Overview
- In tests with two participants, one typed about 110 characters per minute (around 22 words per minute) with a 1.6% word error rate, while the other reached roughly 47 characters per minute at about 81% accuracy.
- The iBCI records signals from microelectrode arrays in the motor cortex and uses AI to map intended finger and hand movements to letters on a standard keyboard, supported by a predictive language model.
- Calibration required as few as 30 sentences, and both participants used the system from home, indicating potential for translation to at-home use.
- Performance differences aligned with electrode placement and count in the dorsal precentral gyrus and with participants’ diagnoses, with T18’s six arrays (about 384 electrodes) enabling faster, more stable typing than T17.
- Researchers and external experts caution the approach remains early-stage and invasive with ongoing calibration needs, and no similar device has U.S. FDA approval even as China has authorized an invasive BCI.