Kurzweil Scorecard: Zoom Killed the Commute Kurzweil Assigned to VR

Ray Kurzweil spent a whole chapter in The Singularity Is Near answering
critics who argued the rich-poor divide, government regulation, and theism
would break his curves. His answers rested on one mechanism: information
technology becomes almost free, spreads everywhere, and in doing so
collapses distance, cost, and inequality. Two decades on, the mechanism
half-works. The outcomes are arriving — just not through the channels he
named.

The clearest example sits at the center of this batch. Kurzweil argued
that full-immersion virtual reality would reduce the need to travel.
That substitution happened. Global business travel spend hit a projected
$1.52 trillion in 2025, but Deloitte’s corporate travel survey reports
that virtual meetings have permanently displaced about 30 percent of
pre-pandemic trips. Counterpoint Research put global VR headset shipments
down 12 percent year-over-year in 2024, the third straight annual
decline. Apple’s Vision Pro, the flagship immersive device launched in
2024, sold an estimated 370,000 to 420,000 units for the full year and
then fell 43 percent quarter-over-quarter in Q4. The substitute Kurzweil
described is real. The technology doing the substituting is a grid of
rectangular video tiles on a laptop.

The predictions

The twelve statements cluster around three themes: the rich-poor divide
as a test of exponential progress, virtual reality as a replacement for
physical presence, and the shrinking lag between early-adopter and
ubiquitous technology. A twelfth — that most humans will eventually
believe nonbiological entities are conscious — belongs to a timeframe
that hasn’t arrived.

Where we actually are

The inequality thesis held. The mechanism held only halfway.

Kurzweil wrote that “the rich-poor divide in access to advanced
technologies will be rapidly mitigated because information technologies
become inexpensive and ubiquitous after early adoption”
(ch. “The Criticism from the Rich-Poor Divide”). He cited Xavier
Sala-i-Martin’s 2005 finding that “eight measures of global inequality
among individuals had all declined over the previous quarter century”.
The latest evidence vindicates the core claim. Branko Milanovic’s
Global Income Inequality in Numbers and its 2024 update show the
global Gini coefficient fell from a peak of 69.4 in 1988 to 60.1 in
2018 — a level last seen in the late nineteenth century. In 2024
Sala-i-Martin himself published a new NBER working paper showing that
within-country inequality peaked shortly after the global financial
crisis and has been declining through the second half of the 2010s.
Using the World Bank’s $3.65-a-day and $6.85-a-day poverty lines, the
share of humanity below those thresholds has dropped to roughly 30 and
50 percent of their 1990 levels. On the headline verdict, Kurzweil was
right.

The mechanism is weaker. Milanovic credits the collapse in global
inequality overwhelmingly to China’s growth, and China’s contribution
is essentially spent. His projections, echoed by the World Bank’s 2024
Poverty, Prosperity, and Planet report, expect only about 69 million
people to escape extreme poverty between 2024 and 2030, compared with
150 million between 2013 and 2019. The World Bank’s March 2024 update
tracked extreme poverty at 8.5 percent of the global population —
roughly 700 million people — and concluded that eradicating it by 2030
is no longer plausible. The driver of the past three decades was an
industrial, not an information-technology, expansion.

Web access in Africa expanded, but slowly compared to what “almost
free” would imply.

Kurzweil wrote that “Web access is expanding rapidly even in the
poorest countries of Africa”. The rate is impressive; the absolute
level is not. ITU data shows internet penetration in Africa was 2.4
percent in 2005 and 9.8 percent in 2010, growing at an average annual
rate of 16.7 percent. By 2025 it reached 38 percent — still the
lowest of any ITU region and well below the 68 percent global average.
Entry-level fixed broadband in Africa still costs a median of 13.4
percent of gross national income per capita, far above the 2-percent
Broadband Commission target. Rural penetration is 23 percent against
57 percent urban. The “almost free” prediction is accurate at the
bit-delivery layer but fails at the service layer: the compute has
gotten nearly free, the access hasn’t.

The immersive-reality prediction got the social change right and
the hardware wrong.

Kurzweil wrote that “as increasingly realistic high-resolution
full-immersion virtual reality emerges, more of our need to be together
physically will be met through computation and communication”
(ch. “The Criticism from Lock-In”). In The Singularity Is Nearer
he restated it more confidently: “the VR and AR of the late 2020s
will merge into a compelling new layer to our reality. In this digital
universe, many products won’t even need a physical form at all … a full
virtual meeting, with the ability to interact with coworkers and easily
collaborate as though together in person; a virtual concert, with the
fully immersive auditory experience of sitting in a symphony hall.”
He also acknowledged the pandemic pulled forward some of the shift:
“At the peak during the pandemic, up to 42 percent of Americans were
working from home … the old model of nine-to-five sitting at a desk in
a company office has been obsolete for years.”

