Limits of semantic and inner-speech decoding

Decoding meaning, whether inner speech, imagined language, or the gist of thought, is qualitatively harder than decoding overt movement. Non-invasive semantic decoders, for example fMRI-based reconstructions of the meaning of heard or imagined narratives, recover coarse gist rather than verbatim wording, are trained per-subject, and typically require cooperation, degrading when the subject engages a simple mental countermeasure. Invasive speech BCIs decode attempted articulation well, but they tap the motor plan, not free unspoken thought.

Several barriers look fundamental rather than merely technical: inner speech is sparse, idiosyncratic, and not reliably tied to a fixed neural substrate; there is no ground-truth label for a private thought; and covert cognition is intertwined with attention and volition, so a subject can withhold or corrupt the signal. This bounds both the promise and the privacy threat of thought-reading, and it is a place where honest technical limits and neurorights concerns meet.