The anatomy of a hype cycle
Neurotechnology runs a familiar loop: a genuine result, an extrapolation to a limitless future, funding and press that reward the extrapolation, and a quiet correction when the engineering bites. These contested and overstated claims are not usually fraud — they are a real finding stretched past its evidence. The skill is not cynicism but calibration: hold the finding and the extrapolation apart.
More electrodes is not more mind
The most common technical overreach follows directly from Guide 1. It takes Stevenson's law — the real exponential in channel count — and silently assumes decodable cognition rides the same curve. But information saturates: past the dimensionality of the signal you are decoding, extra channels mostly add redundancy. A thousandfold more electrodes does not imply a thousandfold richer readout of thought.
The fallacy in one line. Channels may grow exponentially (left) while the marginal information per channel falls toward zero (right). Both are empirically supported; equating them is the error. Scaling is necessary for richer BCIs but nowhere near sufficient.
The limits of mind-reading
Can a BCI read your thoughts? The honest answer turns on the unsolved neural code problem and on the limits of semantic and inner-speech decoding. Recent work reconstructing the gist of imagined language from non-invasive recordings is a genuine advance — and, read carefully, a lesson in limits: it needs many hours of per-subject cooperative training, it recovers coarse meaning not verbatim thought, and it collapses if the subject simply resists.
So the accurate summary is neither dismissal nor telepathy. Cooperative, calibrated, coarse-grained decoding of language and intention is real and improving. Covert, verbatim, cross-subject mind-reading of the unwilling is not supported by any published result, and the read/write and coding problems suggest it will remain hard for reasons deeper than data volume. Name both halves precisely.
Telepathy, memory upload, and brain emulation
The grandest claims deserve the plainest audit. Demonstrated brain-to-brain interfaces transmit on the order of one bit per trial through a slow, cued channel — a real proof of principle, not thought-sharing. There is no published system that writes a specific declarative memory into a human brain; the write problem and our ignorance of the memory code place that firmly in speculation.
And whole-brain emulation — scanning a brain and running it as software — is the clearest case of speculation dressed as timeline. Its feasibility rests on unproven premises: that a static connectome suffices (it likely does not — molecular, glial, and dynamical state matter), and that the relevant computation is capturable at all. It may be physically permitted; it is not, on current knowledge, an engineering roadmap. Say so cleanly, and reserve the word soon for things with a mechanism.