Predicting rock-paper-scissors choices based on single-trial EEG signals

Area of Science:

• Neuroscience
• Artificial Intelligence
• Cognitive Science

Background:

• Single-trial electroencephalogram (EEG) analysis is crucial for real-time decision prediction systems.
• Existing methods often average EEG signals, predicting general tendencies rather than specific choices.
• Predicting individual choices in single trials is essential for human-AI interaction.

Purpose of the Study:

• To develop a method for predicting single-trial choices using EEG signals during a multichoice decision-making task.
• To introduce and validate a novel algorithm for feature extraction from EEG data.
• To assess the feasibility of real-time decision prediction in a dynamic game environment.

Main Methods:

• Utilized a rock-paper-scissors game with 40 participants playing 330 trials.
• Proposed the Common Spatial Pattern-Attractor Metagene (CSP-AM) algorithm to extract features from various EEG frequency bands.
• Employed a multilayer perceptron classifier to predict choices based on extracted CSP features.

Main Results:

• The CSP-AM algorithm successfully extracted relevant features from EEG signals during decision-making.
• A multilayer perceptron classifier achieved significantly above-chance accuracy in predicting single-trial choices for 88.57% of participants.
• Validated the effectiveness of CSP features for multichoice decision prediction.

Conclusions:

• The CSP-AM algorithm demonstrates strong potential for predicting specific choices from EEG signals.
• This approach advances the development of proactive AI systems capable of understanding and anticipating human decisions.
• Highlights the value of single-trial EEG analysis for real-time neuroadaptive technologies.