Kurzweil Scorecard: The Fly Got Its Wiring Diagram. The Brain Did Not.

In October 2024, the FlyWire consortium published the complete neuronal wiring diagram of an adult female Drosophila melanogaster: 139,255 neurons, 54.5 million synapses, 8,453 annotated cell types. It took 33 person-years of proofreading. The fly’s brain is the size of a grain of sand.

Six months later, a separate team released the largest mammalian connectome ever reconstructed: a cubic millimeter of mouse visual cortex containing roughly 200,000 cells and 523 million synapses, published in Nature in April 2025.

Both papers are real. Neither tells the story Ray Kurzweil told in 2005 when he wrote that humans would eventually access, permanently archive, and understand the thousands of trillions of bytes stored in each brain (ch. “The Longevity of Information”). A human brain has roughly 86 billion neurons. The trajectory from a grain of sand to three pounds of wet tissue is the story of this scorecard.

The predictions

Batch 58 gathers nine of Kurzweil’s neuroscience claims from The Singularity Is Near — six stated as already-true in 2005, three forecast for the 2030s and 2040s. They cluster into three groups. Plumbing: neuromorphic implants for a growing list of brain regions; a Max Planck “neuron transistor” controlling a leech bidirectionally from a PC; peptide-coated quantum dots binding specific neurons and activating them with light. Accounting: human thinking at about 10^16 calculations per second per brain, 10^26 across all humans; each brain storing thousands of trillions of bytes; neurons replacing their own subsystems in weeks, as evidence that identity persists as pattern rather than fixed matter. The long bet: nanobots creating new neural connections by the 2030s; emotional overlays in shared VR generated by nanobots; and archiving every brain, byte for byte, by the 2040s.

In The Singularity Is Nearer (2024), Kurzweil doesn’t retreat. He restates: “At some point in the 2030s we will reach this goal using microscopic devices called nanobots. These tiny electronics will connect the top layers of our neocortex to the cloud, allowing our neurons to communicate directly with simulated neurons hosted for us online.” The fourth bridge — “being able to back up our mind files digitally” — is “a 2040s technology.”

Where we actually are

Neuromorphic implants by brain region. The plumbing arrived, but in a less ambitious form. Cochlear implants (Kurzweil’s own example) are standard of care. Retinal prostheses had a brief commercial run and largely failed — Second Sight discontinued Argus II in 2020. Ted Berger’s hippocampal memory prosthesis reached small human trials; a 2024 Frontiers paper reports a decoding model improving short-term retention by about 35%, still research.

What actually scaled is electrode-based BCIs, not neuromorphic ones (the distinction matters — neuromorphic means biomimetic circuitry, not just recording). In the patent corpus reviewed for this post, grants tagged with brain-computer interface language rose from 6 per year in 2010 to 108 in 2025. Neuralink implanted 19 patients by late 2025. Synchron’s endovascular Stentrode — described in US 12,508,084, granted December 2025, which claims stent-based electrode arrays inserted through the jugular vein — requires no craniotomy, and its COMMAND trial reported zero serious device-related adverse events across six patients at 12 months. IBM’s US 12,001,942 (June 2024) describes an array of micro-channels each sized to receive a single axon, paired with its own electrode and chip. US 12,333,073 (Duke and Dartmouth, June 2025) claims a multiplexed front-end amplifier placing a low-noise circuit at each electrode site, breaking thermal and wiring constraints that have capped previous arrays.

None of these are neuromorphic. They read neurons; they don’t imitate them.

The “neuron transistor.” Kurzweil’s 2005 reference to Peter Fromherz’s bidirectional silicon-neuron interface was real. What followed is more interesting than the leech demo. The neuromorphic-chip side of the bet paid off unexpectedly well: IBM’s TrueNorth (2015, 1,572 citations) put 1 million spiking neurons on 65 milliwatts; Intel’s Loihi 2 is in labs. Silicon imitating biology is flourishing; silicon-to-biology bidirectional dialogue at scale is not.

Quantum-dot neural activation by light. This is the most interesting miss. The specific mechanism — peptide-coated quantum dots binding neurons and triggering them optically — did not become the standard. Optogenetics won: channelrhodopsin-2, an opsin protein delivered by viral vector, is now the default tool (the canonical 2009 protocol paper has 556 citations). But Kurzweil’s underlying bet — nanoparticles plus light giving remote, cell-type-specific neural control — turned out right via a longer chain of molecular intermediaries. A 2018 Science paper on “near-infrared deep brain stimulation via upconversion nanoparticle–mediated optogenetics” (1,081 citations) used lanthanide-doped nanoparticles to convert tissue-penetrating NIR into visible photons that activate opsins in deep mouse brain — evoking dopamine release, silencing seizures, triggering memory recall, all transcranially. In June 2025, a Science Advances paper reported NIR light converted directly to localized electrical stimulation via photovoltaic nanoparticles, eliminating the opsin dependency Kurzweil’s quantum-dot scheme was always trying to avoid.

