newton lost money in the south sea bubble.

i find this comforting.

there is a remark attributed to him about being able to calculate the movement of heavenly bodies while finding the madness of people beyond calculation. whatever the exact wording of the remark, the investment went badly enough to make the sentiment understandable.

i think about this whenever somebody gives me a probability for the extinction of humanity.

because i have seen a budget.

you sit down with the numbers. you account for the expenses. you make reasonable assumptions about revenue. everybody agrees that the coming year should look approximately like this.

then a customer delays payment.

something breaks.

the supplier who assured you that the price would hold has apparently experienced a change of heart.

and now you are three weeks into the year explaining why the year has failed to behave like the spreadsheet.

government, corporation, small business. the scale changes. i would still like to sit in the meeting where the original assumptions are compared with what actually happened.

so when somebody casually assigns a percentage to the disappearance of the entire human species, i have questions.

quite a few, actually.

we struggle with things we can keep checking

weather forecasting has an extraordinary amount going for it.

satellites. observations. physical models. actual weather arriving afterward so that somebody can check whether the forecast was any good.

and the uncertainty still grows as you look further ahead. the people doing the forecasting explain this themselves.

i appreciate that.

i can make a decision using a forecast that admits what it cannot resolve. i can carry an umbrella. change the date. accept that there is a chance the whole thing goes differently.

insurance is another interesting case.

people build models of future losses, use historical information, compare predictions and eventually have actual claims against which to examine the estimates. there is serious mathematics involved, along with a continuing need to find out whether the mathematics was useful.

uncertainty has somewhere to show up.

that is what i keep looking for in the extinction percentages.

where does the uncertainty show up?

inside the number?

around it?

in a qualification that disappears when somebody quotes the number on a podcast?

and how will we know whether the person producing it has any particular ability to forecast the thing being discussed?

i do not think these are hostile questions. i would ask them before committing money to a business forecast.

apparently asking them about the end of humanity requires a considerably longer explanation.

newton also had a date

the part that really got me was discovering that newton had spent considerable time on theology and prophetic chronology.

and there, in his papers, was 2060.

yes. that newton.

his calculation concerned the end of a corrupt religious age and the anticipated events around Christ’s return. human extinction was not what he was predicting. one route to the date involved interpreting 1,260 prophetic days as years and counting from AD 800.

i can follow the addition.

i would like to spend considerably more time on why those were the numbers being added.

why that starting point? why that interpretation? what makes the correspondence between the text and the historical event reliable?

those questions remain even when the arithmetic is correct.

newton understood some of the difficulty himself. he wrote:

“It may end later, but I see no reason for its ending sooner.”

he also explained that he wanted to restrain rash predictions whose failures brought prophecy into disrepute.

which makes the popular version of the story rather funny.

the man writing against reckless date-setting became another name on the list of people who supposedly knew when the world would end.

i am interested in how that happened too. the qualification required reading. the date fitted in a headline.

the diary is quite full

religious texts themselves can be remarkably reluctant to supply a timetable.

Matthew 24:36 leaves the day and hour unknown. Qur’an 7:187 places knowledge of the Hour with God.

subsequent interpreters have been more accommodating.

a few appointments:

• 22 october 1844. the Millerite expectation of Christ’s return. the failed expectation became the Great Disappointment.
• 1988. Edgar Whisenant published 88 Reasons for expecting the Rapture that year. the year passed without it.
• 2011. Harold Camping’s May expectation was followed by an October deadline. both passed without the publicly anticipated events.
• 21 december 2012. the popular apocalypse attached to the Maya calendar. the calendar cycle ended; the supposed Maya prediction of global destruction was a misrepresentation.

there are differences between these claims. a religious return, a judgment and the death of every human are different events.

still, i find the history useful.

people have arrived at very consequential conclusions before. they have supplied dates, calculations and explanations. some have been deeply sincere.

sincerity does not do much for me here.

neither does the observation that this time the person is exceptionally intelligent.

newton is already in the conversation.

