Agency & Authorship
Agency and authorship concern whether a BCI-mediated action can properly be attributed to the user as its author. A BCI does not transmit intention transparently: a decoder interprets neural activity through a learned model, shared or assistive control may blend user and machine commands, and language models or autocomplete may extend a user's partial input. The ethical question is how much of the output is genuinely the person's, and how to preserve their standing as author of their own communications and movements.
This matters practically for trust, consent, and accountability. If a speller completes a word the user did not intend, or an assistive controller reaches a target the user did not choose, the resulting act is co-authored by the system — with implications for whether the user endorses it, whether it counts as their speech, and who answers for its consequences.
A P300 speller or language-model-assisted typist that autocompletes 'no' into 'not now, thanks' produces an utterance the user may not fully endorse — the sentence is co-authored, and treating it as purely the user's words can misrepresent them.
Autocompletion shifts authorship from user toward system.
Authorship is a spectrum, not a binary; system design should make the degree of machine contribution transparent and give the user the means to review, correct, or veto outputs before they take effect.