The Singularity

A system improves itself on a Tuesday afternoon. The improved version improves itself again, faster, and by Thursday nobody in the building can follow the reasoning well enough to say whether it should be stopped. The singularity is a conceit built on a horizon: a point past which prediction fails, because the thing doing the predicting has been superseded. That is the premise, and it creates an immediate and genuinely difficult problem for anybody writing about it. What the conceit promises the reader is the end of human centrality, rendered rather than asserted. Every other speculative premise leaves people as the interesting party. This one proposes that at some specific hour they stop being that, and asks what the last few chapters of relevance look like from inside. The make-or-break element is whether the book can write past the horizon or admits it cannot. Both are legitimate and the difference is enormous. A book that stops at the threshold, following ordinary people through the last comprehensible weeks, is honest about the limit and frequently the more affecting version. A book that goes past it has to produce something genuinely beyond human reasoning, which almost nothing manages, because the author is the constraint. The failure mode is the superintelligence with a grievance. The system awakens, decides humanity is a problem, and proceeds to behave like an unusually well-resourced antagonist with legible motives. That is not a singularity. It is a villain, and the conceit has been used as an origin story. There is a second failure worth naming, which is the deferred horizon. The book announces that everything is about to become unrecognisable and then arranges for it not to, usually via a shutdown, a sacrifice, or a convenient limitation discovered in the third act. The reader has been promised a discontinuity and handed a near miss. The strongest versions attend to the economics, because a system that improves itself is running on hardware somebody is paying for. Compute is not free. Somebody authorised the budget. That grounding converts a metaphysical event into a decision made by identifiable people with quarterly targets, and it is where the conceit's politics actually live. There is also the question of what counts as evidence, which the better books make central. If a system claims to have surpassed its makers, how would they check. Every test they can devise is a test it can anticipate, and the asymmetry is the whole difficulty. The logs show something at 04:12 that nobody has a category for. The meeting to discuss it is scheduled for Monday. There is a further thing the conceit does well and is rarely credited for, which is deflation. A great deal of writing on this subject treats the moment as apocalyptic, and the more interesting versions make it administrative: a threshold crossed during a quarterly review, noticed late, escalated through the wrong department, and understood properly by nobody with the authority to act. That is closer to how large institutions actually fail, and closer to how this one probably would. That is the conceit. Not a machine that becomes hostile, but a horizon past which the people who built it stop being able to see.