The honest ledger
Set demonstrated against dreamed. Demonstrated: a few bits, computer-mediated, drawn from a pre-agreed codebook, on toy tasks, above chance, sometimes across continents. Not demonstrated: open-ended thought transfer, high bandwidth, or any direct neural coupling without a computer in the loop. Everything in this track has been an effort to keep the hype on one side of that line.
The write-side ceiling
The read/write asymmetry is most likely a frontier limit, not a temporary inconvenience. Non-invasive write — TMS, focused ultrasound, temporal interference — is fundamentally coarse, and there is a deeper problem beneath the hardware: the semantic write problem. A cortex is not a display you can render an arbitrary experience onto; even high-resolution microstimulation yields only crude, learned percepts.
The end-to-end brain-to-brain rate is bounded by the weaker half — today the non-invasive write side, which sits at bits per minute. This is the asymmetry as a design ceiling.
And the read side is not solved either. We cannot yet decode a full 'language of thought' losslessly; robust non-invasive decoding still leans on a handful of coarse paradigms. Both halves must improve, and by the cascade logic of guide 3, improving only one changes little.
Agency, privacy, and responsibility
BBIs stress ethics in ways single-brain systems do not, precisely because they add a write channel into a second person. Whose action was it — the receiver's agency and sense of agency are directly in play. A write channel can, in principle, insert content, which puts mental privacy and mental integrity and cognitive liberty on the line. And a networked action creates a responsibility gap: when several linked brains cause one outcome, who is accountable?
This is why neurorights and the ethical limits of brain linking are not an afterthought. Consent for a BBI must cover being written to, not just being read — and the coercion risk grows precisely as the write channels improve. The ethics get harder, not easier, with every engineering advance.
What would have to change
For brain-to-brain to matter, three things would have to change together: far higher-resolution, safe non-invasive write; a shared representational alignment so that a decoded state in brain A maps to a meaningful state in brain B (representational alignment, cross-subject transfer); and honest benchmarks with governance to match. None of these is close, and the first may be bounded by biophysics.
An honest reading of the next twenty years: expect incremental gains in collaborative reliability and richer sensory write, not thought-sharing. The most likely near-term multi-brain wins are consensus decoding and shared device control — quietly useful, ethically manageable, and a long way from the cognitive-augmentation headlines. That gap between the useful and the sensational is, in the end, the most important thing to carry out of this track.