Neurotech Pub What We’ve Got Here Is Failure To Communicate
Nov 23, 2020
Frank Willett, a Stanford postdoc who decoded imagined handwriting; Sergey Stavisky, a Stanford postdoc specializing in neural decoding; Vikash Gilja, UCSD neural engineering professor; and Beata Jarosiewicz, senior neuroscientist with BrainGate experience. They explore calibration and closed-loop control, handwriting-based thought-to-text breakthroughs, motor cortex organization beyond the simple homunculus, and tradeoffs between handwriting and speech-decoding approaches.
AI Snips
Chapters
Transcript
Episode notes
Closed Loop Feedback Enables Rapid Embodiment
- Closed-loop visual feedback dramatically speeds user learning and gives participants immediate confidence that the BCI is working.
- Sergey Stavisky and Beata Jarosiewicz report participants often stop imagining movements and instead “embody” the cursor once they see real-time control.
Handwriting Expands Neural Separability
- Handwriting-like, high-dimensional trajectories spread characters far apart in neural space, improving separability under noisy recordings.
- Frank Willett notes handwriting produced distinct spatiotemporal patterns, enabling much higher accuracy than nearby-key cursor typing with similar SNR constraints.
Invest In User Interfaces To Boost Throughput
- Prioritize UI research (swipe, graffiti, language models) alongside decoding to squeeze real-world speed gains.
- Sergey and Vikash argue point-and-click is versatile, but industry UI teams can adapt swipe/continuous interactions to increase words-per-minute.