Investigating User Perceptions, Ethical Challenges, and Expectations Regarding Brain-Computer Interfaces (BCIs) for Imagined Speech: A Mixed-Methods Protocol

1. Introduction & Theoretical Framework

1.1 The Evolution of Brain-Computer Interfaces (BCIs)

Brain-Computer Interfaces (BCIs) represent a revolutionary technological breakthrough in neuroengineering, enabling direct communication between the human brain and external computing systems without involving peripheral nerves or the muscular system. The historical development of BCIs began with simple systems that exploited visually evoked potentials (such as the P300 speller paradigm) or motor imagery to control a cursor or artificial limbs.

However, the transfer of these systems to the field of communication faced significant limitations in terms of Information Transfer Rate (ITR) and the naturalness of interaction. The need for faster, spontaneous, and intuitive communication has led modern neurotechnology to focus on Imagined Speech — the process by which an individual mentally reproduces words, phrases, or phonemes without moving the vocal tract and without producing an acoustic signal.

1.2 Neurophysiological Basis of Imagined Speech

Imagined speech relies on the hypothesis of shared neural infrastructure between internal (covert) and overt speech. Neuroimaging studies (fMRI, MEG, high-density EEG) demonstrate that during imagined speech a broad neural network is activated, including:

• Broca's area (left inferior frontal gyrus) for linguistic programming.
• The supplementary motor area (SMA) for planning articulation.
• Wernicke's area and the superior temporal gyrus for auditory processing and feedback.
• The sensorimotor cortex corresponding to the articulatory organs (lips, tongue, larynx).

Despite these similarities, the electrophysiological signals of imagined speech are extremely weak, non-stationary, and characterized by a low Signal-to-Noise Ratio (SNR), especially when recorded using non-invasive methods such as electroencephalography (EEG).

[Neural Programming] ──> [Broca's Area / SMA] │ ▼ [Weak EEG/ECoG Signals] <── [Sensorimotor Cortex] │ ▼ [Low SNR & Noise] ──> [Algorithmic Decoding Required]

1.3 Technological Challenges & Algorithmic Decoding

Decoding imagined speech is among the most difficult problems in machine learning. The main technical challenges include:

1. Inter-subject and Intra-subject Variability: Brain signals differ significantly from person to person, but also within the same individual depending on fatigue, concentration, and emotional state.
2. Continuous vs. Discrete Decoding: The transition from recognizing individual words (a discrete vocabulary of 5–10 words) to continuous, syntactic speech requires complex language models.
3. Deep Learning Architectures: The use of Convolutional Neural Networks (CNNs), Recurrent Networks (LSTM/GRU), and more recently Transformer architectures (Self-Attention) has improved classification performance, but these models require enormous volumes of training data.

1.4 The Research Gap: Human-Centered Design & Neuroethics

While technological research focuses almost exclusively on increasing algorithmic accuracy rates, there is a significant research gap concerning User Acceptance, User Experience (UX), and Ethical Dimensions (Neuroethics).

The ability to decode internal thought raises unprecedented questions:

• How can it be ensured that a BCI will not translate involuntary thoughts (the "Midas Touch" problem in inner speech)?
• What is the tolerable level of cognitive load and training time for a patient?
• How is "Mental Privacy" protected from commercial exploitation or unwanted access to neurodata?

The present research proposal was designed to address precisely this gap, proposing a comprehensive research protocol for evaluating the human and ethical parameters of imagined-speech BCIs.

2. Purpose and Research Questions

The primary purpose of the proposed study is to investigate the expectations, tolerance thresholds, privacy concerns, and interaction needs of potential users (patients and healthy individuals), caregivers, and neurotechnology experts regarding imagined-speech BCI systems.

