center; } } @media (max-width: 420px) { body .markup .install-substack-app-embed .install-substack-app-embed-link= { display: flex; justify-content: center; } } @media screen and (min-width: 500px) { body .header a.logo { width: 42px; height: 42px; border-radius: 12px; } } @media screen and (max-width: 420px) { body .subscription-receipt table:first-of-type .subscription-amount .subs= cription-discount { width: 72px !important; } } @media screen and (min-width: 481px) { body .share-button-container { height: auto; } } @media screen and (max-width: 480px) { body .share-button-container .separator { display: block !important; margin: 0 !important; height: 8px !important; border-left: none !important; } } @media screen and (max-width: 650px) { .digest .item .post-meta-item.audience { display: none; } } @media screen and (min-width: 500px) { .digest-publication .logo img { width: 42px; height: 42px; border-radius: 8px; } } @media screen and (max-width: 650px) { .comments-page .container .comment-list .collapsed-reply { margin-left: calc(10 + 32px - 24px); } } @media screen and (max-width: 650px) { .comment > .comment-list { padding-left: 24px; } } @media screen and (max-width: 650px) { .finish-magic-login-modal .modal-content .container { padding: 24px 0; } } @media (max-width: 650px) { .reader2-text-b3 { line-height: 24px; } } @media screen and (max-width: 650px) { .reader2-text-h4 { line-height: 24px; } } @media screen and (min-width: 541px) { .user-profile-modal { padding-left: 12px; padding-right: 12px; } } @media screen and (max-width: 650px) { .subscribe-widget form.form .sideBySideWrap button.rightButton { padding: 10px 12px; } } @media screen and (min-width: 541px) { .pub-icon:hover .logo-hover, .feed-item-icon:hover .logo-hover { display: block; } } @media screen and (max-width: 650px) { .post-ufi.single-full-width-button .post-ufi-button-wrapper { width: 100%; padding: 16px; } .post-ufi.single-full-width-button .post-ufi-button-wrapper:empty { display: none; } .post-ufi.single-full-width-button .post-ufi-button { width: 100%; justify-content: center; } } @media screen and (max-width: 768px) { .file-embed-wrapper { padding: 0; } } @media screen and (max-width: 768px) { .file-embed-wrapper-editor { padding: 0; } } @media screen and (max-width: 768px) { .file-embed-wrapper-editor:active { padding: 0; } } @media only screen and (max-width: 650px) { .file-embed-button.wide, .file-embed-error-button.wide { display: none; } } @media only screen and (min-width: 630px) { .file-embed-button.narrow, .file-embed-error-button.narrow { display: none; } } @media screen and (min-width: 541px) { .audio-player-wrapper .audio-player { min-width: 500px; } } @media screen and (max-width: 650px) { .audio-player-wrapper .audio-player .audio-player-progress { border-left-width: 16px; border-right-width: 16px; } } @media screen and (max-width: 650px) { .audio-player-wrapper .audio-player .audio-player-progress .audio-player-= progress-bar .audio-player-progress-bar-popup { top: -54px; } } @media screen and (max-width: 650px) { .audio-player-wrapper-fancy .audio-player .audio-player-progress { border-left-width: 16px; border-right-width: 16px; } } @media screen and (max-width: 650px) { .audio-player-wrapper-fancy .audio-player .audio-player-progress .audio-p= layer-progress-bar .audio-player-progress-bar-popup { top: -54px; } } @media (min-width: 250px) { .audio-player-wrapper-fancy .audio-player { padding: 32px; } .audio-player-wrapper-fancy .audio-player .btn-group { display: flex; } .audio-player-wrapper-fancy .audio-player .btn-group .button:last-of-type= { display: