Text-to-Image

Prompt given to Copilot:

Create an image showing the way AI processes information and turns it into an image.

Copilot image produced:

Second prompt given to Copilot:

Create an image of the AI actually turning the information it’s given into an image, showing the process.

Copilot image produced:

Third prompt given to Copilot:

Create an illustration showing an AI system transforming a text prompt into an image. The scene should include a human user typing a prompt into their computer, the AI represented as a network (data and symbols), and break down the way that AI processes that information and turns it into the image.

Copilot image produced:

Reflection:

I initially gave Copilot a general prompt because I wanted to see what it would depict as a starting point. I told it to create an image showing the way AI processes information and turns it into an image. As a result, it depicted AI (itself) as a robot that creates an image of mountains and the sun, and, using its finger, it points to the image on the screen. It is a simplistic depiction that closely follows the instructions of a simple and straightforward prompt, giving the general idea of how AI might work.

For the second prompt, I continued with simple and straightforward instructions, again asking the image to depict the AI turning the information it is given into an image. The result differed from the first because there was “noise” and a diffusion process depicted. However, the image did not actually explain what the “noise” was or how it functioned. While it appeared slightly more technical, it still lacked clarity in terms of explaining the process.

In my third prompt, I intentionally increased the level of specificity, directing the AI to represent itself as a network using symbols and to break down the stages of image generation. My goal was to push beyond the earlier simplifications and gain clarity about the role of “noise.” The resulting image was significantly more detailed and aligned with my expectations. The AI was no longer depicted as a robot, and instead, it appeared as a network of data connections linking the human user, the device, and the final output. The image also referenced processes such as latent space, diffusion, and language understanding, suggesting the processes involved in AI image generation, and that more precise prompting resulted in a more detailed image.

Overall, the resulting images across the three prompts were relatively accurate. The simpler my prompt, the simpler the image. However, it was interesting to see that the AI initially depicted itself as a robot. I was careful not to shape its self-representation in my early prompts because I wanted to observe whether it would present itself as a network, symbols, a robot, or a digital device. It was interesting that it first defaulted to a robot behind the computer screen. This suggests that within its programming, training, and human interactions, the AI absorbs and reflects cultural understandings of what AI is: robotic. So, when it’s given prompts that are not specific, it relies on this stereotyping and even identifies itself in a robotic form, rather than as a complex system. Additionally, referring to its own processes as “noise” demonstrates that the model is inherently abstract and designed to provide surface-level information, rather than complex and detailed explanations from the start. Overall, AI is heavily shaped by its user, and its outputs ultimately reflect the clarity, specificity, and assumptions embedded in the prompts it receives. Although these were not exactly the images I had in mind when I first prompted it, I predicted the AI would produce something closer to the third image from the start. This process helped me realize how much specific guidance the AI needs to produce what I am picturing, reinforcing the idea that meaningful and precise outputs depend heavily on intentional, detailed prompting.