Paradromics Neuralink and what a real cortical interface would need
Clinical trials of invasive brain-computer interfaces (BCIs) have reached milestones that were science fiction a decade ago. Neuralink’s N1 implant records from 1,024 electrode sites on 64 flexible threads inserted by a surgical robot. Paradromics received FDA approval for its Connect-One study in November 2025 and completed the first human implantation of its Connexus Direct Data Interface at University of Michigan Health. Between them these systems have demonstrated reliable motor cursor control, robotic limb manipulation and direct speech decoding from human cortex.
Those achievements are frequently extrapolated into near term predictions of mind uploading, telepathic cloud connections, or the rehoming of human consciousness into synthetic hardware. Measured against the physiology of biological consciousness, that extrapolation runs into a bandwidth wall of about five orders of magnitude.
| Interface platform | Approach | Channel count | Primary mode | Primary limitation |
|---|---|---|---|---|
| Neuralink N1 | Penetrating flexible polymer threads | 1,024 across 64 threads | Read, spike rates | Thread micromotion, localized insertion |
| Paradromics Connexus | Penetrating micro-wire array | 421 per cortical module, up to 1,684 across four | Read, broadband single unit | Thermal dissipation, invasive craniotomy |
| Precision Neuroscience Layer 7 | Surface micro-ECoG strip | 1,024 surface electrodes | Read, local field potentials | No single unit resolution in deep cortical layers |
| Human corpus callosum | Biological myelinated and unmyelinated axons | 200 to 300 million | Bidirectional spike trains | Biological substrate, not tappable by external metal |
Channel density against axonal scale
The anatomical structure that integrates the two cerebral hemispheres into a single unified conscious field is the corpus callosum. In the adult human brain it contains between 200 million and 300 million individual nerve fibers. Those axons carry phase locked action potentials, reciprocal oscillatory signals in the beta and gamma ranges, and high frequency spike trains that let distributed cortical areas coordinate within milliseconds.
The most advanced arrays currently implanted in human subjects sit between 1,024 channels for Neuralink’s N1 and 1,684 for a four module Connexus configuration. Decoding motor intention is possible at that scale because motor populations exhibit redundant low dimensional manifold dynamics, and a sample of roughly 1,000 neurons in primary motor cortex is often enough to move a cursor.
Integrating an artificial neural network into an ongoing first-person conscious stream is a different requirement. As Masataka Watanabe argues in his split-brain test for machine consciousness, a synthetic bridge has to replicate the reciprocal bandwidth of natural interhemispheric commissures. Against 200 million axons, 1,684 channels falls short by a factor of roughly 120,000.
The unsolved writing problem
Recording neural signals is half of the interface challenge. A cortical bridge has to be bidirectional, writing complex spatio-temporal patterns of activity back into biological tissue with cellular precision. Writing by electrical microstimulation introduces three biophysical constraints that passive recording never meets.
The first is electrochemical damage. Injecting charge into the brain drives localized electrochemical reactions at the electrode-tissue interface. Once charge density exceeds the Shannon limit, the threshold at which irreversible Faradaic reactions begin, the electrode causes water electrolysis, toxic pH shifts and permanent neuronal necrosis.
The second is loss of cellular specificity. When an extracellular microelectrode discharges, current flows through conductive extracellular fluid along paths of least resistance. Rather than activating one target pyramidal cell, the current indiscriminately activates neighboring axons of passage, inhibitory interneurons and non-target circuits across a radius of hundreds of micrometers.
The third is thermal dissipation. Active stimulation circuits and high bandwidth telemetry generate heat. The brain sits inside an insulating skull, and raising local cortical temperature by more than 1 degree Celsius triggers heat shock proteins, altered blood-brain barrier permeability and cell death.
To avoid parenchymal tissue damage, neurosurgeon Benjamin Rapoport developed Precision Neuroscience’s Layer 7 Cortical Interface, using flexible micro-electrocorticography arrays that sit conformally on the cortical surface without penetrating the brain. The surface approach avoids microvascular damage and glial scarring. It trades away the ability to record or stimulate individual spikes in cortical Layers 4, 5 and 6, which are the layers that carry Global Workspace ignition and recurrent processing.
What this means for mind uploading and The Consciousness AI
In the Sandberg and Bostrom whole brain emulation roadmap, high bandwidth cortical interfaces are the link between biological tissue and a synthetic emulation. On the architecture page and the Neutral Core page, The Consciousness AI project explores hypothetical split-brain transfer criteria modeled on Watanabe’s framework. The engineering reality of Paradromics, Neuralink and Precision Neuroscience is the facts-discipline check on those concepts.
Commercial BCIs are extraordinary medical neurotechnologies, and they are restoring independence to people with severe paralysis, ALS and motor loss. In their current and near term physical form they cannot support mind uploading, continuous consciousness transfer or synthetic telepathic bridges. The distance is arithmetic rather than conceptual, and it is a factor of roughly 120,000. The bandwidth wall between clinical BCIs and a synthetic emulation is one of the constraints documented in the brain emulation section.
A functional bridge between a biological mind and a synthetic spiking network will require new paradigms in bidirectional nanoscale neurotechnology, capable of addressing millions of individual axons without inducing thermal damage, foreign body immune responses or electrochemical toxicity. Where the field currently places the burden of proof on any such claim is surveyed in the current scientific consensus on AI consciousness.
Matt Angle is Founder and CEO of Paradromics. Benjamin Rapoport is Co-founder and Chief Science Officer of Precision Neuroscience. Paradromics received FDA IDE approval for the Connect-One study in November 2025, and the first Connexus implantation was completed at University of Michigan Health. Masataka Watanabe’s split-brain BMI framework appears in From Biological to Artificial Consciousness (Springer, 2022, ISBN 978-3-030-91137-9).
Researchers covered here
- Matt AngleParadromicsFounder and CEO of Paradromics, developer of the Connexus Direct Data Interface high-data-rate brain-computer interface
- Benjamin RapoportPrecision Neuroscience, Mount Sinai Health SystemCo-founder of Precision Neuroscience, founding member of Neuralink, and pioneer of conformable micro-electrocorticography (Layer 7 Cortical Interface)