Kurzweil Scorecard: Auditing the Facts, Not the Forecasts

Most of The Singularity Is Near is about the future. A smaller, quieter
layer of the book is about the present — the factual claims Kurzweil laid
down as the foundation on which his predictions rested. Batch 42 of the
prediction set is an unusual one: ten “circa 2005” claims Kurzweil asserted
as already true when he finished the book. Not forecasts. Footings.

Twenty-one years later, the verdict is surprisingly mixed. Some of the
factual load-bearing statements in the book aged worse than some of the
predictions. The headline number Kurzweil used to describe the human
body — ten trillion cells, one hundred trillion microbes — is wrong in
both directions. The anticancer drug he said was “already being produced
in the milk of transgenic goats” turned out to be something different from
what got approved. A few of the medical-device claims, by contrast,
undersold how far the technology would travel.

The predictions

These are all “circa 2005” claims pulled from four chapters: “Who Am I?
What Am I?”, “on the Human Body”, “Chapter 8: Promise and Peril of
GNR”, and “Chapter Nine: Response to Critics”. They are the factual
scaffolding for arguments about merging with machines, reprogramming
bodies, and surviving our own technology. If the scaffolding sags, the
argument wobbles.

Where we actually are

The body’s cell count is off by a factor of three. Kurzweil wrote
that “a biological human has about ten trillion cells with the person’s
own DNA” and that the digestive tract holds “about one hundred trillion
microorganisms” (ch. “Who Am I? What Am I?”). The 10:1 microbe-to-cell
ratio was orthodox in 2005 — it traced back to a 1972 estimate from
Thomas Luckey that had been repeated uncritically for decades. In 2016
Ron Sender, Shai Fuchs, and Ron Milo published a careful re-derivation
in PLOS Biology (doi: 10.1371/journal.pbio.1002533) and found the real
numbers: approximately 3.0 × 10¹³ human cells and 3.8 × 10¹³ bacteria in
a 70-kg reference adult. A 2023 PNAS paper by Hatton and colleagues
(doi: 10.1073/pnas.2303077120) compiled updated figures across 1,200
cell types and arrived at roughly 36 trillion human cells. Kurzweil was
low by three-to-four-fold on human cells and high by nearly threefold
on bacteria. The ratio he cited — and hung arguments about bodily
complexity on — isn’t 10:1. It’s closer to 1.3:1.

This is not Kurzweil’s error so much as the field’s error that Kurzweil
repeated. But it matters, because the “outnumbered by our own bacteria”
framing became a rhetorical pillar for later arguments about symbiosis,
reprogramming, and nanomedicine.

The transgenic-goat drug was real, but misidentified. Kurzweil
described “a promising anticancer antiangiogenesis drug, aaATIII,” as
already being produced in goat milk (ch. “Why Is Cloning Important?”).
The goat-milk technology was genuine: GTC Biotherapeutics (formerly
Genzyme Transgenics) did engineer goats to secrete recombinant human
antithrombin in their milk. But the product that actually reached patients
is ATryn, approved by the European Medicines Agency in 2006 and the FDA
in February 2009 — as an anticoagulant for hereditary antithrombin
deficiency in patients undergoing surgery or childbirth. The antiangiogenic
form Kurzweil referenced, “aaATIII,” is a cleaved/latent conformation
studied in preclinical oncology but never commercialized. A 2009 review
in Clinical and Applied Thrombosis/Hemostasis covers the approval history
in detail. The chassis (transgenic goats) worked. The indication
(anticancer antiangiogenesis) is not what shipped.

Deep brain stimulation was, and is, real — but the claim about
multiple sclerosis and cerebral palsy oversold it. Kurzweil’s two DBS
claims (ch. “on the Human Body”) split along a clean line. Subthalamic
and ventral-posterior-nucleus stimulation for Parkinson’s disease was FDA-
approved (Medtronic Activa) in 2002 and has compounded since: the Medtronic
Percept PC closed-loop system reached the U.S. market in 2020, and the
Percept RC rechargeable implant in 2023. Literature-wide, papers on DBS
for Parkinson’s and tremor grew from 121 in 2005 to roughly 700 in 2025,
and granted U.S. patents on adaptive, closed-loop neuromodulation continue
to accumulate — e.g. US 12,415,076 (“Multidisciplinary design optimization
of neuromodulation systems,” granted September 2025), which uses a
wearable inertial-sensor system to iteratively tune stimulation parameters
in real time. In The Singularity Is Nearer (2024), Kurzweil himself
describes DBS as “a very primitive version of this today with certain
parts of a brain — for example, when treating essential tremor or
Parkinson’s disease” — a fair characterization.

