Kurzweil Scorecard: The Footnotes That Bit Back

Most Kurzweil predictions are about the curve going up and to the right: faster chips, cheaper genomes, more neurons emulated per dollar. The harder ones sit buried in the rebuttal chapters of The Singularity Is Near, where Kurzweil defends the Law of Accelerating Returns against its loudest critics. Penrose and Hameroff, he wrote, were wrong about microtubules. Jaron Lanier was wrong about software getting slower. Quantum entanglement would secure our communications. Computers that run hot would beat computers forced to stay cold.

This batch is those footnotes. Twenty years later, several of the critics got the last laugh.

The predictions

The ten claims come from five different chapters but share a posture: Kurzweil, in 2005, drawing lines against people who said the brain was different, software was broken, or physics had ceilings he was ignoring. The interesting question is not whether he was right about the overall arc — the arc has been kind to him. It is whether the specific bricks holding up his argument have survived the last two decades.

Where we actually are

The microtubule line he’s still defending. Three of the ten predictions are variants of the same calculation. Kurzweil wrote that “if microtubules increased neuron complexity by a factor of 1,000, reaching brain-level computational capacity would be delayed by only about nine years” (ch. “The Criticism from Microtubules and Quantum Computing”), and that factors of a million or a billion would delay things by seventeen or twenty-four years respectively. The math was a shrug at Penrose and Hameroff: even if you are right, I am not much wrong.

In The Singularity Is Nearer (2024), he is more dismissive. The Orch-OR theory, he writes, “has so far failed to gain much acceptance within the scientific community… I have not seen any compelling evidence that quantum physics is needed to explain functional consciousness in the brain.”

That sentence did not age well. In September 2024, Mike Wiest’s lab at Wellesley published in eNeuro that rats treated with the microtubule-binding drug epothilone B took significantly longer to go under isoflurane anesthesia — a result consistent with anesthetics acting on microtubule quantum dynamics. A 2025 review in Neuroscience of Consciousness (doi:10.1093/nc/niaf011) concluded that “functionally relevant quantum effects occur in microtubules at room temperature” — precisely the decoherence objection Max Tegmark raised in 2000 that Kurzweil had leaned on. Our literature index shows microtubule-quantum papers jumping from ~30 per year through 2022 to 444 in 2025 and 365 already in 2026 — a ten-fold surge in a field Kurzweil treated as settled.

This does not overturn Kurzweil’s delay arithmetic. His nine- or twenty-four-year buffer still likely holds. But the confident 2024 footnote that quantum effects in the brain are a fringe concern now reads as a premature closing argument.

The software-responsiveness complaint he swatted away. Kurzweil, in 2005, quoted Jaron Lanier’s lament about software bloat and responded that “software from twenty years earlier was not qualitatively or quantitatively better than current software, despite nostalgia about responsiveness” (ch. “The Criticism from Software”). He argued Lanier was romanticizing the past.

Dan Luu’s 2017 latency measurements, using two high-speed cameras, found a 1981 Apple IIe registered keypresses on its display in 30 milliseconds. A PowerSpec g405 with a 7th-generation Intel i7 and a 60 Hz LCD took 200 milliseconds for the same test. A 2017 iPad needed eleven distinct software layers to surface a single keystroke where the Apple IIe needed a keyboard scan loop running at 50 kHz. That is not nostalgia. That is a 6.7× regression on the most basic measure of whether software responds when you press a key.

Niklaus Wirth’s 1995 warning — “software is getting slower more rapidly than hardware becomes faster” — was the subject of a widely read 2024 IEEE Spectrum piece titled “Why Bloat Is Still Software’s Biggest Vulnerability.” Lanier’s complaint was a measurement problem Kurzweil did not take seriously.

The quantum-entanglement encryption claim that was technically right, narrowly. Kurzweil, in the astrobiology chapter, stated that “commercial encryption products based on quantum entanglement already existed” in 2005. MagiQ Technologies shipped the first commercial QKD system in 2003; ID Quantique followed in 2004. Both were prepare-and-measure BB84 systems, not entanglement-based. Entanglement-based QKD (Ekert’s E91 protocol, 1991) has remained commercially marginal. Our patent corpus shows 21 U.S. grants with the explicit “quantum entanglement + encryption/cryptography” combination through 2024, and then a jump to 13 in 2023–2025 alone, including US 12,355,869 (System and method for quantum key distribution with entangled photon pairs, July 2025), US 12,386,236 (Korea Electronics, multi-wavelength entangled photon pair device, August 2025), and US 12,598,062 (2-qubit multi-user QKD, April 2026). The entanglement-based product category Kurzweil implied existed in 2005 is really an early-2020s phenomenon.

