Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Human-AI Teaming Using Large Language Models: Boosting Brain-Computer Interfacing (BCI) and Brain Research

Maryna Kapitonova, Tonio Ball

TL;DR

The paper argues that human-AI teaming, not full autonomy, is the most promising path for AI in brain research and BCI. It introduces the Janusian Design Principles and the ChatBCI toolbox to enable bidirectional collaboration between humans and LLMs, with PyTorch-based processing and GPT-4o interfaces. Using the BCI Competition IV2a EEG dataset, it demonstrates how the AI can generate ideas, assist data exploration, and co-create a CNN decoder and training loop, achieving faster, more interpretable iterations than purely human workflows. The study highlights how domain knowledge can be transferred to AI agents, outlines potential extensions to AutoML for brain signals, and discusses long-term implications for a “brain-grokking AI” that deepens understanding of human brain function.

Abstract

Recently, there is an increasing interest in using artificial intelligence (AI) to automate aspects of the research process, or even autonomously conduct the full research cycle from idea generation, over data analysis, to composing and evaluation of scientific manuscripts. Examples of working AI scientist systems have been demonstrated for computer science tasks and running molecular biology labs. While some approaches aim for full autonomy of the scientific AI, others rather aim for leveraging human-AI teaming. Here, we address how to adapt such approaches for boosting Brain-Computer Interface (BCI) development, as well as brain research resp. neuroscience at large. We argue that at this time, a strong emphasis on human-AI teaming, in contrast to fully autonomous AI BCI researcher will be the most promising way forward. We introduce the collaborative workspaces concept for human-AI teaming based on a set of Janusian design principles, looking both ways, to the human as well as to the AI side. Based on these principles, we present ChatBCI, a Python-based toolbox for enabling human-AI collaboration based on interaction with Large Language Models (LLMs), designed for BCI research and development projects. We show how ChatBCI was successfully used in a concrete BCI project on advancing motor imagery decoding from EEG signals. Our approach can be straightforwardly extended to broad neurotechnological and neuroscientific topics, and may by design facilitate human expert knowledge transfer to scientific AI systems in general.

Human-AI Teaming Using Large Language Models: Boosting Brain-Computer Interfacing (BCI) and Brain Research

TL;DR

The paper argues that human-AI teaming, not full autonomy, is the most promising path for AI in brain research and BCI. It introduces the Janusian Design Principles and the ChatBCI toolbox to enable bidirectional collaboration between humans and LLMs, with PyTorch-based processing and GPT-4o interfaces. Using the BCI Competition IV2a EEG dataset, it demonstrates how the AI can generate ideas, assist data exploration, and co-create a CNN decoder and training loop, achieving faster, more interpretable iterations than purely human workflows. The study highlights how domain knowledge can be transferred to AI agents, outlines potential extensions to AutoML for brain signals, and discusses long-term implications for a “brain-grokking AI” that deepens understanding of human brain function.

Abstract

Recently, there is an increasing interest in using artificial intelligence (AI) to automate aspects of the research process, or even autonomously conduct the full research cycle from idea generation, over data analysis, to composing and evaluation of scientific manuscripts. Examples of working AI scientist systems have been demonstrated for computer science tasks and running molecular biology labs. While some approaches aim for full autonomy of the scientific AI, others rather aim for leveraging human-AI teaming. Here, we address how to adapt such approaches for boosting Brain-Computer Interface (BCI) development, as well as brain research resp. neuroscience at large. We argue that at this time, a strong emphasis on human-AI teaming, in contrast to fully autonomous AI BCI researcher will be the most promising way forward. We introduce the collaborative workspaces concept for human-AI teaming based on a set of Janusian design principles, looking both ways, to the human as well as to the AI side. Based on these principles, we present ChatBCI, a Python-based toolbox for enabling human-AI collaboration based on interaction with Large Language Models (LLMs), designed for BCI research and development projects. We show how ChatBCI was successfully used in a concrete BCI project on advancing motor imagery decoding from EEG signals. Our approach can be straightforwardly extended to broad neurotechnological and neuroscientific topics, and may by design facilitate human expert knowledge transfer to scientific AI systems in general.
Paper Structure (8 sections, 5 figures, 1 table)

This paper contains 8 sections, 5 figures, 1 table.

Table of Contents

  1. Introduction
  2. ChatBCI toolbox overview and Janusian Design Principles
  3. Analyzing the BCI competition IV2a dataset using ChatBCI
  4. Idea generation
  5. Experimental interation
  6. Exploratory data analysis and data validation
  7. LLM-generated deep neural network decoder and training loop
  8. Conclusions

Figures (5)

  • Figure 1: A conceptual illustration of AI-driven scientific research processes for Brain-Computer Interfacing (BCI) and brain research, based on the concept as proposed by lu2024ai_scientist. In contrast to their concept, we consider numerical results and data visualization as an integral part of the experimental iteration loop. Also in contrast to their fully autonomous approach, we support flexible levels of shared autonomy for each phase of the cycle (color coded to reflect the example BCI project as reported here; automatic paper write-up was not addressed)
  • Figure 2: In the Janusian Vision in designing human-AI workspaces, we embrace a dual-facing approach: one face directed toward empowering human expertise, and the other toward amplifying AI capabilities (DALL-E rendering of this idea).
  • Figure 3: ERP waveforms across all trials of all subjects' training data. Trial timing: The cue in the form of an arrow pointing either to the left, right, down or up, corresponding to one of the four classes left hand, right hand, foot or tongue) appeared and stayed on the screen for the duration indicated by the black box (1.25 s). Grey box: Time window of fixation cross presentation.
  • Figure 4: (A) Zoom-in from Fig. \ref{['fig:erp_figure']}. (B) Same with 4-Hz high-pass filter. (C) Direction of the arrows used as cues for the 4 classes and colored corresponding to (A) and (B); placement of the EOG1 and EOG3 channels as described in bci_competition_iv_2a.
  • Figure 5: (A) Zoom-in from Fig. \ref{['fig:erp_figure']}. (B) Same with 4-Hz high-pass filter. (C) Direction of the arrows used as cues for the 4 classes and colored corresponding to (A) and (B); placement of the EOG1 and EOG3 channels.