The thalamic-DBS claim for tremors in multiple sclerosis and cerebral
palsy is a different story. A 2020 review in Brain Sciences (PMC7761254)
and a 2022 connectomic analysis in Brain Communications find that
roughly half of MS-tremor patients achieve any durable benefit from VIM
DBS, only about 30% reach >50% improvement, and outcomes are frequently
washed out by disease progression. Cerebral-palsy DBS remains a case-
series indication, not a standard-of-care therapy. The device existed;
the effectiveness didn’t generalize.

The 20-reservoir drug chip was real, and then it stalled. Kurzweil
cited a design with “up to twenty substance-containing reservoirs that
can release cargo at programmed times and locations in the body” (ch.
“on the Human Body”). That device — the MicroCHIPS implant developed
by Robert Langer and Michael Cima at MIT — matured through the 2000s
and reached a first-in-human trial in 2012: Farra et al., Science
Translational Medicine 4(122):122ra21, implanted a 20-reservoir chip
subcutaneously in eight postmenopausal women and delivered daily doses
of teriparatide for up to twenty days, with pharmacokinetics and bone-
marker responses matching daily injections. Kurzweil’s factual claim is
confirmed. What stalled was the commercialization: Microchips Biotech
pivoted repeatedly, was later absorbed into Daré Bioscience’s pipeline,
and the 20-reservoir programmable implant class never displaced the
simpler drug-eluting implants and subcutaneous auto-injectors that filled
the adjacent niche. The design existed. The market didn’t form.

Hepatocyte nanoparticle delivery was real, and the field pivoted.
Kurzweil invoked the Yamada et al. 2003 Nature Biotechnology paper
(doi: 10.1038/nbt843) describing hepatitis-B-derived “L particles” as
hepatocyte-targeted delivery vehicles. That paper is real and correctly
dated. The interesting part is what happened next: the L-particle
chemistry didn’t become the dominant modality. Lipid nanoparticles did.
Patisiran (Onpattro) reached FDA approval in 2018 for hereditary ATTR
amyloidosis, followed by givosiran (Givlaari, 2019) and lumasiran
(Oxlumo, 2020) — all hepatocyte-targeted LNP siRNA therapeutics. The
patent record shows the shift: granted U.S. patents mentioning LNPs plus
liver or hepatocyte delivery went from single digits a year through 2017
to 23 in 2022, 34 in 2023, and 62 in 2025. Literature grew from a handful
of papers per year in 2005 to 62 in 2025 in our index. Kurzweil’s claim
was correct in 2005; the mechanism that won was different.

Dark Winter was exactly as described. The Johns Hopkins / CSIS /
ANSER / Memorial Institute tabletop exercise held June 22–23, 2001
simulated a covert smallpox release in three shopping malls — Oklahoma
City, Philadelphia, and Atlanta — and, within its thirteen-day scenario,
projected approximately 3 million infections and 1 million deaths if the
outbreak ran unchecked. Kurzweil’s citation is accurate (ch. “Chapter
8: Promise and Peril of GNR”). In hindsight, the exercise did the job
Kurzweil suggested it would: it prompted the 2002 Strategic National
Stockpile smallpox-vaccine procurement and conditioned the policy
environment that later responded to anthrax, Ebola, and COVID-19.

The scorecard

Ahead of schedule: 2. On track: 4. Behind schedule: 1. Overtaken by events: 2. Wrong mechanism: 1.

What Kurzweil got right about 2005, and what he got wrong

Two patterns. First, where Kurzweil cited device-level facts — DBS is
real, the microchip implant exists, the Yamada paper exists, Dark Winter
happened — he was reliable. Devices and published exercises are
checkable, and Kurzweil had researchers and citations. His factual claims
about hardware were accurate enough that The Singularity Is Nearer
(2024) could restate several of them without embarrassment.

Second, where he cited biological background numbers — ten trillion
cells, one hundred trillion microbes — he was inheriting rather than
measuring, and the inheritance was wrong. The 10:1 microbe ratio was
the most-repeated factoid in twenty-first-century biology and it took a
2016 paper from the Milo lab to kill it. It’s a small caution about
forecasting: the places where a forecaster is most confident are often
the places they didn’t verify. The predictions get scrutinized. The
background facts get copied.

Method note

Ten “circa 2005” claims from Kurzweil’s original text were checked
against (a) our index of 9.3 million U.S. patent documents, covering
grants and pre-grant publications through September 2025; (b) our index
of roughly 357 million scientific works (OpenAlex); (c) the full U.S.
clinical-trials registry; and (d) targeted web research on FDA approvals,
exercise records, and journal publications. The verdicts reflect what
can be verified today, not what was orthodox in 2005. Primary sources
include Sender/Fuchs/Milo (PLOS Biology 2016), Farra et al. (Science
Translational Medicine 2012), Yamada et al. (Nature Biotechnology
2003), and the Johns Hopkins Center for Health Security exercise record
for Dark Winter.