a new prediction deserves to have its mechanism examined. the failures above cannot settle it. they do make me reluctant to be impressed by the presentation before i have understood the claim.

somebody has put a percentage on it

i do not object to somebody putting their uncertainty into numbers.

i object to losing sight of what the number represents.

the survey Thousands of AI Authors on the Future of AI asked 2,778 researchers about future capabilities and outcomes. substantial numbers assigned at least a ten per cent probability to outcomes as bad as human extinction. the same survey also recorded considerable uncertainty and a majority expecting good outcomes to be more likely than bad.

i want to read those answers.

i also want to understand how the respondents arrived at them.

what experience supports the estimate? which parts of the forecast lie inside their expertise? how much agreement comes from independent work, and how much comes from people reading the same arguments?

suppose twenty researchers have all read the same account of how an artificial intelligence could cause a biological catastrophe.

their agreement may be informative.

but i would still like somebody to inspect the biology.

asking more people for their estimate does not, by itself, resolve a shared assumption.

then there is the forecast itself.

what exactly is supposed to happen?

does humanity disappear before the date given? does a process begin that eventually makes extinction unavoidable? does society undergo a catastrophe from which it might recover?

i want that written down before the discussion gets emotional.

because once we have said “the end of humanity,” it is very easy for the sheer size of the claim to make the details feel petty.

they are the claim.

the route through the world is the thing being predicted.

i would like the relevant people in the room

during the pandemic, i read Sunetra Gupta’s Pandemics: Our Fears and the Facts.

it was published in 2013. it asks questions about the conditions under which pandemics arise and our ability to anticipate or prevent them.

i wish more of the people discussing biological extinction would read it.

and then read other people. including people who disagree with her.

i am not a virologist. the reading gave me more questions, which seems a fairly reasonable outcome.

what irritates me is how quickly a discussion can move from an improvement in an AI model to a conclusion about the fate of a species, passing through several fields as though the difficult work in each has already been completed.

a model becomes better at answering a biological question.

what follows from that?

which practical capability improved? what still requires experimentation? what has been demonstrated outside the conditions of the evaluation?

then, further down the argument, what makes the proposed threat work across different people, places and environments?

i would like the people studying those questions to examine the assumptions.

i would like their disagreements included in the account.

and i would like an objection to survive long enough to receive an answer before somebody says that a sufficiently intelligent system will simply overcome it.

perhaps it will.

that possibility is worth investigating. it does not tell me how confidently to assign success to every unresolved step.

this applies to me as well. i cannot borrow Gupta’s expertise, attach it to a claim she never made, and announce that biology has settled the matter.

i want the work done.

the percentage can wait while we establish what it is supposed to describe.

how does it reach everybody?

i went down the nuclear-war question with what felt like a stupid problem.

how do the effects reach everybody?

i was thinking geographically. where would the bombs fall? how much territory would be directly affected? what happens to the people nowhere near those places?

the serious answer takes you into the atmosphere and the food supply.

a Rutgers-led study published in Nature Food modelled soot from nuclear-war firestorms, the resulting climatic effects and damage to food production. its large US–Russia war scenario estimated that more than five billion people could die of hunger.

that is an argument i can begin examining.

there is a chain. assumptions about the conflict, fires, soot, climate, crops and food availability.

i can ask where the estimates came from. i can look for disagreements about individual steps. the researchers themselves identify areas requiring more detailed work.

but when the discussion moves to extinction, i still want to know what happens to the people who remain.

where are they?

what can they produce?

how long do the conditions last?

what prevents any viable population from surviving them?

the deaths do not become less horrifying because someone asks those questions. the questions become necessary because extinction is the event we are now discussing.

it does not have to happen everywhere simultaneously. survivors could face conditions that eventually make their continued existence impossible.

then include that period in the argument.

i am still reading.

the Doomsday Clock gave me another small surprise. i had “less than five minutes to midnight” in my head.

the setting announced in january 2026 was 85 seconds.

the Bulletin also explicitly says the Clock is not a forecasting tool.

i appreciate the clarification, because seconds are unusually persuasive units for something that is not actually counting down.