block; } } @media (min-width: 300px) { .audio-player-wrapper-fancy .audio-player .btn-group { display: block; } .audio-player-wrapper-fancy .audio-player .btn-group .button:first-of-typ= e { display: block; } } @media (min-width: 350px) { .audio-player-wrapper-fancy .audio-player .audio-player-substack-logo { display: block; } .audio-player-wrapper-fancy .audio-player .audio-player-title { margin-top: 16px; } .audio-player-wrapper-fancy .audio-player .audio-player-hero-image-contai= ner { padding-top: 15%; width: 15%; display: block; } .audio-player-wrapper-fancy .audio-player .btn-group .button:first-of-typ= e { display: block; } .audio-player-wrapper-fancy .audio-player .audio-player-substack-logo { display: block; } } @media (min-width: 350px) { .audio-player-wrapper-fancy .audio-player .audio-player-hero-image-contai= ner { padding-top: 25%; width: 25%; display: block; } .audio-player-wrapper-fancy .audio-player .btn-group { display: flex; } .audio-player-wrapper-fancy .audio-player .btn-group .button:first-of-typ= e { display: block; } } @media (min-width: 400px) { .audio-player-wrapper-fancy .audio-player .audio-player-hero-image-contai= ner { padding-top: 40%; width: 40%; } } @media (max-width: 400px) { .audio-player-wrapper-fancy .audio-player .btn-group { margin-top: 12px; } .audio-player-wrapper-fancy .audio-player .btn-group .button { font-size: 13px; padding: 6px 12px; height: auto; margin-top: 10px; } } @media (min-width: 600px) { .audio-player-wrapper-fancy .audio-player .audio-player-hero-image-contai= ner { padding-top: 55%; width: 55%; } } @media (max-width: 650px) { .poll-editor-modal { min-width: calc(100% - 20px); } } @media (max-width: 750px) { .poll-embed .poll-anchor-target .poll-anchor-copy-button { left: 8px; top: 45px; } }3D""
The Problem With P(doom)
͏     ­= 47;     ­͏     ­͏     = ­͏     ­͏     ­͏  =   ­͏     ­͏     ­T= 7;     ­͏     ­͏     &= #173;͏     ­͏     ­͏   =   ­͏     ­͏     ­͏= ;     ­͏     ­͏     &#= 173;͏     ­͏     ­͏   &= #8199; ­͏     ­͏     ­͏=     ­͏     ­͏     = 73;͏     ­͏     ­͏   &#= 8199; ­͏     ­͏     ­͏ =     ­͏     ­͏     = 3;͏     ­͏     ­͏   = 199; ­͏     ­͏     ­͏ &= nbsp;   ­͏     ­͏     ­= ;͏     ­͏     ­͏   Q= 99; ­͏     ­͏     ­͏ &n= bsp;   ­͏     ­͏     ­= ͏     ­͏     ­͏   ̳= 9; ­͏     ­͏     ­͏ &nb= sp;   ­͏     ­͏     ­&= #847;     ­͏     ­͏    = ; ­͏     ­͏     ­͏ &nbs= p;   ­͏     ­͏     ­&#= 847;     ­͏     ­͏    = ­͏     ­͏     ­͏  = ;   ­͏     ­͏     ­= 47;     ­͏     ­͏     = ­͏     ­͏     ­͏  =   ­͏     ­͏     ­T= 7;     ­͏     ­͏     &= #173;͏     ­͏     ­͏   =   ­͏     ­͏     ­͏= ;     ­͏     ­͏     &#= 173;͏     ­͏     ­͏   &= #8199; ­͏     ­͏     ­͏=     ­͏     ­͏     = 73;͏     ­͏     ­͏   &#= 8199; ­͏     ­͏     ­͏ =     ­͏     ­͏     = 3;͏     ­͏     ­͏   = 199; ­͏     ­͏     ­͏ &= nbsp;   ­͏     ­͏     ­= ;͏     ­͏     ­͏   Q= 99; ­͏     ­͏     ­͏ &n= bsp;   ­͏     ­͏     ­= ͏     ­͏     ­͏   ̳= 9; ­͏     ­͏     ­͏ &nb= sp;   ­͏     ­͏     ­&= #847;     ­͏     ­͏    = ; ­͏     ­͏     ­͏ &nbs= p;   ­͏     ­͏     ­&#= 847;     ­͏     ­͏    = ­͏     ­͏     ­͏  = ;   ­͏     ­͏     ­= 47;     ­͏     ­͏     = ­͏     ­͏     ­͏  =   ­͏     ­͏     ­T= 7;     ­͏     ­͏     &= #173;͏     ­͏     ­͏   =   ­͏     ­͏     ­͏= ;     ­͏     ­͏     &#= 173;͏     ­͏     ­͏   &= #8199; ­͏     ­͏     ­͏=     ­͏     ­͏     = 73;͏     ­͏     ­͏   &#= 8199; ­͏     ­͏     ­͏ =     ­͏     ­͏     = 3;͏     ­͏     ­͏   = 199; ­͏     ­͏     ­͏ &= nbsp;   ­͏     ­͏     ­= ;͏     ­͏     ­͏   Q= 99; ­͏     ­͏     ­͏ &n= bsp;   ­
<= tbody><= tr>
Forwarded this = email? Subscribe here for more

