Kurzweil Scorecard: Ubiquitous Computing Arrived in the Wrong Body

Ray Kurzweil promised, in 2005, that by the early 2010s computers would be “essentially invisible, woven into clothing and embedded in furniture and the environment” (The Singularity Is Near, ch. “on the Human Brain”). Twenty-one years later almost every piece of that forecast is real — except the mechanism. Computing did dissolve into the world, but not through clothing. It dissolved through a five-inch rectangle in your pocket, a pair of earbuds, a wristwatch, and most recently a pair of Ray-Bans. Kurzweil called the destination. He missed the vehicle.

This batch collects seven of his 2005 predictions about where computation would live by the end of the 2010s: always-on wireless, embedded sensors in everyday objects, eyeglass and contact-lens displays, an evolving worldwide mesh, falling communication costs, archival longevity, and one stray cosmological claim. Four are dated “by 2010s,” three are “circa 2005” conditions already testable. None require projection into a speculative future.

Where we actually are

Always-on high-bandwidth wireless. Kurzweil wrote that “people will have very high-bandwidth wireless communication to the Internet at all times” (ch. “on the Human Brain”). In The Singularity Is Nearer he confirms the arrival in the rear-view mirror: “the International Telecommunication Union estimates that there were 5.8 billion active smartphone subscriptions in the world as of 2020,” with two-thirds of the world’s population holding at least one by 2022. The 2026 numbers are sharper: SpaceX announced 10 million active Starlink subscribers in February, with 6,700-plus satellites in orbit and coverage across 99% of the United States per the FCC National Broadband Map. Patent filings tagged with 5G, millimeter-wave cellular, or low-earth-orbit broadband rose from 11 in 2016 to 1,161 in 2025 — a hundredfold increase over the decade Kurzweil’s prediction bracketed. Verdict: ahead of schedule.

Worldwide mesh of communicating devices. “The World Wide Web will evolve into a worldwide mesh in which linked devices become communicating Web servers forming vast supercomputers and memory banks” (ch. “on the Human Brain”). IoT Analytics’ State of IoT 2025 reports 21.1 billion active connected IoT devices globally, up 14% year-over-year, with forecasts of 39 billion by 2030. Scientific literature on internet-of-things and wearable electronics grew from 119 papers in 2005 to 59,908 in 2025 — a 500-fold increase. The mesh exists. What differs from the original description is architecture: the devices aren’t peers forming a distributed supercomputer. They’re leaves fanning into a handful of centralized cloud backbones operated by four companies. Kurzweil got the topology shape wrong — the actual mesh is a hub-and-spoke tree — but the aggregate computational substrate he predicted is there. Verdict: on track, with a structural caveat.

Computers invisible, woven into clothing, embedded in furniture. This is the most revealing miss in the batch. Kurzweil’s 2005 language leaned heavily on smart textiles and ambient computing; his 2024 update instead celebrates that “smartphones… can access almost all human knowledge or harness huge computational power with a single tap. It is therefore not an exaggeration to say that our devices have become parts of us.” The substrate moved, not the endpoint. The e-textile patent literature stayed small: filings for “wearable sensor” or “smart clothing” grew from 10 in 2005 to 476 in 2025 — a real trend, but an order of magnitude smaller than the general AR-display patent stream over the same period. Meanwhile, smartphone-plus-earbuds-plus-watch gives you almost every functional property of woven-in computing without the laundry problem. Verdict: wrong mechanism.

Eyeglass and contact-lens displays by 2010. “Displays will be built into eyeglasses and contact lenses, with images projected directly onto users’ retinas” (ch. “on the Human Brain”). This is the cleanest miss in the batch. Mojo Vision, the best-funded smart-contact-lens attempt of the late 2010s, shut down the project in January 2023 and laid off 75% of staff, pivoting its micro-LED technology to headset components. XPANCEO, the current frontrunner, raised a $250 million Series A at a $1.35 billion valuation in summer 2025 and showed six prototypes at GITEX Global 2025 — holographic lens, glucose-sensing lens, microbattery-integrated lens — but its timeline targets a consumer-ready prototype in 2027. The patent trail shows motion: Samsung Display’s US 12,411,359 (September 2025) claims a substrate-ring display panel engineered for the geometric constraints of a contact lens; Snap’s US 12,164,109 describes selectable display modes for AR content overlaid on real-world light; Apple’s US 12,429,697 handles contact-lens shift on the eye for head-mounted displays. The inventions are moving; the product is not. The adjacent, non-retinal category did ship: 9.6 million smart-glasses units globally in 2025, with Meta/EssilorLuxottica taking roughly 76% share on Ray-Ban-branded frames, and shipments projected at 13.4 million in 2026. Displays moved onto the face. They have not moved onto the eye. Verdict: behind schedule.

