Kurzweil Scorecard: The Dishes Never Came. The Search Did.

In 2005, Ray Kurzweil pointed to two radio telescopes as the leading edge of humanity’s hunt for company in the universe. One was the Allen Telescope Array, then a 32-dish prototype on track, he wrote, to grow to 350 dishes by 2008. The other was an experimental antenna array at Ohio State University, called Argus, that he described as a coming all-sky radio camera. Both projects pointed at a future where SETI would stop nibbling at the sky and start swallowing it.

Neither happened the way he described. The Allen Telescope Array stopped at 42 dishes, ran out of money in 2011, and never reached the 350 figure Kurzweil cited. Ohio State’s site for Argus had already been bulldozed for a golf course in 1998, seven years before The Singularity Is Near went to print. And yet — by almost any other measure — the search itself has expanded past anything Kurzweil sketched. The instruments he named died. The thing he wanted them to do is happening anyway, on different hardware, with different funders, and with a vocabulary he didn’t use.

This is the pattern this batch keeps revealing: the predictions about which machine tend to be wrong. The predictions about which capability tend to be right.

The predictions

Batch 101 collects ten claims clustered around Kurzweil’s chapter “On the Intelligent Destiny of the Cosmos” and adjacent passages on uploading, consciousness, and the scalability of human intelligence. Most are dated to roughly 2005 — claims about the state of the world at the time of writing — but a handful project further. Together they form a tight three-part question:

• Can we listen to the universe well enough to know whether anyone is talking?
• Can we read and reproduce a human brain?
• Can we tell, in principle, whether anything is conscious?

Where we actually are

The Allen Telescope Array. Kurzweil wrote that the array “has 32 dishes scheduled to be online in 2005, growing to 350 dishes by 2008,” equivalent to a 10,000-square-meter dish listening across 100 million frequency channels (close paraphrase, ch. “On the Intelligent Destiny of the Cosmos”). The array hibernated in April 2011 when funding ran out. Twenty-one years past Kurzweil’s deadline, it still consists of 42 dishes. The SETI Institute has stated the full 350-dish build would require another $40 million it has never received. What the array did get, between 2020 and 2025, was a sensitivity upgrade — cryogenically-cooled receivers across all 42 antennas, covering 1–14 GHz for transient and technosignature work. Verdict: Behind schedule. The promised array was never built.

The Ohio State Omnidirectional Search System. Kurzweil described OSU as building an instrument that would “use interferometry and intelligent computation to image the entire sky from simple antenna arrays” (paraphrase, same chapter). The project — Argus, designed by Steve Ellingson and others at OSU — consisted of 22 broadband spiral antennas, expandable to 32, in the 1200–1700 MHz band. More damning: the host facility, the famous “Big Ear” telescope that recorded the 1977 Wow! signal, had been demolished in 1998 to make way for a golf course expansion. The land was sold without the observatory’s knowledge. Argus’s lead designers dispersed to Virginia Tech and CSIRO. Technical updates trail off after 2008. Verdict: Overtaken by events. The institutional base for the prediction was gone before the prediction was printed.

The nine-dimensional parameter space. Kurzweil’s most defensible SETI claim was the most modest one: that as of 2005, “only very thin slices of SETI’s nine-dimensional parameter space had been explored” (paraphrase, same chapter). Twenty-one years later, the slices are wider. Breakthrough Listen, funded with $100 million by Yuri Milner in 2015, surveyed 1,327 nearby stars at 1.10–3.45 GHz (Price et al. 2020, 105 citations) and in 2023 reported a search of 97 nearby galaxies at four frequency bands. China’s FAST and the Murchison Widefield Array have added blind searches of millions more stars toward the galactic center. The word “technosignature” — barely visible in scientific publishing when Kurzweil wrote — went from three matched papers in 2017 to 165 in 2025. Margot et al.’s 2023 Nature Astronomy deep-learning search of 820 nearby stars (DOI 10.1038/s41550-022-01872-z) prefigured a November 2025 Berkeley result reporting a detection pipeline running 600× faster than the previous benchmark. None of this happened on the instruments Kurzweil named. Verdict: Ahead of schedule on the search; behind schedule on the hardware Kurzweil bet on.

The brain as readable hardware. Kurzweil claimed “the human brain’s hierarchical complexity does not exceed the level of complexity humans are already capable of handling” (ch. “Reverse Engineering the Brain: An Overview of the Task”). For most of the two decades after he wrote that line, this looked like wishful thinking. The state of the art was the C. elegans worm: 302 neurons. Then in October 2024, the FlyWire consortium published the full connectome of an adult Drosophila brain — 139,255 neurons, more than 50 million synapses, 8,453 cell types — in Nature (Dorkenwald et al., 447 citations and counting; DOI 10.1038/s41586-024-07558-y). It took about 33 person-years of human proofreading on top of AI-assisted reconstruction; the same project without AI would have taken nearly 50,000 person-years. In April 2025, Princeton’s contribution to the MICrONS Project published a half-billion-synapse functional connectome of mouse visual cortex in Nature (DOI 10.1038/s41586-025-08790-w). Nature Methods declared EM-based connectomics its Method of the Year for 2025. In The Singularity Is Nearer, Kurzweil hardens the bet: “One of the major research projects of the next two decades will be figuring out what level of brain emulation is sufficient” and lists five categories from functional through connectomic, cellular, biomolecular, quantum. Verdict: On track. The complexity has, so far, been handleable — though no one has produced a working mammal-scale emulation yet.