Not Even Wrong

The Problem W= ith P(doom)

Harry Law
<= /td>
 
<= td>
= <= tr height=3D"16">
<= /table>
3D""
3D""
=
READ IN APP3D""
&= nbsp;
<= td style=3D"text-align: center;">
3D""
The = Fall of the Rebel Angels by Pieter Brueghel the Elder (1562)

John Graunt was a merchant who = lived in 17th century London. In 1662, in a city familiar with plague, Grau= nt consulted parish death records to figure out how long Londoners could ex= pect to live. He could tell you that roughly 36 out of every 100 children b= orn in the city would die before their sixth birthday. This probability cou= ld be checked by anyone willing to do the same sums using the same sources.=

Three and a half centuries later, AI researchers are= calculating a morbid probability of their own. They wonder about thinking = machines seeing off the human race, and refer to the likelihood of that pos= sibility as P(doom). The term began life as an inside joke in rationalist s= paces in the closing years of the 2000s, but shot to prominence after resea= rchers signed an open letter about the threat posed by AI developmen= t in 2023. In the AI world, everyone has their number. Yudkowsky puts it at= roughly 95–100 percent. Geoffrey Hinton sits around 10–20 perc= ent, while Marc Andreessen reports a bullish 0 percent (a logical impossibi= lity because you cannot rule out something that has never happened and whos= e conditions have never been tested). The mean estimate among surveyed AI re= searchers is about 14 per cent.

We have so= me who believe that AI is almost certain to wipe out humanity and others wh= o ascribe a 0 per cent probability to the same outcome. This is curious for= a supposedly hard-headed prediction made by some of the world’s most= credentialed scientists. Imagine a roomful of cardiologists who, given the= same scans and the same patient history, exposed to the same training and = the same clinical standards, disagreed on whether the probability of heart = failure was 0.1 percent or 99 percent.

You would not conclu= de that this was a hard problem on which reasonable people differed. In fac= t, you would probably conclude that at least some of our medics may need to= visit neurology down the hall. With P(doom), the estimates from AI researc= hers range from 0 percent to 99 percent (and a few nines then some). When s= omeone tells you they assign a 15 per cent probability to AI wiping out hum= anity, it is worth asking what kind of claim they are actually making. It s= ounds like a statement about the world. It is not.

To be cl= ear: my objection is not to probabilistic reasoning, which is straightforwa= rdly indispensable in domains where estimates can be calibrated against out= comes over time. My problem concerns what we might call proximate and ultim= ate formulations. Proximate estimates for COVID were things like the hospit= alization rate and case severity, while an ultimate estimate was the infect= ion fatality rate. For P(doom), proximate estimates include the conditions = under which models exhibit deceptive behavior and the observed failure rate= s of current alignment techniques. Here, the ultimate estimate is the thing= itself.

Someone might know that current models can be made= to exhibit deceptive alignment under laboratory conditions or that existin= g alignment techniques have specific failure modes. But to cram those findi= ngs into a container labeled “15% chance of extinction” involve= s a series of judgments about unprecedented transitions for which there is = no model connecting proximate estimates to their ultimate counterpart. Know= ing that fine-tuning on specific tasks can induce broader misalignment tell= s us little about how well alignment solutions will generalize or how geopo= litical actors will respond. Compare with COVID, where hospitalization and = case-severity data fed epidemiological models that could be tested against = actual deaths.

A natural objection is that some people are = good at formulating predictions in lots of different domains, so perhaps th= eir assessment of P(doom) ought to carry weight. After all, some superforec= asters assign 17 per cent to various events and see those events happen rou= ghly 17 per cent of the time. You can check a forecaster’s guesses ac= ross many different predictions in such a way as to mean we’re dealin= g with a property of the forecaster’s overall policy rather than of a= ny individual estimate. When a superforecaster says 17 per cent for a cease= fire in some conflict, that number is useful because they also say 17 per c= ent for hundreds of other things and roughly 17 per cent of those things ha= ppen.

Superforecasters earn their calibration (that is, the= extent to which confidence lines up with their track record over time) usi= ng predictions with short time horizons, a history of similar events to dra= w on, and a mechanism that allows the forecaster to correct course when nee= ded. For P(doom), there is no source of comparable predictions against whic= h to calibrate. A forecaster might be superbly calibrated on elections and = geopolitical high drama, but that tells us nothing about whether their numb= er for an unprecedented event is any good.

