Kurzweil Scorecard: The Economy Compounded. The Experience Didn’t.

Kurzweil wrote, in 2005, that “by the end of the 2000s decade, going to a website will mean entering a visual-auditory virtual-reality environment” (The Singularity Is Near, ch. “Deflation … a Bad Thing?”). Twenty years later, Meta’s Reality Labs has burned roughly $83.6 billion trying to build that environment. Meta briefly announced the shutdown of Horizon Worlds on VR in March 2026, then reversed course inside 48 hours after user backlash; at its 2022 peak the flagship had fewer than 200,000 monthly users. You still go to a website by clicking a link in a 2D browser.

And yet in the same span, U.S. venture capital deal value ran from roughly $30 million (1974 dollars) in 1974 to $215.4 billion in 2024 across 14,320 deals. Kurzweil’s claim that by 2005 the number was “almost one hundred times” the 1974 figure “even during the 2000–2003 high-tech recession” (ch. “The Singularity as Economic Imperative”) was correct then and spectacularly more correct now. His economic skeleton was right. The flesh on it — the way people would actually live and work — is still mostly the one he thought would be extinct by now.

This batch of twelve predictions, from the economic and society chapters of The Singularity Is Near, is the sharpest test we have of that split. When Kurzweil reasoned about dollars, deals, and aggregate productivity, he was usually on-track or ahead. When he reasoned about what people would do and feel — retire, shop, socialize, go to a website — he was usually behind, and the gap is not closing.

The predictions

Six are about the economic engine: stock market updraft, venture deals, GNR (genetics, nanotechnology, robotics) growth outstripping the 1.7% government estimate, the knowledge economy crowding out everything else, the twenty-first century compressing twenty thousand years of progress, and the decade-long technology-adoption lag collapsing to half a decade. Three are about the shape of experience: websites becoming VR, experiences becoming “increasingly” virtual, and the post-Singularity dissolution of the human/machine boundary. Two are temporal anchors: the Singularity in the first half of the 21st century, and people refusing to retire at sixty-five because they’ll have “the bodies and brains of thirty-year-olds.” One is a warning: the genetics revolution that cures disease and aging also “create[s] the potential for new bioengineered viral threats.”

The economic machinery: mostly on-track, sometimes ahead

Start with VC, the cleanest win. Kurzweil’s 100x claim was made for a 2003 trough. Twenty years later even a down year crushes that multiple. The S&P 500, which he expected to enjoy a “strong long-term updraft” from the spreading recognition of exponential growth, returned about 10–11% annualized since 2005 with dividends reinvested; 2025 alone posted 17.88%. Two global financial crises (2008, 2020) did not derail the trend. Verified.

The shape of growth was messier. U.S. real GDP growth has averaged roughly 2% since 2005, not the acceleration one would expect if GNR were “greatly outstrip[ping] the 1.7 percent annual estimates used in government projections” (same chapter). Individual GNR sectors compounded fast — the cost of sequencing a human genome fell roughly seven-hundred-fold from 2005 to 2025, a trajectory Kurzweil documents in The Singularity Is Nearer — but the aggregate number refuses to move. Total factor productivity growth has in fact slowed. This is the Solow paradox in its 2020s form: you can see the AI and the sequencers everywhere except in the GDP statistics. Wrong mechanism: the sectors accelerated, but the reallocation of the rest of the economy didn’t, so the aggregate still hugs the long-run average.

The knowledge-economy prediction fares better. By 2024, information, professional services, finance, and health care account for well over half of U.S. GDP; software alone passed 9% of nonresidential fixed investment. Kurzweil’s “another couple of decades” deadline lands roughly now. Verified.

The claim that the twenty-first century will compress “on the order of twenty thousand years of progress at today’s rate” (ch. “The Intuitive Linear View Versus the Historical Exponential View”) is an endpoint claim. Kurzweil’s own post-hoc defense in The Singularity Is Nearer — 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” — suggests the law of accelerating returns, on the compute substrate at least, remains intact. Too early to call on the headline; on-track on the substrate.

The adoption-lag prediction is the most interesting miss. Kurzweil said the decade-long gap between leading-edge and trailing-edge technology use would shrink to half a decade within ten years. For smartphones, he was right. But broadband tells the opposite story. In 2024, approximately 98% of urban U.S. households had high-speed broadband access, versus roughly 64.4% of rural households. In 32 states, the urban–rural high-speed gap widened in 2024. Washington state showed 68% of urban users with 100/20 Mbps or better, versus 31% of rural users. That’s a structural gap federal BEAD dollars are only now beginning to close. Behind schedule for infrastructure; ahead for consumer devices. The prediction had no provision for the asymmetry.

The VR website that never was

The claim that “going to a website will mean entering a visual-auditory virtual-reality environment” by the end of the 2000s is the batch’s clearest failure, and the one Kurzweil has quietly walked back. In The Singularity Is Nearer (2024), he rewrites the timeline: “over the next couple of decades, brain-computer interface technology will become much more advanced. Ultimately this will allow full-immersion virtual reality that feeds simulated sensory data directly into our brains” — pushing the mechanism to the 2040s and the substrate from screens to BCIs.

