Current BCIs fit discrete choice, not replacing continuous pointing or text input

What it is

Wired to a task, current BCIs match “pick one of a few options”: lights, call a nurse, step down a menu. They do not match continuous pointing of the mouse kind, or a character stream of the keyboard kind. When the task topology is discrete choice, low bandwidth and coarse classes can still work. When it is a trajectory or text, the channel fails in control structure first, not only in speed.

Why it happens

Pointing needs ongoing two-dimensional error correction: opposite velocity as soon as error appears. Scalp-separable dimensions are few and slow; closing that loop oscillates or only hits large blocks. Text’s information demand is letter entropy; even with a language model it still wants a stable symbol stream. P300 spellers use matrix flash and temporal averaging and land at a few characters per minute, while occupying vision. Discrete choice quantizes the world into slots first, sends (\log_2 N) bits per step, and can raise confidence by stacking trials without holding a continuous servo. Language prediction can shrink the number of slots; it cannot turn the channel into a keyboard—when prediction is wrong, repair still spends those scarce selections.

Studying it

Run three tasks on the same user and electrode set: N-choose-1, two-dimensional Fitts, sentence copy. Outcomes: throughput, time per correct result, mid-task abandonment. Split out the language model: character rate with prediction off is what the channel itself is. Counting “a sentence was eventually produced” as text-input success bills prediction and caregiver help to BCI capability.

Where it stops holding

Invasive systems have produced usable cursors and substantial character rates in some participants; that is task-topology expansion bought with surgery, not a roadmap for scalp products. Gaze spelling and switch scanning are often already faster than non-invasive EEG on discrete choice; BCI becomes a reasonable text candidate only when those channels also fail. Using EEG as an “energy bar” in a game is not a pointing substitute and should not enter a capability claim.

Applying it

• Rewrite the product task as a few mutually exclusive options before considering EEG; leave pointing and typing to hand, eye, or switch.
• If spelling is offered, quote capability as character rate with prediction off, and keep a scanning or gaze fallback.
• If a continuous cursor exists only for demos, keep it out of unassisted daily flow.
• Verify by comparing EEG menu choice, gaze choice, and switch scanning on the same communication script with the same users, not only a successful copy-typing video.

Related

• Same group: C9.07.1 Non-invasive EEG can reliably separate only a few coarse intents, not continuous fine control · C9.07.2 Invasive electrodes have substantially higher bitrate, but surgical risk confines them to medically necessary settings · C9.07.4 Capability bounds move with signal-processing progress; they are not fixed physical limits
• Adjacent: C1.15 Pointing Device Metrics and Throughput · C8.02 Dwell Selection
• Search: discrete BCI selection · P300 speller · Fitts law BCI