Kurzweil Scorecard: The Social Predictions Were Right About Speed, Wrong About Bodies

In 2005, Ray Kurzweil made a batch of predictions about how the next two decades of technology would rearrange daily life — how fast new things would diffuse, how we would experience the web, whether people would be cloned, whether babies would be designed, whether the underclass would still exist. Twenty-one years on, the batch scores unusually cleanly. The pattern is striking: where Kurzweil was predicting the behavior of information, he was right, often early. Where he was predicting the behavior of bodies — flesh, embryos, eyeballs — he was consistently wrong about the mechanism, and sometimes wrong about direction too.

Where we actually are

Adoption is faster than he predicted, by orders of magnitude.

Kurzweil wrote that “a decade from 2005, the lag from early-adopter to ubiquitous inexpensive adoption for information technologies will shrink to about five years” (ch. “The Criticism from the Rich-Poor Divide”). In The Singularity Is Nearer (2024), he notes that “one dollar buys about 11,200 times as much computing power, adjusting for inflation, as it did when The Singularity Is Near hit shelves,” framed as the engine behind compressing adoption curves.

The five-year number is now quaint. ChatGPT crossed 100 million monthly users in two months after its November 2022 launch — the fastest consumer adoption curve on record. Generative AI reached 39% U.S. household penetration in two years; the internet took five, personal computing twelve. Kurzweil set a floor that reality drove through almost immediately.

Full-immersion VR did not arrive via the retina.

This is the most specific, and most wrong, prediction in the batch. Kurzweil wrote that “early in the second decade of the twenty-first century, the Web will provide full-immersion visual-auditory virtual reality with images written directly to our retinas from eyeglasses or lenses and very high-bandwidth wireless Internet woven into clothing” (ch. “Government Regulation”). Almost every clause is wrong.

Mojo Vision, the best-funded attempt at a direct-view microLED AR contact lens, canceled its consumer program in early 2023 and laid off roughly 75% of staff. The state of the art is instead head-worn near-eye displays using pupil-replicating waveguides. Apple’s Vision Pro (2024) packs 23 million pixels across two 1.41-inch micro-OLED panels at 3,386 pixels per inch — but as a roughly 600-gram headset. U.S. patent grants claiming “waveguide near-eye display” architecture climbed from one patent in 2013 to 39 so far in 2025. Retinal-projection patents through contact lenses or eyeglasses barely register. US 12,585,094, granted March 2026, is typical of the dominant paradigm: an “anamorphic near-eye display apparatus” with a spatial light modulator, an input anamorphic lens, and an extraction waveguide bouncing light through partially reflective elements back to the eyebox. That is a pupil-replicating projector on your forehead, not light written to your retina.

The web was not the medium either — Meta Quest (roughly 80% of unit share) runs native apps in a closed store. Adoption has been weak. Apple sold an estimated 45,000 Vision Pro units in Q4 2025 versus 390,000 in all of 2024; the VR headset market shrank 14% year over year; Apple reportedly cut Vision Pro marketing by as much as 95% and paused a follow-on model. Reality has given us the parts of Kurzweil’s prediction that don’t depend on the body — 90-Hz panels, 34 pixels per degree, sub-millisecond tracking — and not the retinal writing, clothing networks, or ubiquity.

Human cloning did not happen. Something else absorbed the ambition.

Kurzweil wrote that “human cloning will occur soon once cloning technology is perfected for safety” (ch. “Human Cloning: The Least Interesting Application of Cloning Technology”), supporting it with the claim that contemporary somatic-cell nuclear-transfer methods using an “electric spark” produced high defect rates — the Dolly problem. The safety claim held. The prediction it supported did not.

No verified human reproductive cloning has ever been performed. SCNT for reproduction remains prohibited in most jurisdictions. U.S. patents mentioning “somatic cell nuclear transfer human” peaked at three grants in 2018 and have averaged one or two a year since. Instead, induced pluripotent stem cells absorbed nearly all the therapeutic ambition — Kurzweil’s own 2024 book describes iPSC cardiac therapy, iPSC-derived retinal transplants, and iPSC regenerative medicine across “a broad variety of biological tissues.” He called cloning “the least interesting application” in 2005; the field appears to have agreed, and skipped it.

Designer babies arrived early, by a route Kurzweil missed.

The 2005 prediction: “technology for designer babies will not be available for another ten to twenty years” — so 2015 to 2025 — and the revolution would be “slow” and “not a significant factor in the twenty-first century” (Ned Ludd / Ray dialogue).

