Kurzweil Scorecard: The Brain Was Both Easier and Stranger Than He Predicted

In April 2025, a consortium of labs released a 1.6-petabyte reconstruction of a cubic millimeter of mouse visual cortex — 200,000 cells and roughly half a billion synapses across four contiguous cortical areas, mapped at the level of individual dendritic spines. That same month, Cortical Labs began shipping a $35,000 biological computer whose processing layer is a living sheet of human neurons cultured on a silicon electrode array. And in August 2024, a group at Wellesley College reported that anesthetic drugs lose potency in rats whose brain microtubules have been stabilized — the first experimental evidence for a theory of consciousness that Ray Kurzweil had dismissed.

Twenty years ago, The Singularity Is Near predicted that within two decades, we would have detailed models of the several hundred information-processing regions of the human brain. The deadline has arrived. What was delivered looks almost nothing like what he described.

The predictions

This batch bundles ten brain-modeling claims and forecasts Kurzweil packed into two chapters — “A Panoply of Criticisms” and “The Criticism from Microtubules and Quantum Computing.” Read together, they describe a vision of how the brain would yield to reverse-engineering. Neurons and synapses are already well-modeled. Brain regions are simpler than the neurons they contain. The genome encodes the brain with compact probabilistic rules. Microtubules do not do anything strange. Within twenty years, the whole thing falls.

He was right about the trend. He was wrong about the destination. And he was confident about microtubules in a way that is not aging well.

The big-ticket prediction

Kurzweil wrote that “within twenty years, we will have detailed models and simulations of the several hundred information-processing organs collectively called the human brain” (ch. “The Criticism from Software”). In The Singularity Is Nearer he doubles down, estimating that computers will be able to “simulate human brains in all the ways we might care about within the next two decades or so.”

Here is what actually landed by the 2025 deadline. The MICrONS dataset, published in Nature (DOI 10.1038/s41586-025-08790-w), delivered dense calcium imaging of about 75,000 neurons across four areas of a mouse visual cortex — VISp, VISrl, VISal, and VISlm — co-registered with an electron-microscopy reconstruction of roughly 200,000 cells and half a billion synapses. It is the largest mammalian-cortex wiring diagram ever assembled, built across Baylor, the Allen Institute, and Princeton over six continuous months of EM imaging.

It is a triumph. It is also one cubic millimeter of one mouse, in a region dedicated to one sensory modality. The human brain is roughly a million times larger by volume. The two public programs that targeted whole-brain work both ended short: the EU-funded Human Brain Project closed in 2023 after spending €600 million without producing a working whole-brain simulation, and the Blue Brain Project at EPFL shut down at the end of 2024. A 2024 projection published in a related review projected mouse whole-brain cellular simulation around 2034 and human not before 2044.

Verdict: behind schedule, by roughly a decade and a species.

What Kurzweil got right about the trend

The claim that brain scanning and modeling are accelerating is straightforwardly true. Our literature database contains about 4,400 papers mentioning brain organoids between 2013 and 2025, growing from 17 in 2013 to more than a thousand in 2025 alone — a 61× increase. Papers on whole-brain connectome work grew from roughly one per year before 2010 to hundreds per year now. Connectome 2.0 — the ultra-high-gradient MRI scanner described in a 2021 NeuroImage paper (DOI 10.1016/j.neuroimage.2021.118530) — is now an operating instrument, with 171 citations. Electron-microscopy connectomics tooling has matured to the point that neuPrint, the open-access analysis platform described in Frontiers in Neuroinformatics in 2022, is now a standard dependency.

The realistic-neuron-model claim also largely holds up. A 2024 Cell paper on neurotransmitter classification from electron-microscopy images at synaptic sites in Drosophila (DOI 10.1016/j.cell.2024.03.016) infers chemical identity at single synapses directly from EM. Kurzweil’s claim that we already had “realistic mathematical models and computer simulations of neurons and interneuronal connections” was optimistic in 2005 — and accurate by 2025.

Verdict on scanning acceleration and neuron-level modeling: on track to ahead.

The mechanism surprise: living tissue

Kurzweil’s prediction that hybrid biological/nonbiological neural networks would perform similarly to purely biological ones (ch. “The Criticism from Microtubules and Quantum Computing”) was meant as a stepping stone toward silicon-brain replacement. What happened instead is that the biological side kept pulling ahead of the hybrid.

