The pop song generator: designing an online course to teach collaborative, creative AI
Project Overview
The document outlines the development and assessment of an innovative online course aimed at harnessing generative AI for creativity in education, specifically through the creation of pop songs. Utilizing advanced AI technologies such as GPT-2 for lyric generation, Music-V AE for music composition, and Diffsinger for synthesizing singing voices, the course employs a constructivist, hands-on pedagogical approach that emphasizes creativity and practical application. Through both qualitative and quantitative evaluations, including feedback from students and experts, the course has yielded valuable insights for enhancing its content and delivery. It also addresses the inherent challenges of teaching AI in creative settings, while highlighting the importance of incorporating ethical considerations into AI education. Overall, the course exemplifies how generative AI can be effectively integrated into educational frameworks to foster creativity and innovation among learners.
Key Applications
Pop song generator using AI models (GPT-2, Music-V AE, Diffsinger)
Context: Online course for undergraduate Computer Science students at Goldsmiths, University of London, with plans for wider access through a MOOC format.
Implementation: Implemented as part of a 20-week AI course on Coursera, structured into five weeks focusing on different components of the pop song generator.
Outcomes: Students learn to create a complete pop song, engage with AI models, and develop creative and technical skills. Quantitative analysis shows student engagement and the course's effectiveness.
Challenges: Technical complexity in understanding AI systems, resource-intensive requirements for running models, and varied student backgrounds leading to difficulties in engagement.
Implementation Barriers
Technical Barrier
AI systems can be hard to describe and understand due to their complex structure and parameters. They often require powerful hardware to run.
Proposed Solutions: Use simplified models (like GPT-2 instead of GPT-3), provide virtual labs to eliminate setup complexities, and target the course design to be user-friendly.
Pedagogical Barrier
Teaching AI effectively to a diverse audience (including non-technical students) is challenging due to the complexity of the subject matter.
Proposed Solutions: Adopt a constructivist, hands-on learning approach, and design the course to accommodate both technical and creative perspectives.
Ethical Barrier
There is a lack of awareness regarding ethics in AI, especially concerning copyright and the implications of using AI in creative practices.
Proposed Solutions: Incorporate content on ethics and legal implications, emphasizing the responsibility of students when using AI technologies.
Project Team
["Matthew Yee-king", "Andrea Fiorucci", "Mark d"Inverno"]
Researcher
Contact Information
For information about the paper, please contact the authors.
Authors: ["Matthew Yee-king", "Andrea Fiorucci", "Mark d"Inverno"]
Source Publication: View Original PaperLink opens in a new window
Project Contact: Dr. Jianhua Yang
LLM Model Version: gpt-4o-mini-2024-07-18
Analysis Provider: Openai