i can understand the warning. i still cannot convert the position of the hand into a probability that humanity disappears by a particular year.

i keep going back to the stones

i have spent an unreasonable number of hours looking at megaliths.

some of them look like humanity showing off.

i mean that affectionately.

you look at the scale of the undertaking and start wondering about the person who proposed it. then the people who had to figure out how to do it. somebody must have spent an awful lot of time dealing with the gap between the idea and the stone.

the finished object makes it easy to forget that process.

i find myself imagining the failed attempts. the arguments. the person insisting a different method would work.

that is speculation on my part. it is also why i am interested in the builders.

the flood stories pull me in for a related reason.

they keep appearing in my reading. then i look at the dating and the apparent unity becomes harder to maintain.

even within Mesopotamia, the archaeological record contains distinct flooding episodes. the deposits discussed at Kish, Shuruppak and Lagash do not all belong to one simultaneous event.

i can imagine a flood ending the world someone knew.

their home is gone. the fields are gone. the institutions that organized life cannot function. the people who knew how things worked may be dead.

what does “the world ended” mean to someone living through that?

i do not think we should casually answer on their behalf.

but i do want to know what happened afterward, and what happened beyond the area they could see.

humanity survived the history that produced us. that observation cannot supply a probability of surviving every new threat. we would not be here discussing the histories in which no humans remained.

still, when forecasting the future, i want adaptation examined.

what are people doing while the disaster develops?

what do they learn? where do they move? what becomes possible once they understand more of what is happening?

some responses will fail. some may worsen the situation. i want those possibilities included too.

but an account extending decades into the future needs to explain how it treats human ingenuity during those decades.

i am particularly interested when the forecast concerns intelligence.

how much additional capacity to understand and respond to a threat is being assumed? who has access to it? what prevents it from being useful?

those questions belong in the same analysis.

i want to see the old forecasts

this is where integrity becomes practical.

keep the forecast.

keep the original wording. keep the date. keep the assumptions that made the estimate seem reasonable at the time.

when the number changes, explain why.

i have no problem with somebody changing their mind. i am much more interested in whether i can follow the change.

perhaps a capability arrived earlier than expected. perhaps an experiment exposed a failure. perhaps a control worked better than anticipated.

all of that can matter.

but there has to be some imaginable evidence that lowers the estimate too.

otherwise i do not understand what i am being invited to participate in.

suppose a model behaves dangerously and the risk estimate increases. fair enough.

suppose it behaves well and the estimate increases because the behaviour might be strategic.

that second possibility may deserve investigation. i still want to know what result would count against it, and how the experiment distinguishes between the explanations.

i cannot keep granting certainty to the worst interpretation simply because it remains possible.

i would also like some shorter-term commitments.

a forecast reaching to 2060 presumably depends on things happening before 2060. which capabilities do you expect in the next few years? which bottlenecks should disappear? what would surprise you?

write those down as well.

a probability forecast does not become wrong simply because an event assigned a small chance fails to happen once. i understand that. but a history of forecasts gives us more to examine than a reputation and a compelling essay.

and please include the misses.

i have read enough forecasts to know how much more informative they become when the person making them lets you inspect what previously went wrong.

my own rather inconvenient number

i would take the survival side through 2060.

my immediate, argumentative answer is 0% chance humanity ends before then.

yes, i can hear the objection already.

i have not established impossibility. that zero is a deliberately provocative wager, and it should not be mistaken for a measured probability.

i owe that distinction to anyone reading an article with this title.

what i can say plainly is that the extinction arguments i have examined have not persuaded me. i keep finding steps that require more work, assumptions that need people from other fields to inspect them, and conclusions that become less definite once i follow them beyond the initial catastrophe.

i am willing to keep reading.

i would particularly like to see more work on the people who remain, the conditions they face and the capacity they have to change those conditions. i still have questions about the food-production assumptions in the nuclear-war studies.

2060 leaves some time to go through the material. i would like the original forecasts and their revisions to remain available while we do.