Framing AI Images (copy)

Framing AI Images

"Framing AI Images" DALL-E Generated 9/17/23

Overview:

Generative AI tools are well positioned to play a significant role in the future of human thought, work, and creative expression. They also present unique challenges and threats in each of these areas that, when exposed, can strike at the very nature of what it means to be human. These conditions create a need for educational exercises that require both engagement with and critical reflection on AI processes, so that students can experience the potential and limitations of generative AI both on their own terms and as part of a cohort of engaged learners tasked with identifying best use practices. The following discussion pertains to a writing activity for early undergraduates designed to spark interest in the evolving relationship between images and language, and an individual's ability to generate one with the other through technological assistance. The multimodal nature of this activity provides a visual medium for thinking about the constructed and mediated nature of AI content and authorship. There’s an immediate need for an acquisition of this sort of critical framework, as students suddenly find themselves part of the first generation of scholars with near instantaneous access to these powerful new tools. Evidence drawn from student work included below is used to advance discussion on several key topics related to the assignment. You can view a full account of student free writing here.[1]

Assignment:

Ask students to create an image using an original phrase or brief set of phrases as a text prompt, and a free generative AI platform of their choice. Encourage them to take some time to experiment with the different features and permutations of 'prompt language' before settling on an image. Once they settle on an image, have them save or screenshot and upload it to a new document. For the writing assignment, ask them: “In several sentences, describe the image. What do you notice? What is surprising or unexpected? How do the recorded results compare with what you were thinking before generating the image? What, if anything, does the image say about the AI you used to create it or what contexts or sources the AI may have drawn from? To what degree do you think the image is authentic, and why?” These questions, while flexible, are meant to help students gain a more critical understanding and appreciation of the actual source generated from their initial text prompt, in part by using visual analysis tools endorsed by the Library of Congress and in part developed to encourage theorization around issues of authorship. Encourage students to share their work and findings together! A great reflective activity is to ask students about why they think the exercise took them from a small amount of writing (the text prompt) to a much larger amount of writing in the response.

Fostering AI Literacy Through Creative Discovery:

As educators grapple with the promise and perils of AI in and beyond the classroom, one intention of this assignment is to direct students towards an AI utilization strategy in which machine generated content is used to help them build outward from their own ideas, starting with their own originating concept for the image, and one where content is not an end product but an important and problematized starting point. At UC Berkeley, undergraduate Discovery is described as "a journey of engaged creativity and self-actualization" that "foregrounds experiential learning as a doorway to educational purpose and community building." Discovery practices aim to push knowledge building beyond didactic exchanges, and in this case encourage students to foster their own relational understanding with the tools that stand to redefine the nature of thinking and writing. With this in mind, student responses to the assignment offer a range of insight into the mediating role that AI plays in content creation. Some suggest novel takes on the errors, imperfections, or discrepancies/distance between intended outcomes and actual results. Others venture into issues like the nature of expression and the powerful danger of misinformation. This variety of responses is significant if for no other reason than its demonstration of how opinions exist in diversities that AI notoriously flatten or distort, and cannot be relied upon to replicate:

"The images...created by AI, each show a grand castle, as the AI was told to create. Each image shows the castle with a sky with clouds as a background, and some kind of trees or body of water surrounding the castle. None of the images follow exactly what I was looking for in the prompt, which was a grand castle floating in the sky. The grandness of the castle has been accurately captured by the AI, but it failed to depict the castle floating in the sky, untethered to the ground. I tried several different searches, from “a castle on a floating island” to “a castle in the sky” but none of these prompts led to an image that exactly matched the description of a castle floating in the sky. It seems that the AI can only create images of castles on the ground, because those are the only types of images that it has access to" -Example taken from student free write

"The reason why I chose red tulips was because they hold sentimental value to me and I was hoping that my emotions could be reflected in the painting. However, I think the AI failed to do this. When I look at the image, I don't feel some sort of connection even though red tulips themselves are something I connect to. I think the AI fails to be authentic because it doesn't carry the experiences that I have. Authenticity in art to me means a piece that carries the emotions of the artist and the things they have health with. The AI only carries a prompt but it will never be able to convey the real feelings of someone." -Example taken from student free write

