FakeForward: Using Deepfake Technology for Feedforward Learning

Document Title

FakeForward: Using Deepfake Technology for Feedforward Learning

Document Information

• Subject Area: Technology-Assisted Learning and Skill Enhancement
• Keywords: Deepfake, Video Self-Modelling (VSM), Skill Learning, Psychology, Dietary Health, Public Speaking, Deep Learning, Self-Identity, Training Support Tools

Research Background and Problem Statement

• Existing Problems:
  • Video Self-Modelling (VSM) is a learning technique that enhances skill acquisition by allowing learners to watch videos of themselves performing skills in an idealized manner. However, traditional VSM techniques are complex and time-consuming to produce, requiring extensive manual video editing.
  • Deepfake technology offers new opportunities for customizing video content, but its applications are often accompanied by ethical challenges (e.g., data security, malicious use). To date, it has not been fully explored in educational and training contexts.
• Research Significance:
  • This study explores how deepfake technology can reduce the technical and labor barriers of VSM, providing an innovative solution for training and psychological issues in the context of skill development and psychology research.
• Research Motivation:
  • To combine deepfake technology with simplified video processing workflows to meet the personalized needs of skill learning scenarios.
  • To investigate how to overcome the ethical challenges of deepfake technology and turn it into a tool for promoting individual learning and psychological development.

Solution

• Proposed Method:
  • Developed a method called "FakeForward," which uses deepfake technology to replace the learner's face in a high-performing model video, generating realistic skill demonstration videos.
• Innovations:
  • First application of deepfake technology in the field of video self-modelling and learning interventions.
  • Proposed a set of technical and ethical protocols to ensure the effectiveness and safety of FakeForward videos.
  • Enables users to quickly generate personalized training videos that depict the potential future state of their skills, enhancing users' mirroring responses and immersive learning experiences.
• Implementation Steps and Key Technologies:
  • Technical Protocol:
    • Video Collection: Ensure that the model video and the user share similar appearances (e.g., wearing the same clothing, with consistent background and lighting).
    • Deepfake Video Generation: Use DeepFaceLab software to replace faces in the video.
    • Design video processes tailored to various training tasks (e.g., physical exercise or public speaking) to accommodate multiple tasks.
  • Ethical Protocol:
    • The video generation process is strictly limited to user authorization and non-commercial purposes, with data protection technologies to prevent misuse.
    • Verify the legality of video sources to prevent malicious use.

Research Outcomes

• Experiments:
  1. Physical Training Experiment:
     • Hypothesis Validation: Example tasks included squats, planks, etc.
     • Key Results:
       • FakeForward videos significantly improved performance in several physical training exercises (e.g., squats and static core exercises).
       • Results indicate that watching videos with higher self-identity may serve as a potential cognitive driver for improving physical performance, rather than being driven by traditional psychological variables.
       • The quality of video content significantly affects the training effectiveness of deepfake-generated videos.

1. Public Speaking Experiment: