Large language models integrated into brain-computer interfaces for communication and control: a systematic review.

Area of Science:

• Neuroscience and Artificial Intelligence
• Human-Computer Interaction
• Assistive Technology

Background:

• Large language models (LLMs) are increasingly integrated with brain-computer interfaces (BCIs) to enhance communication and control systems.
• Current BCI-LLM systems exhibit considerable diversity in their architectural design, data acquisition methods, and performance evaluation.

Purpose of the Study:

• To systematically review and categorize existing studies combining LLMs with BCIs for communication or control.
• To identify common integration patterns, hardware/software pipelines, and LLM prompting strategies.
• To assess the reporting quality of latency, outcomes, and risk of bias in current BCI-LLM research.

Main Methods:

• A systematic review adhering to PRISMA guidelines was conducted on eleven studies integrating LLMs with BCIs.
• Studies were analyzed for their BCI paradigms (e.g., P300, SSVEP), LLM integration patterns, hardware, decoding pipelines, and prompting strategies.
• Risk of bias was assessed using an adapted ROBINS-I framework, categorizing studies into online, offline-simulation, and system-proposal.

Main Results:

• Five distinct BCI-LLM integration patterns were identified: autocomplete, post-edit correction, intent expansion, dynamic interface generation, and affective support.
• Performance varied, with copy-spelling tasks showing over 60% keystroke savings and an intent-based ALS task achieving 42 characters per minute with 88% semantic accuracy.
• Significant limitations include the absence of motor-impaired participants in all studies, reliance on remote LLM endpoints in most cases, and sparse reporting of end-to-end latency and failure modes.

Conclusions:

• A novel taxonomy for BCI-LLM integration and a checklist for future reporting standards are proposed.
• Distinguishing between supported findings and speculation is crucial for advancing the field.
• Further research is required to demonstrate tangible clinical benefits for individuals with motor impairments.