The 2005 mechanism has been falsified in the marketplace. Meta’s Reality Labs has posted operating losses every year since 2020, cumulatively $83.6 billion through 2025, including a $6.02 billion Q4 2024 loss against revenue of roughly 1% of Meta’s total. Horizon Worlds failed to hit a revised target of 280,000 monthly actives — itself halved from 500,000 — and has accumulated about $1.1 million in lifetime consumer spend. The Horizon Worlds VR shutdown announced March 18, 2026 was reversed inside 48 hours after user backlash, but the direction is clear: Meta is pivoting its metaverse future to mobile, not VR.

The patent record confirms the quiet. Patents mentioning “virtual reality website” are a trickle: one a year in the 2010s, peaking at four in 2020, then back to one to three per year. Behind schedule, wrong mechanism: if immersive presence ever replaces the browser, the evidence points to BCIs or AR glasses, not head-mounted VR.

The broader claim that “experiences will increasingly take place in virtual environments” is more defensible. COVID-19 pushed telework from roughly 7% to a peak of 42% of American workers and settled near 25–30% hybrid. Discord, Roblox, Fortnite concerts, and always-on video are “virtual environments” in the colloquial sense if not the sensory-immersion sense Kurzweil meant. Verified, wrong mechanism.

The bioengineered threat that already arrived

The 2005 prediction that “the genetics revolution will … create the potential for new bioengineered viral threats” is one of the most prescient lines in the book. The Singularity Is Nearer makes this explicit: “The 2019–2023 coronavirus pandemic offers us a pale glimpse of what such a catastrophe could be like.”

Reality has overtaken the prediction. A December 2024 House select subcommittee report concluded COVID-19 “most likely” emerged from a laboratory incident at the Wuhan Institute of Virology, citing NIAID-funded gain-of-function work on coronaviruses — though the scientific origin debate remains unsettled, with Nature reporting in late 2024 that no close genetic relatives to SARS-CoV-2 have been found in archived WIV samples. Either way, the policy response has moved: a 2014–2017 federal gain-of-function funding pause, a May 2024 policy tightening oversight, and Executive Order 14292 on May 5, 2025 directing OSTP to rewrite it. Bills like S.81 (118th Congress) have tried to codify a permanent moratorium.

In our literature corpus, papers on gain-of-function research covering influenza, SARS, and coronaviruses grew from a handful per year before 2010 to 48 in 2021 and 34 in 2025. A 2014 paper, “Moratorium on Research Intended To Create Novel Potential Pandemic Pathogens” (mBio, 67 citations), and a 2015 paper, “The Reemergent 1977 H1N1 Strain and the Gain-of-Function Debate” (mBio, 126 citations), frame a twenty-year institutional argument that Kurzweil saw coming. Verified.

Retirement: the part he got backward

Kurzweil predicted people “will not seek to retire at sixty-five when they have the bodies and brains of thirty-year-olds” (ch. “The Singularity as Economic Imperative”). By the 2040s deadline we are not there. But something superficially similar is already true: in 2024, 19.5% of Americans aged 65 and older were in the labor force (23.4% of men, 16.2% of women), up from roughly 13% in 2000. The 65-to-74 cohort’s participation rose from 18.0% in 2004 to 22.4% in 2014, and has kept climbing.

The fine print inverts the mechanism. 38.3% of employed over-65s in 2024 worked part-time, versus 14.2% of those aged 55–64. BLS data and ProPublica reporting describe forced late-career work after job loss, thin retirement savings, and declining health coverage — not people with thirty-year-old bodies chasing purpose. Verified outcome, wrong mechanism.

The scorecard

What Kurzweil missed (and what he nailed)

A pattern comes into focus. Kurzweil’s aggregate financial claims landed — stock market updraft, VC compounding, knowledge-economy dominance, compute cost curves — because those are averages over enormous numbers of uncorrelated decisions, and averages smooth over human resistance. His human-experience claims almost always assumed that a better substrate (VR, BCIs, digital-human companions) would be adopted at its engineering speed. It wasn’t. Meta spent $83.6 billion trying to force the adoption curve and the curve stayed flat. The 2005 book systematically under-weighted the friction budget of a biological species: the nausea, the social awkwardness, the fact that a 2D browser is still easier.

The bioengineered-threat prediction is the opposite lesson. When Kurzweil described a dark scenario that required no human-adoption curve — just the steady compounding of biotech capability in the hands of a few thousand labs — he was nearly prescient to the year. The mechanism worked because nobody had to buy a headset.

Takeaway for R&D forecasting: when a prediction depends on a billion people changing their daily sensory interface, discount aggressively. When it depends only on capability plus a few thousand practitioners, take the timeline seriously.

Method note

Predictions and chapter attributions come from our extracted index of The Singularity Is Near (2005). Restatements are pulled from the full text of The Singularity Is Nearer (Viking, 2024). Patent and literature counts come from a 9.3M-document U.S. patent archive and a 357M-work scientific literature archive, queried by full-text search and ranked by citations. Macro data (S&P 500 returns, U.S. real GDP, VC deal value, broadband access, older-worker labor force participation) come from FRED, BEA, PitchBook/NVCA, FCC reports, and the BLS; links are available via the live session search log. Where a paper has a DOI or PMCID, we read beyond the abstract. Verdicts follow a six-level rubric (ahead / on-track / behind / wrong mechanism / too early / overtaken), and we prefer “wrong mechanism” to “behind” whenever the underlying phenomenon is happening via a path Kurzweil didn’t describe.