Kurzweil Scorecard: The Brain Got Read. It Didn’t Get Decoded.

In January 2024, a paralyzed man named Noland Arbaugh moved a cursor across a screen with a thought. By early 2026 he had logged thousands of hours on the implant, and the company that built it had put electrodes in more than a dozen people. A second firm, Synchron, threaded its device into the brain through a blood vessel — no skull opened — and is now lining up the pivotal trial it needs before asking the FDA for the first market approval of an implantable brain interface. A third, Precision Neuroscience, already cleared an FDA review in April 2025 for a film thinner than a hair that lays across the cortex.

So the brain has become readable. Wires go in, intentions come out. This is exactly the kind of progress Ray Kurzweil was betting on in 2005. And yet the specific neuroscience predictions in this batch — three of them, all clustered around the late 2020s — are mostly running behind. The gap between “we can read signals off the brain” and “we have decoded the brain” turns out to be the whole story.

The predictions

Writing in The Singularity Is Near (2005), Kurzweil made three claims about where neuroscience would be roughly now.

The first was a deadline. By the late 2020s, humanity will have completed the reverse engineering of the human brain (ch. “The Vexing Question of Consciousness”). Not a sketch — completed.

The second was a capability with a hard date. By 2029, experience-beaming technology will let one person’s brain experience another person’s sensory experiences, and possibly the neurological correlates of emotions (ch. “The Vexing Question of Consciousness”). Telepathy by instrument, in other words, three years from now.

The third was a humble note about the starting line. As of 2005, humanity does not yet have the means to access and back up the neural knowledge patterns that constitute a person (ch. “Ich bin ein Singularitarian”). He flagged this as the thing that had to change before a mind could ever be copied.

Where we actually are

Start with the deadline, because it anchors the other two. “Completed reverse engineering of the human brain” has a concrete technical meaning, and the field has a 2025 status report for exactly this question. The honest summary: not close.

The largest piece of mammalian brain ever densely reconstructed is a single cubic millimeter of mouse visual cortex — about 120,000 neurons and 523 million synapses. A human brain has on the order of 86 billion neurons. The complete connectomes that do exist are for the roundworm C. elegans (~300 neurons) and, as of the last couple of years, the fruit fly (~140,000 neurons, male nervous system and female brain both reconstructed). The field’s own framing is blunt: no organism’s complete brain has been recorded at single-neuron resolution. Not the zebrafish (~80% coverage), not even the worm. The bottleneck is data acquisition, not computers. Proofreading a single reconstructed mouse neuron runs around $1,000; a full mouse connectome is a roughly billion-dollar undertaking at current prices, and those prices have to fall by orders of magnitude before a human is on the table. Fewer than 500 people worldwide work directly on brain emulation.

This is behind schedule on the human brain — by a lot. But it deserves a split verdict, because at the small scale the prediction came true on time. We do have completed wiring diagrams for a worm and a fly. Kurzweil’s error was one of scale, not direction.

Now the telepathy claim, which is where the mechanism story lives. Sharing signals between two brains is real and old: in 2013, Miguel Nicolelis’s lab wired two rats so a “decoder” rat could act on what an “encoder” rat perceived, and a University of Washington team sent a signal from one human brain to another that twitched the recipient’s hand on a video game. Those are genuine brain-to-brain links. But twelve years later they have not scaled into anything resembling “experiencing another person’s sensory world.” What’s been transmitted is a bit — a yes, a go, a single motor command — not a sunset, not a smell, not grief.

What the patents are building is telling. US 12,572,208, granted March 2026, describes a wearable interface that reads electrical signals, hands them to a machine-learning model that “extracts a pattern” and infers the user’s intent, then transmits that intent over a telecommunications network. Read the claim carefully and you see the substitution: it is not decoding the brain’s representation of an experience and replaying it in another head. It is training a statistical model to guess what you mean, and sending the guess as data. The richer US 12,566,502, granted a week earlier, goes further — it communicates with a brain region that “forms concepts without sensory input” and converts the concept to computer-readable data. Closer to mind-reading in spirit, but still: a model mapping signals to outputs, not a wiring diagram being run in reverse.

That is the pattern across the whole field. Patents mentioning brain-computer interfaces grew from 2 in 2012 to 42 in 2025; ones mentioning neural decoding went from a couple a year to 259 in 2025. The scientific literature on the connectome climbed from roughly 280 papers in 2014 to over 2,100 in 2025. Registered clinical trials for brain interfaces have run hot the last three years. The assignee list reads like consumer tech, not just medicine — Meta, Snap (which absorbed NextMind), HI LLC, Neurable, Cognixion, alongside the medical players Synchron and Precision. Everyone is shipping decoders. Nobody is shipping a decoded brain. The 2029 experience-beaming target is behind schedule, and arriving by a different mechanism than Kurzweil described — narrow learned mappings from signal to intent, not a read-out of the brain’s own code.

The third claim is the easy one, and the most quietly damning. Kurzweil’s 2005 statement that we couldn’t yet back up the neural patterns of a person was true then and is true now. We still cannot. Score it verified — and still standing. What’s notable is the author’s own correction. In The Singularity Is Nearer (2024), Kurzweil moves the goalposts on himself: “This scenario will become realistic as we approach the Singularity in the mid-2040s,” he writes of backing up a mind, “once we have backed ourselves up, how could we die, anyway?” (The Singularity Is Nearer, ch. on radical life extension). The man who in 2005 implied the means were coming soon now files the same capability under the 2040s — a roughly two-decade slip, made by the original forecaster.

The scorecard

What Kurzweil missed (and what he nailed)

The pattern in this batch is consistent enough to be a rule. Kurzweil was right about the direction of every one of these and wrong about the depth. Brain interfaces did arrive, on something like his timeline — that’s a real hit, and the people moving cursors and speaking through implants in 2026 are proof of it. But the deep version of each prediction — a finished wiring diagram, shared subjective experience, a backed-up self — required understanding the brain’s code, and that understanding has not come. The field built a workaround instead: train a model to predict the output without ever learning the rules that produce it.

That is the same trick that made large language models work, and it may be the defining surprise of this whole era. Kurzweil assumed the road to thinking machines and copied minds ran through a complete theory of the brain. The actual road went around it. We got systems that behave intelligently without anyone reverse-engineering intelligence, and interfaces that read intent without anyone decoding the brain. For the cursor-mover, the workaround is enough. For the person hoping to back themselves up, the difference between predicting the output and understanding the system is everything — and it is precisely the difference Kurzweil’s own 2024 revisions push another twenty years out.

Method note

Verdicts here draw on full-text searches across roughly nine million patent records and a corpus of several hundred million research papers, filtered by year to establish trends, then read closely for the specific inventions named. Two recently granted brain-interface patents were pulled in full to check what they actually claim. Clinical-trial counts come from the public US registry. The state-of-the-field figures on connectomes and brain emulation come from a 2025 field report and primary reconstruction studies, and the current status of the leading interface companies from contemporary reporting. Kurzweil’s predictions are quoted from The Singularity Is Near (2005) and his restatements from The Singularity Is Nearer (2024). Every number above traces to a query run or a source read while writing this.