The patent record tells a quieter story than that. Grants tagged to
video conferencing, virtual meetings, or remote collaboration rose
from 159 in 2019 to 274 in 2023 — an enterprise-software build-out,
not an immersive-hardware one. US 12,603,091, granted April 2026 and
titled “Immersive collaboration of remote participants via media
displays”, describes video-conferencing over a transparent electrochromic
display that adjusts its opacity based on sun position and weather to
composite a remote colleague into the local room — incremental work
around a flat display, not a headset. US 12,598,270 (“Generating and
providing in-meeting coaching for video calls”) is an AI layer bolted
onto a Zoom-style call. US 12,579,964 is literally “Acoustic echo
suppression and cancellation for web-based virtual meeting platform”.
The inventive energy of the 2020s has gone into making the flat video
call better.

Contrast that with VR-plus-avatar patents also granted this spring:
US 12,602,891 for a “Teleportation system combining virtual reality
and augmented reality”, or US 12,597,361 for “Deterring sexual
harassment of avatars in virtual reality”. Both assume a future where
avatars meet in persistent 3D space. Both also sit inside an
immersive-hardware industry that shipped 12 percent fewer units in
2024 than the year before. Kurzweil predicted a substitute for travel.
He got one — running on the technology stack of 2005’s WebEx, scaled
up, not on the brain-computer interface stack he expected.

The adoption-lag prediction was split by layer.

Kurzweil wrote that “the lag from very expensive early-adopter
versions of information technologies to very inexpensive ubiquitous
adoption is now about a decade”, and that in twenty years — by 2025
— “the lag from early-adopter to ubiquitous inexpensive adoption
for information technologies will be only two to three years”. That
holds decisively at the software layer. The cost to run AI at the
GPT-3.5 performance level fell from about $20 per million tokens in
November 2022 to about $0.07 per million tokens by October 2024 — a
roughly 280-fold drop in under two years, described in a recent arXiv
analysis of algorithmic efficiency. 5G networks went from first launch
in 2019 to coverage by 209 operators across 83 countries within five
years. But at the hardware layer the curve bent the other way. VR
headsets have been in market for a decade, and the technology is still
shrinking as a category. The two-to-three-year diffusion forecast is
ahead of schedule for bits and models; it’s behind schedule for
devices that require a screen strapped to a face.

The scorecard

What Kurzweil nailed, and what he missed

Kurzweil was right on the direction of almost every arrow. Global
inequality fell. Poverty fell. Information technology got almost free.
Virtual meetings substituted for travel. Africa came online. The
adoption lag for software compressed to months, not years.

Where he erred, he erred on the physical substrate. He over-weighted
the role of hardware breakthroughs — full-immersion VR, brain-computer
interfaces, autonomous transport — in delivering those gains. Again
and again the substitute he described turned out to be a software
layer on existing commodity hardware. WebEx and its descendants, not
head-mounted displays, carried the load of the remote-work shift.
Mobile web on a $40 smartphone, not neural implants, carried African
connectivity. GPT-4-class inference on commodity GPUs, not radically
new architectures, carried “almost free” intelligence.

The other systematic error is about the tail. Extreme poverty reduction
was extraordinary from 1990 through 2019 and has slowed to a crawl
since. The last mile of the rich-poor divide is harder than the first —
the countries remaining in poverty are smaller, more fragile, and more
climate-exposed than the China-shaped engine that drove the curve for
three decades. Kurzweil’s price-performance logic predicts convergence.
The data predicts partial convergence with a stubborn floor.

Exponential price-performance is real, but it flows unevenly across
the stack. Fast in bits, slower in atoms, slower still in the
institutions that have to absorb the bits. When Kurzweil predicted
VR would replace travel by the 2020s, he was wrong about which layer
the substitute would use. He was right that one would arrive.

Method note

We checked the twelve statements in this batch against our US patent
index (9.3 million documents), a 357-million-record scientific literature
index, and current reporting on global inequality, extreme poverty,
internet penetration, VR shipments, AI inference pricing, and
business-travel recovery. Patent numbers are USPTO identifiers. Quotes
from The Singularity Is Nearer are verbatim from the 2024 Viking
edition. Each score corresponds to evidence in a named source, not a
vibe.

Next scorecard: batch 51.