What I Talk Ab= out When I Talk About P(doom)

There are two influential in= terpretations of probability: subjective and objective. Where the former de= als with how confident you are in a claim, the latter stresses what actuall= y happens when you test it. Anyone who has spent even a little time thinkin= g about the reliability of P(doom) metrics probably has a sense that the ca= lculations ought to be taken with a pinch of salt. Many in rationalist or r= ationalist-adjacent circles are well aware of the difference between these = two accounts, but lots of people who absorb these views are not.

That goes for those who adopt the terminology to signal their m= embership of the in-group, and those who hear talk of P(doom) on the news a= nd wonder why one of the world’s most cited researchers thinks there&= #8217;s a one in five chance that AI will wipe out the human race. When a c= hief global strategist at one of the largest investment research firms tells C= NBC it’s 50/50 whether AI destroys humanity by mid-century,= the average viewer — incorrectly but understandably — assumes = that figure estimates the likelihood that it will happen.

Every single P(doom) you hear is a subjective probability. It is a me= asure of the degree of rational belief one holds in a proposition given the= available evidence — a property of an agent’s epistemic state = rather than of the world. When you say “I think there’s a 50 pe= r cent chance this meeting will be a waste of time,” you’re not= drawing on a frequency table of past meetings. You’re expressing how= confident you feel given what you know about the agenda and who’s at= tending. All the framework requires is that your degrees of belief are cons= istent with one another. Beyond this internal consistency, there is no furt= her requirement that your probabilities correspond to the shape of reality.=

If your beliefs are internally coherent, you can mul= tiply your probability for any outcome by the value of that outcome —= however you choose to measure it — and arrive at its expe= cted value, a single number that tells you how good or bad a bet= looks on average. From there you can derive a utility function and compare= any set of actions on a common scale. The framework of cause prioritizatio= n, wherein resources are allocated to whichever cause produces the greatest= expected good per dollar, inherits this apparatus. This approach is popula= r with rationalist-adjacent communities, especially Effective Altruism, bec= ause it transforms questions of moral life into properties that can = be measured and optimized.

Objective accounts = of probability take a different approach. Here probability is what actually= happens in the world when you repeat something many times. This species of= probability is a physical property of systems, one that is empirically tes= table. If you flip a coin a thousand times roughly half will land heads, so= we can say the probability is that ratio. The same l= ogic applies to complex phenomena, like radioactive decay, where about half= the atoms in a given sample will decay within the half-life period (and yo= u can go and check).

The subjective interpretation i= s internally coherent and mathematically elegant. But coherence should not = be confused for empirical content. My objection to P(doom) flows from the s= imple fact that subjective probability claims are, by their nature, unfalsi= fiable. That’s generally fine in many instances — like casual e= veryday judgments or betting on the Super Bowl — but not when the sta= kes are high enough to redirect billions of dollars and make or break gover= nment policy. Statistical reasoning about unlikely but potentially devastat= ing scenarios is obviously useful, but only when properly grounded in claim= s we can falsify.

Your Problem Too

For a probability to tell you about something other than the speaker’= ;s state of mind, it needs a collection of similar events from which a rati= o can be drawn. To say that a coin lands heads 50 percent of the time, we n= eed a set of coin flips to draw from. This is our reference clas= s.

Clearly there is no reference class fo= r human extinction. As far as every AI researcher in the world knows, the e= vent is singular and unprecedented. Traditionally, the reference class prob= lem has been associated with the objective interpretation of probability. T= his is because objective probability is calculated according to the ratio o= f how many times something happened out of how many times it might have hap= pened. If the event has never happened and can’t be repeated, then th= ere’s nothing to compute.

But the reference cla= ss problem isn’t only relevant for objective ac= counts of probability. Alan Hájek famously made the case that i= t affects every interpretation of probability when applied to singular even= ts. As he put it, “the reference class problem is your problem too.&#= 8221; The argument goes something like this: before you can assign a probab= ility to a one-off event, you have to decide what kind of event it is. Is A= I-caused extinction a case of “new technology going wrong,” = 220;species-level catastrophe,” or “unprecedented transition in= the nature of intelligence”? Each framing suggests a different proba= bility, and there is no principled way to choose between them. The number y= ou arrive at depends on how you describe your target.