The weapons-simulation claim that was conservative. Kurzweil wrote that “weapons will increasingly be developed and tested in simulation to shorten development cycles that otherwise leave systems obsolete by deployment” (ch. “on Warfare: The Remote, Robotic, Robust, Size-Reduced, Virtual-Reality Paradigm”). By 2010s, he said. He undershot. In December 2024, the F-35 Program Office opened a dedicated digital twinning facility. Waymo — not a weapons program, but the closest civilian analog — runs 20 million simulated driving miles per day, matching in a day what it accumulated on real roads from 2009 to 2018. Patent US 12,313,387 (May 2025, Device and method for simulating a dynamic munition-related environment for a projectile) describes a physical simulation rig where a real projectile is flown through a digitally generated flight environment, blurring the line Kurzweil drew between simulation and test.

The archivists are still at it, and Gordon Bell died still logging. Kurzweil named Gordon Bell, DARPA, and the Long Now Foundation as actively building archival systems. Bell died in May 2024 at 89, having continued to feed data into MyLifeBits through the end of his life. The Long Now Foundation’s Rosetta Project now holds over 100,000 pages covering more than 2,500 languages, with physical Rosetta Disks distributed to new institutions in 2024. The institutions he named kept running — but MyLifeBits never generalized into consumer lifelogging. We all lifelog now, just into Google’s servers, not Gordon Bell’s file system.

The thermodynamic ceiling is getting tested. Kurzweil’s claim that “allowing the ultimate computer to run hot could increase its computational power by as much as 10⁸-fold over the cold-computing limit” (ch. “The Criticism from Malthus”) was a rebuttal to critics invoking thermodynamic limits. A Nature Physics 2025 paper experimentally probed Landauer’s principle in the quantum many-body regime. London-based Vaire Computing taped out its first reversible-computing prototype in Q1 2025, aimed at an AI-inference chip for 2027 that recovers energy from arithmetic circuits rather than dissipating it. Kurzweil’s specific 10⁸ claim remains a theoretical upper bound, but the field he invoked to defend the claim is now producing silicon.

The two untestable axioms. Two predictions — “digital computing can be functionally equivalent to analog computing, but analog cannot simulate all functions of a digital computer” (ch. “Is the Human Brain Different from a Computer?”) and “computation requires a nonlinearity; weighted sums alone are insufficient” (ch. “Trying to Understand Our Own Thinking”) — are mathematical facts Kurzweil deployed as rhetorical weapons, not falsifiable predictions. The second has become load-bearing in modern deep learning: every transformer activation function is a specific implementation of the nonlinearity requirement Kurzweil flagged in 2005. Twenty years later, a well-read twelve-year-old knows why ReLU exists.

The scorecard

What Kurzweil missed (and what he nailed)

Two patterns fall out of this batch. First, Kurzweil is consistently strongest where he is forecasting engineering trend lines and weakest where he is adjudicating physics edge cases that non-engineers care about. The weapons-simulation call was correct and early; the dismissal of Lanier’s latency complaint was the opposite. The user-facing annoyance of software that takes 200 milliseconds to echo a keystroke is not solved by cheaper transistors.

Second, the microtubule saga is a cautionary tale about over-committing to a rebuttal. Kurzweil could have written, in 2005 and again in 2024, “if Penrose-Hameroff turn out to be right, my timeline still works and here is the math.” He did, exactly — the delay calculations in this batch are that hedge. But in 2024 he volunteered a stronger claim, calling Orch-OR unsupported. Eleven months later, a Wellesley anesthesia lab published the first behavioral experiment consistent with microtubule-level quantum effects. The math still holds. The posture does not.

The pattern, if there is one, is that Kurzweil forecasts hardware well, software poorly, and physics controversially. The Law of Accelerating Returns has survived. Some of its defensive footnotes have not.

Method note

We scored this batch by first counting the shape of the evidence — patent grants mentioning quantum key distribution jumped from 15 per year in the mid-2010s to 75 in 2025; papers on microtubule quantum effects went from roughly 30 per year through 2022 to 444 in 2025 — and then reading the actual inventions and findings underneath those counts. Patent numbers cited above were retrieved from a 9.3-million-document U.S. patent index. Paper citations came from a 357-million-work OpenAlex literature mirror. Web sources covered the Wellesley anesthesia experiment, Dan Luu’s input-latency measurements, Vaire Computing’s reversible-computing prototype, and the F-35 digital twinning facility. Every number in this scorecard was surfaced by a query we ran today or a source we fetched today.