The fraction-of-a-second scan claim. Kurzweil argued that an upload would only need to capture “a person’s state within the natural amount of change occurring over a fraction of a second or a few minutes to pass a personalized identity test” (paraphrase, ch. “Uploading the Human Brain”). This is a claim about what would suffice for identity continuity rather than a claim about anything we can do yet. No instrument exists that can scan a living human brain at the resolution and speed required. Bostrom and Sandberg’s 2008 Whole Brain Emulation: A Roadmap — the document Kurzweil cites in the 2024 update — estimated that anything beyond functional-level emulation requires destructive methods. The whole question remains theoretical. Verdict: Too early to call. Nothing has been done that would test it.

The hard problem of consciousness. Kurzweil claimed in 2005 that “because subjective experience cannot be resolved entirely through objective measurement and analysis, philosophy will continue to play a critical role in questions of consciousness” (paraphrase, ch. “The Vexing Question of Consciousness”), and separately that “there exists no objective test that can conclusively determine the presence of consciousness.” In May 2025, Nature published the results of the Cogitate consortium’s adversarial collaboration — a seven-year, six-lab, 256-participant, pre-registered experimental shootout pitting Integrated Information Theory against Global Neuronal Workspace Theory using fMRI, magnetoencephalography, and intracranial EEG (Cogitate Consortium 2025, DOI 10.1038/s41586-025-08888-1). The result was a draw. Both theories were partially supported and partially falsified. IIT was contradicted by a lack of sustained posterior synchronization. GNWT was contradicted by limited prefrontal representation and missing “ignition” at stimulus offset. After seven years and an unprecedented amount of money and equipment thrown at the question, no objective test for consciousness emerged. Twenty years on, Kurzweil’s claim is intact. The literature this batch generated — over 5,000 papers tagged with “integrated information consciousness” published in 2025 alone — is the sound of a problem refusing to resolve. Verdict: Verified. The objective test has not arrived and the hard problem has eaten every attempt to dissolve it.

Wormholes for transport and communication. Kurzweil suggested that “engineered or naturally occurring wormholes may allow transportation and communication to distant locations in the universe much faster than ordinary light-travel routes” — drawing on Morris, Thorne, and Yurtsever’s 1988 work. The math has moved. In 2021, Blázquez-Salcedo et al. published the first traversable wormhole in Einstein-Dirac-Maxwell theory (Physical Review Letters, 148 citations; DOI 10.1103/physrevlett.126.101102), followed by Maldacena, Milekhin, and Popov’s humanly-traversable construction using Casimir-like fermion energy. Traversable-wormhole papers grew from 25 in 2005 to 363 in 2025 — about a 14× increase, driven by the ER=EPR conjecture. In late 2022, a Google team running an SYK model on the Sycamore processor reported observing wormhole-like dynamics — qubit teleportation through a circuit whose behavior matched the gravitational dual of a traversable wormhole. (The interpretation was contested; the experiment was real.) None of this produced a wormhole you could drive a probe through. It produced a theoretical infrastructure that did not exist when Kurzweil wrote. Verdict: Wrong mechanism. The wormholes physicists now study live in entanglement geometries, on chip.

The hemoglobin throughput number. A small but testable claim: that the human body creates hemoglobin about 500 trillion times per second, for roughly 1.5 × 10¹⁹ ribosomal read operations per minute on hemoglobin alone (ch. “Life’s Computer”). About two million red blood cells enter circulation per second, each containing roughly 270 million hemoglobin molecules — multiplying out to about 5.4 × 10¹⁴ produced at steady state, broadly compatible with Kurzweil’s order of magnitude. Verdict: Verified historical.

The scorecard

What Kurzweil missed (and what he nailed)

The errors here are almost entirely errors of institutional naming. Kurzweil pointed at the wrong telescope, the wrong observatory, the wrong funder — and the right capability. Breakthrough Listen, using existing dishes at Green Bank, Parkes, and MeerKAT, ate the role he gave the 350-dish ATA. The Murchison Widefield Array, in Australia, ate the role he gave OSU’s Argus. The 350-dish vision arrived as something stranger — a smaller array with sharper receivers, plus a million-channel back-end at a different facility, plus an AI pipeline that didn’t exist when Kurzweil was writing.

The same pattern shows up in brain emulation. Kurzweil’s claim that the brain’s complexity is within human handling capacity looked overconfident for most of two decades and now looks plausible. But the breakthrough came from a non-obvious combination: electron microscopy at scale, citizen-scientist proofreaders, and machine-learning segmentation models. Kurzweil sketched a future where we would understand the brain by reverse-engineering its algorithms. FlyWire understood it by mapping every wire in a fruit fly’s head. Different mechanism, similar destination.

The single prediction Kurzweil got cleanly right may be the philosophical one. He insisted, against the optimism of his own technological narrative, that no objective test for consciousness was coming. Twenty years later, with hundreds of millions of dollars of neuroimaging and the most rigorous pre-registered study consciousness research has ever seen, that remains true. The hard problem is, so far, hard.

A forecasting lesson sits at the bottom of this batch: betting on a capability aged better than betting on an instrument. The instrument died and was replaced; the capability got farther than the original instrument’s roadmap promised. That pattern recurs through this scorecard project.

Method note

For each prediction, we counted patents and high-citation papers using full-text indexes over a corpus of 9.3 million U.S. patents and 357 million scientific papers, then read the most relevant 2024–2025 papers in full where they were open-access. Trend curves come from year-by-year counts of papers whose title and abstract match the relevant terms. Hardware and funding history for the Allen Telescope Array, Big Ear, and Argus came from the SETI Institute, UC Berkeley, the Ohio State Radio Observatory memorial archive, and contemporary press. The Cogitate consortium result is the May 2025 Nature paper (DOI 10.1038/s41586-025-08888-1). The fruit fly connectome figures are from Dorkenwald et al., Nature, October 2024 (DOI 10.1038/s41586-024-07558-y). Restatements from Kurzweil are from the full text of The Singularity Is Nearer (2024).