This is because good explanation= s resist variation. If Graunt had doubled his estimate of childhood mortali= ty, the parish registers would have corrected him. If an epidemiologist in = early 2020 had put the infection fatality rate at 10 per cent rather than 1= per cent, the incoming hospital data would have shown the estimate to be w= rong. The explanation is enmeshed with the world in such a way that it cann= ot be led to wherever you would like it to go. With P(doom), you can swap t= he causal pathway from deceptive alignment to resource competition and infl= ate the number from 5 per cent to 7.5 per cent. The estimate can contort to= accommodate any figure and any causal chain because the only constraint it= faces is internal coherence.

In the context of P(doom), d= efenders might say something like “We don’t need a reference cl= ass because we’re expressing a degree of personal belief.” This= is perfectly coherent on the subjective view, but it is also the move that= severs the claim from its empirical content. If your probability doesnR= 17;t need to correspond to any feature of the world, then it can’t be= wrong about the world. And a claim that can’t be wrong about the wor= ld tells you nothing about it.

We might call the resu= lt a Gettier probability, a credence that is internal= ly justified — and that might even correspond to reality — but = that lines up with the world through luck rather than judgement. In epistem= ology, a Gettier case is a belief that is true and justified but not true <= /span>because of its justification. Any P(doom), where the j= ustification and the truth are aligned, will (assuming the estimate was cor= rect) have the same structure. A researcher’s estimate might happen t= o match whatever the actual risk turns out to be, but the match would be ac= cidental relative to the method.

Gettier probabiliti= es show up everywhere, like when a consulting firm estimates that a given m= arket will grow at 7.3 per cent and lo and behold they are correct. The sam= e is true for a geopolitical risk score that looks precise but cannot be up= dated by any observation short of the catastrophe it purports to predict. P= (doom) is distinctive only in that the stakes are high enough that the prob= ability escapes containment and ends up on the morning news.

Bayesian probability demands that credences be precise.= When estimates range from 0% to 99% with no mechanism to adjudicate, the f= ramework is not being applied so much as invoked. P(doom) manages the impre= ssive trick of satisfying neither the Bayesian standard (precise credences = that can be adjudicated between) nor the falsifiability standard (empirical= claims that can be proved wrong).

I’m not argu= ing that we ought to stop reasoning about novel events. Subjective probabil= ity is useful because it helps us think under uncertainty without good data= =2E Some of the best calls in recent A= I history were made by researche= rs who formed strong priors about scaling laws and bet on them before the e= vidence was in. Those bets were vindicated because they were testable. Scal= ing laws either held or they didn’t; capability thresholds either arr= ived on schedule or did not. My concern is that P(doom) borrows the confide= nce of claims like these without sharing their vulnerability to refutation.=

I Am Sure You Are Very Sure

Forget, for a moment, whether P(doom) can be tested. Consider instead whe= ther it is the kind of thing that can be meaningfully quantified in the fir= st place. In 1921, the economist Frank Knight drew a distinction between risk and uncertainty. Risk is wha= t you face when you can calculate the odds as with flicking a small ball on= to a roulette wheel. Uncertainty is what you face when you can’t, eve= n in principle, because the situation is too poorly understood to yield a n= umber.

Knight argued that the two are different in k= ind.

The Bayesian tradition’s response to Knight was = to argue that subjective probability does away with the distinction: if you= can always express a degree of belief, then there is no situation that res= ists quantification. P(doom) is a textbook application of this move, but th= e dissolution works only if the resulting number can be tested via real-wor= ld feedback. For P(doom), the Bayesian move converts Knightian uncertainty = into a figure that has the form of risk without any of the mechanisms that = make risk estimates trustworthy.

The same process, wh= ere uncertainty becomes risk, happens wherever a decision-making framework = demands a numerical input and none is available. Firms routinely assign pre= cise probabilities to market scenarios that are genuinely unprecedented. Th= ey don’t have any evidence that warrants a specific number, but they = do have a spreadsheet with cells that need filling. The defense that such f= igures are “rough heuristics” does not help. The problem is tha= t they are treated simultaneously as a casual shorthand and as inputs into = value calculations depending on whatever is most convenient. This is why th= e “shmrobability” move doesn’t fix our problem. It= concedes we’re not really talking about probability, while trying to= preserve its practical content. Either the number is a rough heuristic and= we should stop plugging it into expected value calculations, or it is a se= rious input to policy and we should hold it to the standards that implies.<= /span>

If P(doom) is a subjective credence, there is = no principled way to adjudicate between two people who disagree. A P(doom) = of 95 per cent and one of 50 per cent are both internally coherent insofar = as neither violates any rule of the subjective framework. You can be confid= ent in your P(doom) and I can be confident in mine, and there is no observa= tion or outcome that could settle which of us is right. Attempts to resolve= this from within the Bayesian model, from Aumann’s agreement theorem onwards, all founder on conditions that do not hold (like shared st= arting points, perfect rationality, or common knowledge of each other’= ;s beliefs).

