AI system
- System
-
BrainGate2 intracortical typing neuroprosthesis: six 64-electrode Utah arrays (Blackrock Microsystems) per participant, a five-layer gated recurrent unit decoder trained with connectionist temporal classification loss, and a 5-gram language model implemented as a Kaldi weighted finite-state transducer
- Model
-
Not assessed
- Release
-
Not assessed
- Application developer
-
BrainGate consortium and Neurorecovery, Brown University and VA Providence Healthcare System, with 
Stanford University and UC Davis co-authors - Role
-
Not assessed
- Configuration
-
Not assessed
- Inputs and outputs
-
Not assessed
- Autonomy
-
Not assessed
- Workflow
-
Six 64-electrode Utah arrays per participant placed in precentral gyrus after fMRI-guided parcellation, giving hand-motor electrodes for both hands. A recurrent neural network decoder trained with CTC loss on attempted finger movements, recalibrated from as few as 30 sentences, plus a 5-gram language model. A keyboard setup the participants already knew. Neither was a touch typist, and both learned the layout over sessions. A pilot-trial infrastructure that brings research staff and equipment to the participant's residence.
- Human oversight
-
Not assessed
- Relation to AI
-
The decoder is the component that makes typing possible. It is a recurrent neural network of five gated recurrent unit layers with 512 units each. It was trained with CTC loss on 20 ms bins of threshold crossings and spike-band power. It maps attempted finger movements to characters, and a 5-gram language model corrects them. Without it the arrays only record. The error rate figures are word error rates after the language model. Performance figures come from the developers' own peer-reviewed paper. The paper declares that the MGH Translational Research Center has research support agreements with several BCI companies, for which L.R.H. provides consultative input.
- AI contribution
-
The decoder is the component that makes typing possible. It is a recurrent neural network of five gated recurrent unit layers with 512 units each. It was trained with CTC loss on 20 ms bins of threshold crossings and spike-band power. It maps attempted finger movements to characters, and a 5-gram language model corrects them. Without it the arrays only record. The error rate figures are word error rates after the language model. Performance figures come from the developers' own peer-reviewed paper. The paper declares that the MGH Translational Research Center has research support agreements with several BCI companies, for which L.R.H. provides consultative input.
Benefits
- Outcome identifiers
-
Not assessed
- Benefit
-
T18 reached 110 characters (22 words) per minute with a 1.6% word error rate
- Category
-
Improved Accessibility
- Status
-
Not assessed
- Beneficiary groups
-
People. Society.
- Beneficiaries
-
Not assessed
- Reach
-
Not assessed
- Distribution and access
-
Not assessed
- Duration
-
Not assessed
- Expected?
-
Yes
- Costs and harms
-
Not assessed
Measurement
- Studies
-
Not assessed
- Study design
-
Not assessed
- Measurement
-
Not assessed
- Setting and sample
-
Not assessed
- Measures
-
T18 reached 110 characters (22 words) per minute with a 1.6% word error rate
- Comparison
-
the previous state-of-the-art hand-motor iBCI (handwriting decoding) and able-bodied smartphone typing
- Results
-
Reference: the previous state-of-the-art method, 20 characters per minute slower (90 characters per minute, about 18 words per minute, per Scientific American); able-bodied smartphone typing of 27.3 words per minute for T18's age group. Result: 110 characters per minute (22 words per minute) at a 1.6% word error rate.
- Additional value from AI
-
Not assessed
- Limitations
-
Not assessed
- Denominators
-
Not assessed
- Materials for replication
-
Not assessed
Money
- Kind of value
-
Not assessed
- Amount
-
Not assessed
- Currency and prices
-
Not assessed
- Period and unit
-
Not assessed
- Realization
-
Not assessed
- Recipients and payers
-
Not assessed
- Gross or net
-
Not assessed
- Included costs
-
Not assessed
- Comparison
-
Not assessed
- Method
-
Not assessed
- Return and payback
-
Not assessed
People and places
- Participants
-
Developer:
BrainGate consortium and Neurorecovery, Brown University and VA Providence Healthcare System, with Stanford University and UC Davis co-authors. Operator: BrainGate consortium (BrainGate2 pilot clinical trial; sponsor-investigator Leigh R. Hochberg; FDA Investigational Device Exemption G090003), with research sessions at the participants' residences. - Jurisdictions
-
Developer:
United States. Operator: United States. - Ownership and control
-
Not assessed
- Dependencies
-
Not assessed
- Cross-border connections
-
Not assessed
- Geopolitical implications
-
Not assessed
Follow-up
- Conditions for replication
-
Use was confined to research sessions. The paper describes independent use through online decoder training as future work. Performance differed by participant. T18 reached 22 words per minute with word error below 7% across blocks. T17 reached 47 characters per minute with word error mostly below 15%. Invasive surgery is required. The device is investigational under an FDA investigational device exemption. The paper reports no clinical-trial safety outcomes.
