The singularity is usually discussed as a date: before 2030 for Vernor Vinge, 2045 for Ray Kurzweil. But the idea Vinge presented at a NASA-sponsored symposium in 1993 was first a claim about knowledge – that past a certain point our models of the future fail. This essay takes that claim seriously. It argues that a horizon of prediction is a poor foundation for central plans and a good one for institutions that learn, and that the change the singularity describes is better understood as a permeation than as a point.
Three singularities
The word has carried three different ideas. Stanislaw Ulam, recalling a conversation with John von Neumann in a 1958 memorial, meant ordinary technological progress accelerating until human affairs as we know them could not continue. Vinge in 1993 meant something narrower and stranger: the creation of superhuman intelligence, by AI or by amplifying human minds, after which the human era would end. Kurzweil, from 2005 on, meant a smooth exponential across computing, biology and nanotechnology, ending in a merger of human and machine intelligence that he dates to 2045.
What they share is the image of a horizon. Vinge said he called the event a singularity because it is a point where our models must be discarded. That is an epistemic claim before it is a technological one. It says less about what will happen than about what we will be able to know.
The paradox of a dated horizon
There is an obvious tension in giving a horizon a date. Kurzweil addressed it in his 2011 reply to Paul Allen and Mark Greaves: individual innovations are unpredictable, he argued, but the aggregate trend in price and performance is not. One can doubt the trend, as his critics did, and still accept the logic. A horizon can approach on schedule even if nothing beyond it can be seen.
In 2026 we stand inside Vinge’s window, four years before its far edge and three before Kurzweil’s 2029 milestone for human-level AI. The International AI Safety Report 2026 does not use the word singularity, but its picture has the shape of a horizon. Its four scenarios for 2030 range from stalled progress to AI systems comparable to human remote workers, and it reports little expert consensus on which is most likely. If AI accelerates AI research, it warns, decades of progress could happen in years, creating extreme planning challenges for policymakers. In a forecasting study the report draws on, participants disagreed by several orders of magnitude on some questions. The horizon has not arrived. What has shrunk, on this evidence, is the span over which informed people can agree.
What a limit of knowledge implies
If the future beyond some point cannot be modeled, the first casualty is not optimism but planning. Friedrich Hayek argued in “The Use of Knowledge in Society” (1945) that the knowledge a planner would need is dispersed among many people, much of it local and situational, and never exists in one place. The singularity is the knowledge problem in its extreme form: the relevant knowledge does not yet exist anywhere. A regime of prior permission for AI – deciding in advance which systems may be built, by whom and for what – rests on exactly the kind of forecast the concept says we cannot make.
The strongest objection runs the other way. If we cannot see past the horizon, and what lies there may be irreversible, then prudence says slow down until we can see. This is a serious argument, and it is the one behind calls to pause frontier development. Vinge weighed it in 1993 and doubted it would work: the competitive advantage of the technology in economics, military affairs and even the arts meant, he wrote, that forbidding it would only ensure that someone else got there first. One need not share his fatalism to see the further point. Relinquishment is itself a forecast – that the slow path is the safe one – and it concentrates the decision in the few hands able to enforce it.
The liberal alternative is not indifference. It is a set of institutions built to learn quickly: liability for harm, prohibition of fraud, transparency about what systems can do and how they are used, many independent developers, and open models that let outsiders test claims. These are discovery procedures, not plans. They do not need to know the future in advance. They need to notice it as it arrives, and to correct course faster than a central authority could.
Not a point but a permeation
The singularity pictures change as an event. The AIciety model pictures it as a condition. The definition describes a society in which AI becomes part of how knowledge is produced, decisions are made and work is organized, and the last of the AIciety stages, Level 10, is permeation. Kurzweil’s merger of brains and cloud is the transhumanist version of that end state. The sociological version needs no nanobots. It arrives through delegation, through machines communicating with machines, through institutions reorganized around them, one decision at a time, with no single day on which the world changes.
This matters for the reflexive layer of AIciety as well. The singularity is not only a forecast about society; it is a story society tells about itself, and stories move money. The report notes data-center investments totaling hundreds of billions of dollars. Some of that money is, in effect, a bet on the horizon. A concept meant to mark the limit of prediction has become one of the forces shaping what happens.
What the song does with it
I’m Upping My P(doom) places the singularity in its second verse, which the lyricist osmarks describes as a parallel to the training-run imagery of the first, and which ends in a reference the annotation traces to K. Eric Drexler’s Nanosystems. The song’s singularity is therefore local and sudden, closer to a FOOM inside one machine than to Kurzweil’s gradual merger; it fuses the two ideas this Lexicon keeps apart. The third verse names von Neumann, linked only to his biography, so whether it nods to Ulam’s anecdote is left open. The video by mexicat, posted on YouTube on September 27, 2026, brought the scene to a wide audience. Sung, the singularity is not a date. It is the feeling of being inside the run.
Sources: Stanislaw Ulam, “John von Neumann 1903–1957”, Bull. Amer. Math. Soc. 64 (1958) · Vernor Vinge: The Coming Technological Singularity (1993) · Zoë Corbyn, interview with Ray Kurzweil, The Observer, 29 June 2024 (syndicated copy) · Ray Kurzweil: Don’t Underestimate the Singularity (2011) · International AI Safety Report 2026 · METR: Forecasting the Impacts of AI R&D Acceleration · F. A. Hayek: The Use of Knowledge in Society (1945) · osmarks: p(doom) song, objectively correct interpretation · Lexicon: Singularity

Leave a Reply