The 2023 paper “Organoid intelligence (OI): the new frontier in biocomputing and intelligence-in-a-dish” (Hartung et al., DOI 10.3389/fsci.2023.1017235) — 243 citations — argued that cultured three-dimensional brain organoids could be trained as biocomputers using microelectrode arrays. By 2025 two companies had made that commercial. Cortical Labs shipped the CL1, a $35,000 integrated device that couples cultured human neurons to silicon hardware; FinalSpark, a Swiss startup, rents organoid-based compute remotely for $1,000 per month and reports training cycles that use less than a millionth of the energy of equivalent silicon workloads.

The patent record mirrors this. US 12,502,562 (granted December 2025) claims an hiPSC-derived brain organoid with a layered structure — a core region positive for the neural progenitor markers SOX2 and TLX, and an outer layer positive for BRN2, replicating the ventricular-to-cortical-plate organization of developing cortex. US 12,577,531 (March 2026) describes a method for incubating primitive-like macrophages with a brain organoid to produce microglia-sufficient organoids — the immune cells that the original brain-in-a-dish preparations conspicuously lacked. These are claims on the biology, not the silicon.

Kurzweil expected the path to artificial brains to run through digital reconstruction. The shorter path running in parallel is to grow the real thing and interface with it. Verdict: wrong mechanism, but flattering to the spirit of the original claim rather than the letter.

The microtubule paragraph, reconsidered

The most confident prediction in the batch is also the one most exposed to recent data. Kurzweil wrote, citing Seth Lloyd, that “the brain is a hot, wet environment that is not favorable for exploiting quantum coherence” and that “there is no evidence that neuronal microtubules perform quantum computing or that such quantum computing is required for consciousness” (ch. “A Panoply of Criticisms”). In The Singularity Is Nearer, he maintains a working hypothesis that a model based only on neuronal firing can achieve a working brain simulation, and relegates microtubule-level and quantum models to a low-probability contingency.

In August 2024, Wiest and colleagues at Wellesley published an eNeuro paper reporting that rats given microtubule-binding drugs take measurably longer to fall asleep under isoflurane than controls — suggesting that the anesthetic acts on microtubules, exactly as the Orch-OR theory of Penrose and Hameroff predicts. In 2025, Babcock’s paper “A quantum microtubule substrate of consciousness is experimentally supported and solves the binding and epiphenomenalism problems” (Neuroscience of Consciousness, DOI 10.1093/nc/niaf011) reviewed that result alongside evidence of macroscopic quantum-entangled states in living human brains correlated with working-memory performance. The literature has reacted visibly: our database logs roughly 300 papers mentioning microtubule quantum consciousness in 2025 alone, against single-digit annual counts throughout the 2010s.

None of this proves that consciousness requires quantum microtubule computation. It does mean the theory is no longer empirically empty — and that Kurzweil’s 2005 dismissal was premature rather than prescient. Verdict: wrong mechanism / too early to call on the strong form, behind schedule on the prediction that microtubules would quietly disappear from the conversation.

The scorecard

What Kurzweil missed (and what he nailed)

The pattern is legible. Kurzweil nailed the direction of the curves — scanning resolution, neuron-model fidelity, the rate at which regions came under empirical control — and got the endpoint wrong. He assumed a smooth interpolation from “we can model a cerebellum” to “we can model a brain,” where in practice each order-of-magnitude scale-up required new instruments, new ethics frameworks, and new theoretical commitments about what “modeling” even means.

The deeper miss is about substrate. He was too quick to close the door on alternatives — microtubule-level quantum effects on one side, living-tissue computation on the other. Twenty years later, both doors are wider open. Some of the most interesting work in brain-scale computation in 2026 is happening at Wellesley, at Cortical Labs, and inside Petri dishes at FinalSpark, none of which looks much like the silicon-replaces-neurons scenario the book described. Directionally right and methodologically narrow: the signature failure mode of the book.

Method note

Counts of patents and scientific papers come from full-text searches over our internal corpus of 9.3 million US patents and 357 million scientific papers. Patent numbers are cited as granted; paper claims are drawn from abstracts and, where open-access, from full text. Project-status claims for the Blue Brain Project, Human Brain Project, MICrONS, Cortical Labs, and FinalSpark are sourced from published news coverage and primary papers accessed during this analysis. Verdicts are the author’s judgment from the evidence assembled; the underlying queries and sources are logged.