- Replication attempts
-
Not assessed
- Later findings
-
No development after the 16 March 2026 publication was found in the cited sources.
- Availability
-
Not assessed
- Institutional decision
-
Not assessed
- Grounds
-
Not assessed
- Action and result
-
Not assessed
- Success criterion
-
Not assessed
Evidence
- Source numbers
-
AIBD-0204-source-0001–0008
- Citations
-
Numbered references below.
- Language
-
Not assessed
- Source relationships
-
Developer, Independent
- Original evidence
-
Not assessed
- Access and versions
-
Not assessed
- Supported findings
-
Not assessed
- Evidence links
-
Not assessed
- Evidence strength
-
Not assessed
- Missing evidence
-
Not assessed
Grades
- Assessed benefits
-
People. Society.
- Depth
-
People: 2. Society: 1.
- Reasons
-
People: Two identifiable participants typed sentences of their own choosing at up to 22 words per minute. One of them has anarthria and only eye movement left. The typing happened only in research sessions. The paper lists independent use as future work, so the gain is material rather than life-changing. Society: The authors report higher throughput than current state-of-the-art hand-motor iBCI decoding. UCSF's Edward Chang, who is not on the paper, called it among the fastest motor-cortex typing BCIs yet. There were two participants and no population effect.
- Observed or intended
-
Not assessed
- Rubric
-
Not assessed
- Assessor and date
-
Not assessed
- Alternative assessments
-
Not assessed
- Fill
-
Independently reported
Entry details
- Kind
-
Not assessed
- Related entries
-
Not assessed
- Field
-
Neurotechnology / assistive communication. People with tetraplegia from ALS or cervical spinal cord injury.
- Purpose
-
The typing neuroprosthesis was built to restore communication for people with paralysis. It decodes attempted finger movements into keystrokes. The typing the two participants achieved is exactly that purpose.
- Dates
-
Date basis unassessed: 2026 (precision unassessed). Updated: 22 October 2026.
- Stage
-
Not assessed
- Version
-
SHA-256: b3c6dcf2900e…
References
- Other identifiers
-
oecd_aim: 2026-03-16-f051.
- Category mappings
-
Not assessed
- Brief
-
Not assessed
- Revisions
-
6 revisions. Latest: 22 October 2026.
Sources
-
Peer-reviewed study Developer
Jude, J. J., Levi-Aharoni, H., Acosta, A. J., et al. (2026). Restoring rapid natural bimanual typing with a neuroprosthesis after paralysis. Nature Neuroscience 29, 2613–2622, online 16 March 2026.
Typing speed is user-regulated, reaching 110 characters per minute, resulting in 22 words per minute with a word error rate of 1.6%.
-
Peer-reviewed study Developer
All research sessions were performed at each participant's place of residence.
-
News report Independent author
Lewis, T. / Scientific American (2026). Brain implant allows people who are paralyzed to type using their thoughts at speed of texting. Scientific American, 16 March 2026.
At about 22 words per minute, this is among the fastest motor-cortex typing BCIs yet
-
News report Independent author
Horak, M. / ALS News Today (2026). Brain implant decodes finger moves so paralyzed patients can type. ALS News Today, 18 March 2026.
For many people with paralysis, when losing use of both the hands and the muscles of speech, communication can become difficult or impossible.
-
News report Independent author
Europa Press / Infosalus (2026). Una neuroprótesis experimental permite a personas con parálisis escribir con la mente como si usaran un teclado. Infosalus, 17 March 2026.
Para muchas personas con parálisis, al perder el uso de las manos y los músculos del habla, la comunicación puede volverse difícil o imposible.
-
Press release Developer
Mass General Brigham (2026). Investigational Brain Implant Restores High-Speed Communication. Press release as reproduced by Neuroscience News, 16 March 2026.
Both participants used the device from the comfort of their own place of residence, demonstrating the potential for translation and at-home use in the future.
-
Press release Developer
Brown University (2026). Brain computer interface enables rapid communication for two people with paralysis. Release as reproduced by Medical Xpress, 16 March 2026.
BCIs are on track to become an important new alternative to what's currently offered.
-
Other Developer
ClinicalTrials.gov (n.d.). BrainGate2: Feasibility Study of an Intracortical Neural Interface System for Persons With Tetraplegia. NCT00912041, sponsor Leigh R. Hochberg, MD, PhD.
BrainGate2: Feasibility Study of an Intracortical Neural Interface System for Persons With Tetraplegia