"The phrase I used to compute the image was 'someone sitting alone on a swing during lunchtime' ... Often AI is classified as not being able to compute emotion in what it makes. Therefore, I wanted to test out this theory. This is why I chose a phrase that had a bit of a sad connotation because I wanted to be able to see if it could identify a sense of emotion in the image. The AI could have easily chosen a more upbeat and sunlit atmosphere, which would change the connotation of the image into someone playing happily on a swing. However, the weather and the tree shade emphasizes a feeling of loneliness among the boy. I would have believed that the AI would just compute what I asked for without including a focus on the weather or the color distribution in the image. However, I was proven wrong because I do see emotion and feeling in the art. This shows how maybe AI doesn’t explicitly portray the emotion in the work, but from the computation and gathering art from multiple other sources on the internet, it is able to incorporate some feeling in what it makes." -Example taken from student free write

"Ignoring the image’s technical quality and accuracy, Open Art AI has successfully created a fake image of a possibly real event. This concept in itself connects to the greater issue of AI image generators having the ability to plague the internet with fake information. In fact, it already has. When confirming whether or not Joe Biden and Kim Jong Un have ever met or shaken hands before, I found three AI generated images out of the first nine images to appear." -Example taken from student free write

"While this picture is appealing to the eye, it was unable to draw out what I had in mind. The inspiration for my prompt were the long nights of studying I did both at home and at Berkeley. With each passing hour, my eyes would slowly shut and leave an obscure vision for my eyes to see. I would see colorful dots as I closed my eyes, a colorful image that seemed ironic compared to the exhausted student. This story is something that AI couldn’t integrate into its image which makes it different from an authentic image. Something authentic is able to share the creator’s thoughts and emotions. "

"THE AI-GENERATED IMAGE SUCCESSFULLY PORTRAYED what I had envisioned in my head. This image can be interpreted in various ways: humankind looks up at the wonders of the sky and finds a light shining down upon him; is it the sun? Is it a celestial-like message? Divine perhaps? I, for one, see a man looking beyond the earthly objects surrounding him (trees, in this case) and realizing there is more to life than what lies here. The fun part is that it's open to interpretation... [I] believe that this creation is authentic, it's authentic to me. I thought of the idea (emphasis on the “I”), and then the AI was only a means to an end. It brought my creation to reality; it was my tool. Similarly, an artist’s tool is his brush, and a photographer’s tool is his camera. Mine was the AI generator. Who’s to say this is inauthentic?"

Recursive Development, Rhetorical Strategies, and Project Based Learning:

Opinions formed through informal discovery writing like those highlighted above can become points of departure for sophisticated critical inquiry through a 'return and revise' type strategy for idea building that would look familiar within the field of composition studies even before the AI turn, but is also instructive in demonstrating the nature of critical thinking as fundamentally novel in comparison to AI generated content. Simply put, returning to an idea with new questions and with the intention to reconsider, develop, and innovate is how experienced writers often view their process for developing meaningful writing.[2] However, requesting a text or image generator to provide elaboration or draw conclusions from any given text, let alone a text it generated, is often the most expedient way to expose the limitations of its ability to produce valuable or accurate content.[3] Importantly, embedding both peer and instructor intervention between stages of recursive development supports the values of collaborative learning, establishes a less hierarchical model for providing instructive feedback, and creating opportunities for casual and incremental assessment that promotes greater accountability by making AI plagiarism of final products more difficult and less relevant. For example, responses for this assignment, shared online and in classroom breakout groups, became the basis for further class conversations and controlled opportunities for students to rehearse and debate their ideas. Beyond this, students were provided options to continue their work through a guided investigation into different rhetorical situations devised around their initial creation and reflection.

In an optional second part to this assignment, students were asked draw from their image and response to develop one of the argumentative essays required by the course. (To help inform their perspectives, they were encouraged to draw from or respond directly to things like notes generated from class discussions, one of several popular articles on AI, and Denis Dutton’s scholarly work Authenticity in Art, abridged from the Oxford Handbook of Aesthetics.) For an optional third part of the assignment, students have been invited to take their reflections on AI art and authorship to develop their own works of creative multimodal composition. By relying on an iterative, project-based model for introducing the writing process, one that starting with low stakes writing and encourages students to determine the content and direction of their assignments, students are able to more easily to become invested in an idea and develop it in a way that readies them for the kind of discovery experiences that will hopefully define their undergraduate experience.