The Law of Accelerating Returns

This is the long essay that laid out the argument Kurzweil later expanded into The Singularity Is Near (2005). It is the canonical statement of the idea the wiki tracks as the Law of Accelerating Returns.

Summary

The core error is linear intuition. Kurzweil opens with a joke (reading the essay will earn “you” $40 trillion) and then states the thesis: long-range forecasts consistently underestimate future technology because they extrapolate the current rate of change in a straight line (the “intuitive linear” view) instead of on the curve history actually shows (the “historical exponential” view). His headline number is that the 21st century will not deliver 100 years of progress at today’s rate but something on the order of 20,000 years of it, because the paradigm-shift rate is roughly doubling every decade.

Why the curve bends upward. Evolution, biological and then technological, builds on its own products. Each stage creates better tools for building the next, so the returns of an evolutionary process accelerate. When a process becomes more effective, more resources flow into it, and the rate of exponential growth itself grows exponentially. He calls this a double exponential and sketches a simple model of it for computation.

Moore’s Law is the fifth paradigm, not the first. Price-performance of computing has grown exponentially through five successive technologies: electromechanical calculators (the 1890 census), relays, vacuum tubes, discrete transistors, and integrated circuits. Kurzweil predicted Moore’s Law would run out of room around 2019 as transistor features approached atomic scale, and argued that a sixth paradigm (three-dimensional molecular computing) would carry the trend forward, the way each earlier paradigm handed off to the next. He extends the same pattern to DNA sequencing, communication speeds, memory, and miniaturization (shrinking at roughly a factor of 5.6 per linear dimension per decade, by his count).

What “Singularity” means. He traces the word from mathematics and physics (a point of infinite value, a black hole’s event horizon) through John von Neumann’s remark about approaching “some essential singularity,” I. J. Good’s “intelligence explosion,” and Vernor Vinge’s 1993 NASA paper. His own definition: change so rapid and profound that it ruptures the fabric of human history. Mathematically there is no discontinuity, only very large finite growth; it looks like a break only from our present vantage.

Hardware, then software. Matching the brain’s raw computation is necessary but not sufficient. The rest of the argument is about software: scanning and reverse-engineering the brain region by region (likening it to the human genome project, which also looked impossible), and building neuromorphic models from what is learned. He addresses consciousness and the paradoxes of “uploading” a specific person without resolving them, and expects machines to pass the Turing test within the first decades of the century.

The merger. The most important implication, in his view, is not machines replacing humans but biological and nonbiological intelligence merging. Biological thought is fixed at roughly 10^26 calculations per second for all human brains; nonbiological intelligence grows at a double exponential and will dominate well before 2100. He insists that intelligence is still human, because it is derived from a human-machine civilization.

Economics. A long section argues that the 1990s boom was not a bubble: information technologies are the most deflationary force in history, so measured inflation understates real growth, and a Federal Reserve “speed limit” of about 3.5% growth is built on linear models that ignore bandwidth, MIPS and intellectual property. The $40 trillion joke pays off here: if markets priced the exponential view, he argues, equity values would roughly triple.

Dangers. He closes by engaging Bill Joy’s 2000 Wired essay “Why the Future Doesn’t Need Us.” He grants the risks (self-replicating nanobots, engineered pathogens, who controls the machines) but rejects broad relinquishment, arguing instead for narrow, fine-grained relinquishment plus defensive technology. He ends on a near-theological note: evolution is the purpose of life, and the Singularity is the goal of civilization rather than a danger to avoid.

Notable quotes

“It isn’t the case that we’ll experience 100 years of progress in the 21st century. Rather we’ll witness on the order of twenty thousand years of progress.”

“The Singularity is technological change so rapid and so profound that it represents a rupture in the fabric of human history.”

“Moore’s Law of Integrated Circuits was not the first, but the fifth paradigm to provide accelerating price-performance.”

“The Singularity then is not a grave danger to be avoided.”

How it connects

• Law of Accelerating Returns · Technological Singularity · Singularity — the concepts this essay names and defines.
• Moore’s Law · Gordon Moore · Exponential Growth — the five-paradigm argument.
• Vernor Vinge — whose 1993 paper Kurzweil credits for the modern use of “Singularity.”
• Transhumanism · Nanotechnology — the merger and the danger sections.
• Ray Kurzweil — author.