In November 2018, He Jiankui announced that twin girls in China — Lulu and Nana — had been CRISPR-edited at the embryo stage to knock out CCR5, with a third child (Amy) confirmed in 2022. He served three years for illegal medical practices. In 2025 he announced a new lab targeting Duchenne muscular dystrophy. The direct-edit technology landed inside Kurzweil’s window, via CRISPR rather than the homologous-recombination methods he described.

The commercial path took a different shape: polygenic embryo screening. Orchid Health launched at $2,500 per embryo in late 2023, offering adult-disease risk scores on IVF embryos. Genomic Prediction (LifeView) and MyOme offer overlapping services. Herasight, out of stealth in 2024, screens for predispositions including adult disease, height, IQ, longevity, and depression. The American College of Medical Genetics called the practice premature in 2024. This is selection rather than modification — and Kurzweil did not predict it. The “not significant this century” clause still holds — births from polygenic screening remain a negligible fraction of global births — but the technology is commercially available ahead of schedule.

Productivity and inflation measurement caught up with him.

Kurzweil’s claim that “official productivity statistics substantially understate real gains because they fail to capture quality improvements and entirely new product categories” (ch. “Economic Imperative”) drew on Mark Bils and Pete Klenow’s 2001 American Economic Review work, which estimated that BLS quality adjustments captured only about 40% of actual quality upgrading — so roughly 2.2 percentage points per year leaked into inflation as phantom price growth. That paper has since become canonical, extended by Klenow’s later work on missing growth from creative destruction.

Erik Brynjolfsson’s GDP-B framework, in American Economic Journal: Macroeconomics (October 2025), goes further. Using incentive-compatible choice experiments, his group estimated the median U.S. Facebook user would need $48 to give up the service for one month. The U.S. incremental welfare gain from the internet between 2002 and 2011 averaged about $38 billion per year — roughly 0.29% of annual GDP that traditional accounting does not see. In his 2024 book, Kurzweil makes the same point concretely: a $900 chip delivered ~800,000 computations per second per dollar in 1999 and ~58 billion in 2023. Quality improvement GDP barely registers. Kurzweil did not originate the claim, but he was right about the direction and its persistence.

Extreme poverty is collapsing, but “the underclass” is not disappearing.

Kurzweil wrote that “given the wealth-creation potential of GNR technologies, the underclass will largely disappear over the next two to three decades” (ch. “on Work”). The window closes between 2025 and 2035.

World Bank data (March 2026 update) puts the global extreme-poverty rate at 10.4% in 2024 — about 847 million people at under $2.15 a day in 2017 dollars — down from roughly 37% in 1990. East Asia’s extreme poverty fell 95% between 1990 and 2013. But the March 2026 update actually revised the Middle East / North Africa / Pakistan region upward to 14.4%. “Largely disappears” is not the right verb for a decline toward — but not through — a single-digit floor. Direction right. Speed and geography off.

The scorecard

What Kurzweil missed (and what he nailed)

The information predictions in this batch outperformed. Adoption curves compressed faster than the five-year floor he set. Productivity and inflation measurement evolved in the direction he argued for. These are cases where the substrate — chips getting cheaper by his 11,200x factor — propagates directly into the predicted outcome, because the outcome is information diffusion.

The body predictions all broke. Full-immersion VR did not arrive via retinal projection. Human cloning did not arrive at all. Designer babies arrived, but through embryo selection rather than germline edit. Extreme poverty has fallen sharply, but not to a vanishing point, and not via the GNR path Kurzweil emphasized — most of the gains came from Chinese industrialization and mobile-money-enabled market access in Africa.

The pattern is not that Kurzweil was too optimistic. It is that he underestimated how much human flesh would impose its own timelines on anything touching it. Chips scale exponentially. Retinas do not. Embryos do not. Poverty does not. For anyone betting on when a wearable, an implant, a gene therapy, or a behavior change goes mainstream, the information-curve intuition underestimates by a factor that grows with biological contact. If the pattern holds for the Singularity date itself, the pure-information parts (model capability, compute cost, algorithmic complexity) will arrive early; the merging-with-biology parts (neural interfaces, uploads, radical life extension) will arrive late. The 2045 anchor may split.

Method note

For each prediction we checked U.S. patent filings by year for the relevant claim area, high-citation scientific literature, World Bank poverty data, economic-measurement research from the authors Kurzweil cited, and news coverage of specific launches, failures, and clinical events. Patent claim text was read directly, not just counted. All quoted numbers come from sources accessed this session. The 2024 restatements are from The Singularity Is Nearer; the 2005 predictions are from The Singularity Is Near.