The problem compounds when these = estimates enter expected value calculations, where even a vanishingly small= probability multiplied by the stakes of human extinction produces numbers = large enough to dominate decisions. The rationalist and rationalist-adjacen= t communities recognize this dynamic as Pascal’s mugging, while group= s like GiveWell have cautioned against taking such estimates at face= value.

The reader should not take from this essay t= hat AI risk is unworthy of attention. Rather, my point is that P(doom) is t= he wrong epistemic instrument for thinking about it. The alternative is to = formulate falsifiable conjectures about the world and then try to refute th= em, which is exactly what AI safety researchers do in their actual work. Th= ey propose that particular training regimes will produce deceptive behavior= and they predict that models beyond a certain capability threshold will re= sist correction. They hypothesize that misalignment in one domain will gene= ralize unpredictably to others, and they can test this by measuring the tra= nsfer of learned strategies across task distributions. They ask whether mod= els that pass safety evaluations in deployment conditions will behave diffe= rently when those conditions change. Each of these lines of research produc= es claims that can fail, and this vulnerability is what makes them worth pa= ying attention to.

Individual claims about deceptive = behavior or capability thresholds do not by themselves tell you how much to= spend on alignment versus pandemic preparedness, yet real institutions hav= e treated P(doom) as though it could. The effective altruist community has = directed hundreds of millions of dollars toward AI existential risk reducti= on. Open Philanthropy, the field’s largest funder, used subjective pr= obability frameworks to weigh AI risk against global health, biosecurity, a= nd farm animal welfare. Whether or not that assessment was correct, = the reasoning behind it is only as good as the numbers it relies on.

A figure that ranges from 0 per cent to 99 per cent depen= ding on who you ask and with no way to adjudicate between them is about as = useful as no number at all. Aggregating these estimates might seem to help = inasmuch as a mean of 14 per cent looks more sober than bouncing between ex= tremes. Aggregate or individual, the problem is that neither incur a cost f= or being wrong. A bookmaker’s odds are constrained by the market just= as an insurer’s premiums are shaped by the frequency and nature of c= laims made. In both cases, the issuer’s assessment improves over time= because they suffer when it is incorrect. There is no market to punish a m= ispriced P(doom) and no settlement date on which the estimate is tested.

We might say something like: “Even if P(doom) is = imprecise, it’s still better than no number because we need to alloca= te resources somehow.” But here we’re assuming that internal co= herence still outperforms the absence of a number when you have = to act. We can live with imprecision under these conditions, but= not at the expense of claims that cannot be falsified that get steamrolled= by the headline rating.

Governments fund pandemic p= reparedness efforts without a big round number for the probability of the n= ext pandemic. Institutions allocate resources under genuine uncertainty all= the time by doing things like funding a portfolio of approaches, setting t= hresholds for observable harms, identifying the cheapest reversible interve= ntions, demanding observable milestones before scaling commitment, and buil= ding the capacity to pivot as new evidence arrives.

The physicist Wolfgang Pauli was famously unforgiving of bad theo= ry. When a colleague asked him to assess a young physicist’s paper, h= is verdict was that it was “not even wrong.” He meant that its = central claim was so removed from evidence that it could not be proved fals= e. P(doom) is not even wrong. Graunt’s mortality tables may have been= technically crude, but they were accountable. They could be checked agains= t next year’s parish records, and if they were wrong, the records wou= ld show it.

© 2026 Cosmos Institute
500 W 2nd Street, Suite= 1900, Austin, TX 78701
Unsubscribe

3D"Get3D"Start

 
<= td width=3D"8" style=3D"min-width:8px;">
= 3D""Like
3D""Comment
3D""Restack
 
3D""3D"" --c48b76fee43dc8f2138f38d470ef9a1fcfe44e0c02733df6aaaad835a810--