Event horizon, singularity and escape velocity

People throw these words around about AI without saying what they mean. Here's each one in plain English, and the piece that ties them together: recursive self-improvement.

01The four words

Four borrowed words

Recursive self-improvement (RSI)AI that helps build better AI, which builds better AI again. Each loop is faster than the last.
SingularityThe point where that loop runs so fast that the future becomes impossible to predict from where we stand.
Event horizonIn physics: the boundary around a black hole that nothing, not even light, can come back from. In AI: a point of no return.
Escape velocityIn physics: the speed needed to break free of gravity for good. In AI: the point where progress stops needing a push and keeps going on its own.

The idea is old. In 1965, the mathematician I. J. Good wrote that an "ultraintelligent machine" could design even better machines, setting off an intelligence explosion. He added that the first such machine would be "the last invention that man need ever make", provided it is docile enough to tell us how to keep it under control. In 1993, the writer and mathematician Vernor Vinge called the moment the technological singularity.

RSI is the engine. The singularity is where the road goes off the map. The event horizon is the last exit.

02The four event horizons

In physics there's one event horizon. In AI, I see at least four, and we could cross them at different times.

Four points of no return

1. PrevalenceWhen anyone can build a powerful model or super-agent, from anywhere, for anything. When a code form of a nuclear bomb can come out of a dorm room and no system can catch them all.
2. CapabilityWhen AI's intelligence has grown past our capacity to manage or control it. We become observers, watching a higher intelligence find its own balance for running the world as it sees fit.
3. UtilityWhen useful inventions grow so fast that every human need is met and the environment secured. Inventions arrive faster than needs are felt. Things for needs we don't know we have yet. Eventually, death cured.
4. ScienceWhen RSI pushes AI to solve hard problems across all of science, even ones with no practical use yet, in a mad, endless loop.

Two of these are dangers (prevalence, capability). Two are promises (utility, science). The uncomfortable truth is that they're powered by the same engine. You don't get the cure for death from RSI without also getting the dorm-room bomb risk from RSI. The choice isn't whether to cross. It's which horizon we reach first, and whether the systems in Rogue AI are in place before the dangerous ones.

03How RSI changes the timeline

Three speeds of progress

Humans improve AISteady steps, paced by people.
AI helps humans improve AIFaster steps. This is already starting: labs use their own models to write code and run experiments.
AI improves AI by itselfEach step shortens the next. This is the loop the word "explosion" refers to.

The key point: with RSI, the time between generations shrinks. If the gap between today's model and the next is a year, and each generation halves the gap, you get many generations in the following year. Supervision built for yearly releases won't keep up with weekly ones.

04The short version

RSI is AI improving AI. It's the engine behind every "explosion" scenario.
The singularity is where prediction breaks down; the event horizon is the point of no return.
There are four horizons: prevalence and capability (dangers), utility and science (promises). Same engine, so the order matters.

05Sources

Good, I. J. (1965). Speculations Concerning the First Ultraintelligent Machine. Advances in Computers 6.
Vinge, V. (1993). The Coming Technological Singularity. VISION-21 Symposium, NASA.

All posts

Appointment setting3

Copywriting3

AI services2

Deep Think AI13

Get the next markup by email.

One real piece of copy per issue, rewritten line by line: a cold email, a web page or a newsletter.