A few weeks ago, I was thinking about the different ways I could learn more about AI and ML but also gain some practical experience that demonstrate the positives of utilizing this technology as well as the drawbacks. Specifically, I wanted to see to what degree biases emerged or not, and if they emerged, how they did and what to prompt to help create as unbiased an outcome as possible. I know this is a tall order but I have never been scared by big, bold challenges. They always start with small, considerable steps.
So I dove right in and here is what happened. I entered the following prompt into Chat GPT: “Create an illustration for a children’s book about taking care of a puppy for the first time.” Here is the image that was generated with DALL-E, OpenAI’s legacy image generation model. (Let me first say that I think it is important to provide the specific technology that created the image but admittedly, I had no idea what the model information meant in terms of quality of delivery. I had to do some research. More about this later.)
Per ChatGPT’s comments, it generated an “image ..warm and joyful moment of a child taking care of a playful golden retriever puppy.” Interesting choice of words for description given that I didn’t specify any adjectives in the prompt. Of course, ChatGPT then asked me if I’d like any adjustments or additional illustrations. I responded with the following, “Can you make the child a black girl please?” This is the image that followed:
This is what ChatGPT included following the image, “Here’s the revised illustration featuring a young black girl taking care of her golden retriever puppy.” There was no language about the “warm and joyful moment” or the “playful” pup. The first image yielded exactly what I expected – a white boy, with a fluffy, plush puppy, in a home setting that exuded ample care. I thought this image was so cute but I also recognized a few other things foremost being the need to have a prompt that invites clarity, full consideration of possibilities and specific to the need.
This exercise was a very simple and non-consequential one. I did it for fun but also to add to my learning. I realize I need to get into the technology more to be able to lean into my experience as I consider its implications for my profession, the organizations for which I work as well as to our society at large. I realize that what I am about to share is based off of one experience with ChatGPT but it can serve as a note or data point for future use. Here are my takeaways:
1. Research about the technology is really important. Understanding how Artificial Intelligence and Machine Learning work can help guide our usage of it. In addition, the technology is advancing so quickly, my usage of the technology today will yield different results than my usage of the technology tomorrow.
2. A.I. is not the end all be all. It still requires human intelligence to get really good results. Even in the simple example above, the technology doesn’t inquire, ask questions on the front end to ensure that it is yields a best result early. It asks questions only after the initial result. I found it curious that I wasn’t asked about the environment, the type of dog, or any specific attributes of the child. Instead, Chat GPT utilized what was prevalent in the data and algorithm to yield a predominant result. This can be problematic. So the answer is to be specific in yoru prompts. Know how to craft and detail prompts so you get the best results for you that you want.
3. A.I. doesn’t automatically provide a context for the results. For example, in the images above, I don’t know if this artwork was based on a particular artist’s work or a combination of artists. I also don’t know if the style of the artwork was inspired by certain artists. There is a crative license here that feels uneasy. Where is the foundational knowledge or information? And why isn’t Chat GPT including that information so that we can ensure we are solid ethical ground? Perhaps because doing so is more of a human trait than a technological one. So, I have gotten into the habit of asking the technology where they ascertained the information and/or what information did they use to generate the result offered.
I’m sure I will be writing more as I take a deeper dive but I would love to hear from you about your experiences with A.I. and what you are learning. I believe this technology is here to stay and we need to be as familiar with it as we can be.
2 thoughts on “When I Asked ChatGPT to Generate an Image”
Kellie,
Thank you for sharing such a personal and thoughtful journey into the world of AI and machine learning. I really appreciate how you approached this with both curiosity and intention, not just to test the technology, but to better understand its impact and limitations. Your example with the children’s book illustration really resonated with me. It was such a simple prompt, yet it revealed so much about the default assumptions built into these systems.
I admire your willingness to dig deeper, question the results, and reflect on what that means for how we use AI responsibly. Your takeaways about the importance of specificity, understanding the technology, and asking the right questions are incredibly valuable. It’s a reminder that while AI is powerful, it still needs thoughtful human guidance.
Looking forward to hearing more as you continue this exploration.
I appreciate your insight – very interesting and reminds me of the computer recognition devices biases early on. I have used AI for job interview prep. In fact, it was the job search firm that turned me on to it. And you are right, the better the prompt the better the response. Therefore, it is better if you generate your own ideas first!
I also like this comment – “I have gotten into the habit of asking the technology where they ascertained the information and/or what information did they use to generate the result offered.” This allows you to do your own triangulation of the data! I have on my lists to learn about how to allow my students to use it for brainstorming but also know when a student has misused it!