Even Perfect Decoder-only BCIs will be less precise than motor channels (HANDS) (markusstrasser.org)

A recent argument contends that even a hypothetically perfect, read-only (decoder-only) brain–computer interface (BCI) will be slower and less precise than ordinary motor channels (hands) for tasks that require explicit reference and commitment. Key reasons: muscles provide native “gating” (push-to-submit affordances) that prevents accidental actions, whereas BCIs would need cognitively costly mode-switching to avoid spurious activations; short sessions yield only ~10^2–10^3 informative neural signals, enough for low‑dimensional control but not for recovering a user’s high‑dimensional, idiosyncratic preferences; decoded thoughts poorly bind to external referents (symbol mis‑binding and homonym confusion); and neural statistics drift across sessions, imposing retraining costs that motor output largely avoids. The practical implication for AI/ML and neuroengineering is that read-only BCIs should be evaluated as systems against strong motor baselines and multimodal hybrids (gaze, speech, touch). The author proposes concrete metrics—spurious activation rate, referential error rate, cross‑session throughput decay, and constrained task time—to quantify gating failures, grounding errors, decoder rot, and end‑to‑end task loss. While BCIs can supply continuous implicit signals (error potentials, arousal), the take-away is realistic: zero‑shot brain reading of highly specific preferences is unlikely to match the precision, built‑in throttling, and compositional clarity of hands without costly trade-offs in mental friction.