Research Questions (RQs):

• RQ1: What are users' preferences regarding the type of interface (non-invasive vs. invasive), and what is the minimum acceptable accuracy threshold they require in order to adopt the technology?
• RQ2: How does cognitive fatigue and calibration time affect the desire to use the system in daily life?
• RQ3: What are the main ethical concerns regarding the "involuntary decoding" of inner thought, and how can intentionality-confirmation mechanisms ("Intentionality Gates") be designed?
• RQ4: What institutional and technical safeguards (e.g., on-device processing, neuro-rights) do neuroethics and legal experts consider necessary to protect neurodata?

3. Proposed Research Methodology

To holistically address the research questions, a Sequential Explanatory Mixed-Methods Design is proposed. The study will be divided into two main phases: a large-scale quantitative phase and an in-depth qualitative phase.

┌────────────────────────────────────────────────────────┐ │ PHASE 1: Quantitative Research (Survey/Questionnaire) │ │ • Sample: N = 300 (General Population, Patients) │ │ • Goal: Statistical mapping of preferences/fears │ └───────────────────────────┬────────────────────────────┘ │ ▼ ┌────────────────────────────────────────────────────────┐ │ PHASE 2: Qualitative Research (Semi-structured │ │ Interviews) │ │ • Sample: N = 20 (Experts, Patients, Legal Professionals) │ │ • Goal: Thematic analysis of ethical & UX dilemmas │ └───────────────────────────┬────────────────────────────┘ │ ▼ ┌────────────────────────────────────────────────────────┐ │ SYNTHESIS: Human-Centered Design Framework (BCI-UX) │ └────────────────────────────────────────────────────────┘

3.1 Phase 1: Quantitative Research (Online Survey/Questionnaire)

3.1.1 Sample Design

It is proposed to recruit N = 300 participants through stratified sampling so that three key stakeholder groups are represented:

1. Group A (n=180): General population with familiarity with technology.
2. Group B (n=80): Students and professionals in Health, Neuroscience, and Informatics.
3. Group C (n=40): Patients with motor/speech disorders (e.g., ALS, stroke) and their primary caregivers.

3.1.2 Data Collection Instrument Structure (Survey)

The questionnaire will consist of 20 weighted questions, structured around 4 thematic axes:

• Axis 1: Demographics & Technological Familiarity (age, familiarity with BCIs/AI).
• Axis 2: Technical Preferences & Error Tolerance (Likert scales 1–5 for acceptable accuracy ranging from 70%–99%, preference for EEG cap vs. invasive microelectrodes, tolerable daily training time).
• Axis 3: UX & Cognitive Load (forced-choice questions regarding acceptance of mental fatigue versus communication speed).
• Axis 4: Mental Privacy & Fears (ranking of risks: involuntary decoding, cloud data leakage, cost).

3.1.3 Statistical Analysis

The quantitative data will be analyzed using SPSS/R software. The analysis will include:

• Descriptive statistics (frequencies, means, standard deviations).
• Inferential statistics (ANOVA and Chi-Square tests) for comparative analysis of responses across the three sample groups.
• Multiple Logistic Regression analysis to identify the factors predicting intention to adopt BCI technology.

3.2 Phase 2: Qualitative Research (Semi-structured Interviews)

3.2.1 Sample Design

To gain an in-depth understanding of the social and ethical implications, it is proposed to conduct N = 20 semi-structured interviews with specialized participants:

• 6 Neuroengineering researchers & BCI algorithm designers.
• 7 Patients with severe speech disability & caregivers.
• 4 Neuroethics experts (bioethicists) & technology law specialists.
• 3 Neurologists / physiatrists.

3.2.2 Interview Guide & Thematic Axes

The interviews will last 45–60 minutes and will cover the following specialized topics:

1. The Inner Voice Problem: How can the "thinking voice" (inner monologue) be distinguished from the "message intended for transmission"?
2. Safety Mechanisms (Intentionality Gates): Which interaction solutions (e.g., mental cues/triggers, gaze-based confirmation) are considered functional?
3. Neurodata Governance: What are the requirements for local processing (Edge AI), and how should "Neuro-rights" be defined at the legislative level?

