Appendix C: Creative courses for music, literature, and art

Back to the Report: Generative Artificial Intelligence for Education and Pedagogy

For creative fields such as art and music, AI was a topic of interest and concern before the release of ChatGPT in November 2022. AI tools that “create” artistic images and “compose” music based on existing images and corpi of musical work have been available longer, and artistic communities are already engaging in robust conversations about how AI can contribute to creation. These conversations have expanded since the release of ChatGPT and the greater attention on AI. We use the more general term “AI tools” here, since it is more appropriate than only large language model (LLM)-based tools such as ChatGPT.

Artistic fields have long standing traditions of discussing what it means for an individual to create “original work”, although this conversation is constantly shifting as it engages with new technology. Disciplinary practices around ideas of authorship and practices for attribution seem well suited to guide faculty and students in discussing whether, how, and when to integrate AI tools into their creative practice, and how to properly and ethically attribute AI generated work. This is not to say that there is consensus or agreement on how questions of authorship or originality should be answered (see the recent Supreme Court Andy Warhol case), but that such conversations fit into prevailing disciplinary discourses. The rest of the academy might look to creative fields for help working through their own disciplinary considerations of how AI tools might change notions of original work, attribution, and creation.

Additionally, students in artistic disciplines are highly motivated to develop the necessary skills needed to create and interpret art. They generally do not need to be convinced of their need to learn specific skills, or to be warned away from tools that may be able to do that work for them.


In addition to LLMs that write prose and poetry, artificial intelligence tools can generate music, images, and video. This provides many ways in which instructors can embrace these tools, building them into assignments (including examples shared below). One typical way is for AI tools to generate source material for the student to then build upon, riff off, interrogate, or extend in their work.

These tools are developing and improving rapidly. Text to image or text to video tools such as Dall-E, Stable Diffusion, MidJourney, and Adobe Firefly, are easily available, with new versions often in beta environments. Runway ML is a platform that links various AI models for artists to experiment with, generating and combining text, images, video, music, and more. Artbreeder allows users to create original artworks by combining and evolving existing art.

In addition to GPT-3 and GPT-4, there are several writing tools that can generate many types of writing, from marketing copy to short stories and more. For example, for creative writing, LAIKA lets you collaborate with an AI partner. LAIKA can be trained to coach you using your own writing, or you can “collaborate with writers” ranging from Fyodor Dostoevsky to Agatha Christie.


Concerns around the use of AI include inequitable access to tools; concerns that overreliance on AI tools built from particular databases can actually limit creative innovation; and risks of appropriation without attribution, including harms to individual creators, such as illustrators and musicians whose individual style is easily co-opted by such tools.

These tools often charge for use beyond a limited amount of free trials. There are also many different AI tools, so keeping up to date with available tools will be a significant time commitment. If Cornell decided to provide AI tools to students to reduce inequity issues, selecting representative AI tools for creative fields would be challenging.

The tools currently available draw from a limited range of human artifacts, skewed toward what has been posted online. Artistic creation that draws only on this content may, over time, limit artistic innovation and narrow the scope of artistic work. Critical engagement with these questions will become an important component of education.

Because this source content is disassociated from the identities of its original creators, it also brings unique risks of appropriation and makes attribution more challenging. AI tools may disrupt how artistic skills are valued and compensated. These tools threaten the livelihood of individual creators (especially illustrators and others whose individual style has been used to inform the development of tools like Dall-E and Midjourney). If used too heavily/without consideration, they can exacerbate imbalances in whose experience is represented by skewing any output toward the existing dataset.


When considering the impact of LLMs, and how to best provide an equitable learning environment for students, it will be important to consider the range of AI tools relevant to creative disciplines, and the extent to which they are integrated into the current workplace.

To allow students time to develop their own design sense/creative perspective, instructors should include assignments that explicitly prohibit the use of GAI tools. Given that students value their individual, creative contributions, instructors should build on this sense of pride and focus on skill-building when encouraging students to avoid overreliance on GAI tools.

Instructors should include critical discussions of the sources of data that are used to build these tools in their pedagogy. There are already conversations around the images that can be elicited from these tools. Such artifacts are interesting in themselves and offer a potential site of critique.

Case studies

We are already seeing the use of GAI in assignments. Some examples of encouraged use in music include:

  • as a generative tool for images, video, and text as poetic materials to respond to musically. For example, students are invited to generate images from poetic input prompts to Midjourney, using them as inspiration for their compositions.
  • as parameter mediation in electronic music, where the AI (or more correctly, a neural network) makes lower/mid-level decisions in a given synthesis system so that the performer/composer can operate on a higher, more "instrumental" level of control.
  • in manipulation of large audio data sets (called a "corpus"), letting AI seek, discover, and organize relationships between and among sounds (or pieces of sounds), so they can then be navigated and accessed by the composer or performer for musical purposes.

When an assignment is designed to teach the use of specific tools, then AI might be prohibited. For example, if a lesson is meant to teach students how to create content “by hand” using a specific digital imaging software, they should not be allowed to use other GAI tools such as text-to-content to produce a similar result.

Attribution is expected in any discussion of the artistic process, and instructors may want to explicitly require students to discuss their experiences interacting with AI tools when used.

Back to the Report: Generative Artificial Intelligence for Education and Pedagogy