Speech BCI will restore communication for locked-in patients within 3 years at clinically meaningful accuracy

Two converging breakthroughs make this thesis viable. BIT achieves 10% WER (down from 24.69%) by aligning neural embeddings with audio LLM representations. Stanford demonstrates inner speech decoding from motor cortex, showing that thought patterns are attenuated versions of attempted speech. Together, these open the path from attempted-speech decoding to true thought-to-text communication.

Confidence: 6/10 Supporting evidence:

• BIT framework: 10% WER brain-to-text, down from 24.69% SOTA Evidence: strong (BIT)
• Contrastive learning aligns neural embeddings with audio LLM representations — elegant bridge Evidence: strong (BIT)
• Stanford inner speech decoding from motor cortex, 4 patients Evidence: strong (Stanford)
• Inner speech patterns are attenuated versions of attempted speech — existing BCIs can adapt Evidence: strong (Stanford)
• Synchron's minimally invasive approach could expand patient eligibility Evidence: moderate (Synchron)

Challenging evidence:

• Generalization across patients and recording modalities is unproven
• Real-time performance for practical communication not yet demonstrated
• Open vocabulary vs. constrained vocabularies — clinical use requires open vocabulary
• Privacy implications of inner speech decoding are largely unexplored
• Integration with minimally invasive or non-invasive hardware is an open problem
• "Clinically meaningful" requires >90% accuracy for reliable communication — 10% WER is there but only in controlled settings

Evolution:

• Apr 5, 2026 — Initial thesis at 6/10. The BIT result is remarkable and the Stanford inner speech finding is the conceptual breakthrough that connects these systems to locked-in patients. "Within 3 years" means clinical trials demonstrating practical communication by 2029, not commercial availability. The audio LLM bridge pattern could accelerate progress faster than traditional signal processing approaches.

Depends on: speech-bci, neural-signal-decoding, invasive-vs-noninvasive-bci Would change if: BIT achieves <5% WER in a multi-patient study, or if inner speech decoding proves unreliable outside Stanford's specific electrode placement protocol.