3.2.3 Qualitative Analysis

The audio recordings of the interviews will be fully transcribed. Thematic Analysis following Braun & Clarke (2006) will be applied using NVivo software. The coding process will include:

• Inductive and deductive thematic coding.
• Creation of a thematic map.
• Inter-coder reliability assessment.

4. Expected Outcomes & Contribution to Research

Implementation of this research proposal is expected to yield significant findings that will influence both the academic sphere and the neurotechnology industry.

4.1 Theoretical Contribution

1. Development of a BCI-UX Framework: Creation of a theoretical User Experience model specific to imagined-speech interfaces, linking algorithmic accuracy (%) with the user's emotional and cognitive burden.
2. Development of a Neuroethical Protocol: Clear definition of the boundaries between permissible decoding and violation of mental privacy.

4.2 Technical & Design Guidelines

The research findings will be translated into specific design guidelines for BCI engineers:

• Error Tolerance Thresholds: Determination of the minimum algorithmic accuracy required before a device can be commercially released.
• "Intentionality Gate" Architecture: Proposal of specific interaction patterns (e.g., combining imagined-speech EEG with eye-movement-based confirmation — Hybrid BCI) to eliminate involuntary decoding.
• On-Device Data Protocols: Technical specifications for running deep learning algorithms locally on device hardware, preventing the transmission of raw neurodata to the cloud.

5. Implementation Timeline & Work Program

The research proposal is designed for implementation over a period of 12 months, divided into 4 main Work Packages (WPs):

Work Package (WP)	Description of Activities	Duration
WP1: Preparation & Ethics Approval	Literature review, finalization of the questionnaire & interview guide, submission of the file to the Research Ethics Committee (REC).	Months 1–2
WP2: Quantitative Data Collection & Analysis	Conducting the online survey (N=300), data cleaning, statistical analysis using SPSS/R.	Months 3–5
WP3: Qualitative Data Collection & Analysis	Conducting 20 interviews, transcription, Thematic Analysis in NVivo.	Months 6–9
WP4: Synthesis, Writing & Publication	Integration of findings, formulation of BCI-UX guidelines, writing of the final report and scientific publications.	Months 10–12

6. Ethical Parameters & Assurance of Research Ethics

Because this research touches on sensitive personal data and vulnerable population groups (patients with speech disabilities), the following ethical protocols will be strictly observed:

1. Informed Consent: All participants (or their legal representatives) will receive a detailed information form regarding the purposes of the research and their right to withdraw at any time.
2. Anonymization & Pseudonymization: Quantitative data will be collected completely anonymously. In the interviews, pseudonyms will be assigned and any identifying information will be removed.
3. GDPR Compliance: Data will be stored on encrypted servers of the academic institution, accessible only to the research team.

During the preparation of this work, the author utilized AI tools to assist in formatting of the research protocol. The conceptualization of the research framework, critical evaluation and final review of the content were conducted entirely by the author.

7. References

• Braun, V., & Clarke, V. (2006). Using thematic analysis in psychology. Qualitative Research in Psychology, 3(2), 77-101.
• Martin, S., Brunner, P., Holdgraf, C., Heinze, H. J., Crone, N. E., Rieger, J. W., ... & Knight, R. T. (2018). Decoding imagined speech from intracranial recordings in human cortex. Frontiers in Human Neuroscience, 12, 254.
• Nieto, N., Peterson, V., Wright, H., Baelam, R., & Kamienkowski, J. E. (2022). Thinking out loud, an open-access EEG dataset for inner speech decoding. Scientific Data, 9(1), 52.
• Wolpaw, J. R., Birbaumer, N., McFarland, D. J., Pfurtscheller, G., & Vaughan, T. M. (2002). Brain-computer interfaces for communication and control. Clinical Neurophysiology, 113(6), 767-791.
• Yuste, R., Goering, S., Arcas, B. A. Y., Bi, G., Carmena, J. M., Carter, A., ... & Wolpaw, J. (2017). Four ethical priorities for neurotechnologies and AI. Nature, 551(7679), 159-163.