Two interfaces, back to back
A brain-to-brain interface (BBI) is nothing more exotic than a read-side BCI whose output drives a write-side stimulator on a second person. Brain A's activity is decoded into a message; that message is used to write a stimulus into brain B. The whole system is an encoder-decoder brain pair: a decoder on one head, an encoder on the other, with a wire between them.
Formally the BBI is a composition: a decoder D maps brain-A activity to a discrete message m, and an encoder E writes m into brain B as evoked activity.
The wire matters. In every real system the two brains never touch: the message is a stream of bits carried by an ordinary digital link, often literally the internet. This is why the honest name is brain-computer-brain, or computer-mediated linking — there is no mysterious field coupling one cortex to another. It is also why a BBI is not the same as a bidirectional BCI, which reads from and writes to a single brain.
The founding experiments
Rats came first. In animal brain-to-brain work, an 'encoder' rat's cortical activity was decoded and delivered by cortico-cortical microstimulation into a 'decoder' rat, which used the injected cue to make the correct choice above chance — even with the two animals on different continents, linked over the internet.
Then humans. The first non-invasive human-to-human BBIs decoded a sender's motor imagery from EEG and used it to trigger TMS over a receiver's motor cortex, producing an involuntary hand movement that pressed a game key. A parallel demonstration sent binary-coded words by evoking TMS-induced phosphenes — a flash of light in the receiver's visual field standing for a 1.
Then networks. BrainNet linked three people to cooperatively play a Tetris-like game: two 'senders' encoded a rotate/don't-rotate decision using SSVEP, and a 'receiver', who could not see the bottom row, read the pooled advice as phosphenes and acted. Above chance, three brains solved a task none could solve alone under those constraints.
Why link brains at all?
Three motivations survive scrutiny. First, collaboration: pooling several noisy brains into one better decision (consensus decoding) can beat the best single operator. Second, sensory sharing: a write channel could, in principle, deliver one person's percept to another. Third, and most useful to researchers, a BBI is a clean testbed for the read/write asymmetry — it puts a decoder and an encoder in series and forces you to confront the weaker one.
And the honest scope: every demonstrated channel to date carries only a few bits, slowly, from a pre-agreed alphabet. The science is real but modest. The job of this track is to keep result and aspiration strictly apart.