Nanobots creating new neural connections (by 2030s). Not at any scale. No medical nanobots in human brains are creating connections. There are drugs that nudge neuroplasticity, deep brain stimulation that modulates circuits crudely, focused ultrasound that temporarily opens the blood-brain barrier. The capillary-injected nanobot pathway has no clinical or late-preclinical analog in the literature reviewed for this post.

Emotional overlays in shared VR via nanobots (by 2030s). VR arrived as a consumer product, then stalled. Apple’s Vision Pro launched in early 2024 at $3,499 and pipes content through pass-through optics, not neural input. Meta’s Quest line is the mass-market incumbent. Every consumer VR system in 2026 delivers simulated sensory data through eyes, ears, and sometimes haptics — none pipes emotional content into neural substrate. The nanobot route is not under product development anywhere.

10^16 calculations per second per brain / 10^26 across all humans. Rough extrapolations from neuron counts, firing rates, and synapses. “Calculations” in a biological brain is not a defined quantity, so the claim is not experimentally measurable. Notably, in Nearer Kurzweil himself downgrades to “on the order of 10^14 per second” for the neuron-level computation inside the human brain — two orders of magnitude below the 2005 figure. A quiet but meaningful revision.

Brain stores thousands of trillions of bytes. Same problem: widely cited, no agreed measurement protocol. FlyWire 2024 annotated 54.5 million synapses across 139,255 neurons — roughly 390 synapses per neuron. Naively scaled to 86 billion human neurons, that’s ~3.4 × 10^13 synapses; encoding weights at a few bytes each gets into hundreds of terabytes. Including dendritic compartments, spine dynamics, glial interactions, and neuromodulator gradients plausibly reaches the “thousands of trillions” band.

Neurons replace their subsystems in weeks. Tubulin turnover is real; microtubule dynamics run on minutes-to-hours; dendritic spine remodeling on days-to-weeks. The cellular fact is supported. The philosophical conclusion — identity as pattern independent of matter — is a rhetorical leap, not a neuroscientific finding, and remains contested.

Archive every brain, byte for byte, by the 2040s. Here the fly-to-mouse-to-human trajectory matters most. FlyWire: 139,255 neurons, 33 person-years, published 2024. MICrONS: 200,000 cells, 523 million synapses in one cubic millimeter of mouse cortex, published April 2025, with hybrid AI proofreading. A human brain is five orders of magnitude larger than the fly; the mouse cubic millimeter is about 0.0001% of a human brain’s volume. Whole-mouse connectomic reconstruction is within reach this decade. A whole-human connectome at synaptic resolution, with individual-identity fidelity rather than species-average wiring, is harder to timeline.

The scorecard

What Kurzweil got right, and what he got wrong

On capability — read from neurons, write to neurons, reconstruct circuits at scale — Kurzweil’s direction was right, and in one case (neuromorphic silicon chips with millions of neurons) the capability arrived earlier than his 2005 framing implied. On mechanism, the specific paths he bet on — quantum dots, neuron transistors, nanobots — mostly lost to paths he didn’t fully anticipate: opsins delivered by viral vector, high-density microelectrode arrays, endovascular stent electrodes, AI-assisted electron microscopy.

The systematic bias: Kurzweil consistently underestimated how much of this work would happen through biology-first mechanisms — viral vectors, engineered proteins, cell-type-specific genetics — rather than silicon-first ones. The 2005 book is a silicon-age forecast of a problem that is being solved, so far, mostly with a biology-age toolkit. Optogenetics is the defining counter-narrative: a light-activated ion channel borrowed from a pond alga turned out to be the enabling technology for cell-type-specific neural control, not a peptide-coated inorganic nanoparticle.

The 2040s mind-archiving prediction is the bet that remains genuinely open. If the scaling curve from fly (140K neurons) to mouse cubic millimeter (200K cells) compounds at FlyWire-to-MICrONS pace, the 2040s call is not obviously wrong. If electron microscopy throughput and proofreading economics plateau, it is. That’s the next decade’s scorecard entry.

Method note

This scorecard combined three sources. First, a search of 9.3 million US patent filings for brain-computer interface, neural electrode, optogenetic, and neuromorphic device patents between 2010 and 2026, filtered by publication date and keyword queries on titles and abstracts. Second, a search of 357 million scientific papers (OpenAlex corpus) for the most-cited work in each mechanism area, filtered to 2015 and later with citation thresholds. Third, targeted web searches and full-text reads of recent Neuralink PRIME Study updates, Synchron COMMAND results, FlyWire consortium papers, MICrONS cubic-millimeter papers, and the most recent upconversion-nanoparticle optogenetics literature. Close paraphrases of Kurzweil’s 2005 statements are drawn from The Singularity Is Near; restatements and updates are quoted from The Singularity Is Nearer (2024).