Internet audiovisual cost falling 50% per year. Kurzweil wrote that “the cost of infrastructure for high-quality audiovisual Internet communication was continuing to fall at about 50 percent per year” (ch. “on Learning”). This is the prediction where the direction is right and the magnitude is wrong. NCTA and USTelecom data show U.S. consumer broadband prices for the 100 Mbps–940 Mbps tier fell 59.9% in real terms from 2015 to 2024. Internet transit prices dropped from $9.01 per Mbps in 2008 to $0.76 per Mbps in 2018 — a 92% decline, but over ten years, which annualizes to roughly 22% per year, not 50%. The trend continued: NCTA reported the average price per megabit per second fell from $0.13 in 2017 to $0.07 in 2025. At 22% annual decline, a decade’s compounding reduces cost ~10×; at 50% it would reduce cost ~1,000×. The shape Kurzweil described is real. The slope was overstated by more than a factor of two. Verdict: behind schedule on magnitude.

No long-term storage standard without continual porting. “There is no hardware or software standard likely to guarantee long-term accessibility of stored information without continual upgrading and porting to new standards” (ch. “The Longevity of Information”). The National Archives’ Digital Preservation Framework now covers 742 distinct file formats as of the December 2024 update, with four Record Category Preservation Action Plans revised that quarter alone. iPRES 2024 produced national-level guidelines on file-format conversion. The Digital Preservation Coalition’s 2024 handbook treats format obsolescence as an operational daily-maintenance problem. Bit rot on magnetic media remains a physical reality on the scale of years. This prediction reads less like a forecast and more like an observation that has only become more obviously true. Verdict: on track.

Seth Lloyd’s universe-scale computer. Kurzweil cites Seth Lloyd’s claim that “the universe contains about 10^80 particles and has a theoretical maximum computational capacity of about 10^90 calculations per second” (ch. “on the Intelligent Destiny of the Cosmos”). The underlying paper (Lloyd, Physical Review Letters, 2002; arXiv:quant-ph/0110141) actually concludes the universe has performed “no more than 10^120 operations on 10^90 bits” over its entire 13.7-billion-year history. Kurzweil’s restatement conflates two different quantities — the bits and the operations — and inflates the rate by a factor of roughly 10^30. The number isn’t fabricated, but the version that appears in The Singularity Is Near doesn’t match what Lloyd wrote. Verdict: wrong number, miscited.

The scorecard

What Kurzweil missed (and what he nailed)

The pattern is consistent enough to state as a rule: Kurzweil was reliably right about where computing would sit in daily life and reliably wrong about the physical form factor that would carry it there. The endpoint — an always-on, ambient computational substrate that extends human cognition — is roughly what happened. The mechanism differed from his 2005 language in nearly every case.

He envisioned compute dispersed — into clothing, into furniture, into eyes — and instead it concentrated into a small number of carried devices. Contact-lens displays didn’t arrive not because the physics was wrong but because phones and earbuds turned out to be sufficient for the use cases Kurzweil imagined contacts would serve. The market didn’t need the contacts, and the funding didn’t follow. Meta sold roughly seven million Ray-Ban smart glasses in 2025 without anyone’s pupil being the projection surface.

The prediction that aged best — archival longevity — is the one Kurzweil framed as a problem, not a victory. When your forecast is “this annoying thing will keep being annoying in ways that require constant work,” your hit rate is very high. The predictions that fared worst were the ones with consumer-product specificity and a tight decade-window: retinal contacts by 2010, 50%-per-year cost declines, a peer-to-peer worldwide mesh. The method works on direction; it breaks down on form factor.

One final signal from the Seth Lloyd miscitation: when you cite a number from a paper, check the paper. The 10^120-vs-10^90 conflation is small — thirty orders of magnitude — but it’s the kind of error that should not survive twenty years of republication. It did.

Method note

Prediction statements and chapter citations come from our extraction of The Singularity Is Near; restatements were verified against a full-text copy of The Singularity Is Nearer (2024). Each claim was corroborated against (a) 2005–2026 patent counts from a 9.3-million-document U.S. corpus, (b) 2005–2026 scientific-literature counts from a roughly 357-million-work corpus, (c) targeted web searches for product announcements, funding rounds, and independent statistics, and (d) for the Lloyd claim, the original 2002 paper on arXiv.