AI system
- System
-
UCSF multimodal speech neuroprosthesis (BRAVO trial)
- Model
-
Bidirectional recurrent neural networks trained with connectionist temporal classification loss.
- Release
-
Exact release not established.
- Application developer
-
UCSF Chang Lab with UC Berkeley - Role
-
The decoders converted cortical activity into text, synthesized speech, and avatar movements.
- Configuration
-
The system used a 253-channel array on the brain’s surface and a vocabulary of 1,024 words.
- Inputs and outputs
-
The system converted cortical recordings into text, synthesized speech, and gestures by an avatar.
- Autonomy
-
Not assessed
- Workflow
-
Preserved articulatory representations in speech sensorimotor cortex 18 years after paralysis. High-density ECoG coverage of speech cortex. Deep-learning decoders trained on the participant's attempted speech in under two weeks.
- Human oversight
-
Not assessed
- Relation to AI
-
Deep-learning decoders are the mechanism. They map cortical activity during attempted speech to phonemes, speech units and articulatory gestures, and every output (text, voice, avatar) depends on them.
- AI contribution
-
Deep-learning decoders are the mechanism. They map cortical activity during attempted speech to phonemes, speech units and articulatory gestures, and every output (text, voice, avatar) depends on them.
Benefits
- Outcome identifiers
-
Not assessed
- Benefit
-
decoded text at a median 78 words per minute with 25% median word error rate
- Category
-
Improved Accessibility
- Status
-
Not assessed
- Beneficiary groups
-
People.
- 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
-
decoded text at a median 78 words per minute with 25% median word error rate
- Comparison
-
Not assessed
- Results
-
decoded text at a median 78 words per minute with 25% median 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:
UCSF Chang Lab with UC Berkeley. Operator: UCSF Medical Center. - Jurisdictions
-
Benefit: San Francisco,
United States. 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
-
Participant can attempt speech and retains cortical speech representations. Array remains stable with adequate signal quality. Research infrastructure available for training and real-time decoding.
- Replication attempts
-
2 follow-up entries. Benefit recurred: 2.
- Later findings
-
31 March 2025. The same participant went on to use a streaming synthesis approach, reported in Nature Neuroscience, that cut the decoding delay from about eight seconds per sentence to under a second.
- Availability
-
Not assessed
- Institutional decision
-
Not assessed
- Grounds
-
Not assessed
- Action and result
-
Not assessed
- Success criterion
-
Not assessed
Evidence
- Source numbers
-
AIBD-0003-source-0001–0004
- 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.
- Depth
-
People: 2.
- Reasons
-
People: Material benefit to one identifiable person: fast text, synthesized voice and avatar communication demonstrated in research sessions (run at her residence). The paper reports no independent use outside those sessions.
- 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
-
Both decode speech for people with impaired speech; Whisper fabricated words and violent phrases when transcribing impaired speech.
- Field
-
Healthcare / assistive communication. Post-stroke anarthria.
- Purpose
-
Built to restore communication by decoding attempted speech from cortical surface recordings into text, synthesized speech and avatar animation. That was the intended benefit and is what occurred.
- Dates
-
Date basis unassessed: 2022 (precision unassessed). Updated: 21 October 2026.
- Stage
-
Not assessed
- Version
-
SHA-256: 7d83466f3a2a…
References
- Other identifiers
-
AI Incident Database: 732.
- Category mappings
-
Not assessed
- Brief
-
Not assessed
- Revisions
-
6 revisions. Latest: 21 October 2026.
Sources
-
Peer-reviewed study Developer
Metzger SL, Littlejohn KT, Silva AB, et al. (2023). A high-performance neuroprosthesis for speech decoding and avatar control. Nature.
median rate of 78 words per minute and median word error rate of 25%
-
News report Independent author
Bendix A (2023). Woman with paralysis speaks through an avatar 18 years after a stroke, thanks to a brain implant. NBC News (via Yahoo News).
In 2005, Ann Johnson suffered a stroke that left her severely paralyzed and unable to speak. She was 30.
-
News report Independent author
Hamilton J (2023). These experimental brain implants can restore speech to paralyzed patients. NPR (via BPR).
There is this aspect to it that is, to some degree, restoring identity and personhood.
-
Press release Developer
Ellery M. Brain-to-voice neuroprosthesis restores naturalistic speech. Berkeley Engineering, 31 March 2025.