Generative AI, Ethics & Its Impact on the Environment
I recently saw an article about artificial intelligence that expounded on what is wrong with the technology. I was teaching a pre-college “Creative AI & Design” course for high school students and we talked about issues, the benefits and harms of generative AI, which is a subset of machine learning and deep learning or what some refer to as AI 2.0. One of the main issues author Kim Crawley has is climate change — the ongoing increase in global average temperature — that has an increasingly large impact on the environment. Crawley quoted climate change activist Greta Thunberg who addressed the United Nations in 2018:
Our civilization is being sacrificed for the opportunity of a very small number of people to continue making enormous amounts of money. Our biosphere is being sacrificed so that rich people in countries like mine can live in luxury. It is the sufferings of the many which pay for the luxuries of the few.
There is no argument from me that climate change is a ongoing concern (and it should be for everyone). However, I do want to counter the notion of generative AI systems, which currently represent only a small fraction of data center workloads, being the whole problem. Recently, Grantable published “What is the environmental impact of AI?”, which explores the energy consumption and environmental impact of AI and data centers, massive warehouses containing the actual hardware that runs our modern digital infrastructure. The article references a recent study that notes that AI systems emit between 130 and 1500 times less CO2e per page of text generated compared to human writers, while AI imaging systems emit between 310 and 2900 times less CO2e per image than humans.
In other words: carbon emissions of writing and making images are lower for AI than for humans.
So what does this mean?
It means that humans may be emitting more carbon dioxide (CO₂) and other equivalent greenhouse gases than writers using ChatGPT or artists using Dall-E. Cars are responsible for a significant portion of global carbon dioxide emissions… and I have never owned a car. However, I have generated at least one AI-based image per day, for eight years. I think it’s important for people to be informed about how much CO₂ they emit when using generative AI and doing other tasks such as driving cars. Crawley also negated the practice of prompt engineering, which refers to crafting effective inputs, or prompts, to guide AI models, especially large language models (LLMs), in generating desired outputs.
Artists go beyond “being able to write instructions” when they enter prompts in an AI generator. In the example above, I started with a pencil sketch of one of my students and I used ChatGPT to enhance the image. I’m reminded of an online encounter with a man who was doing research for Adobe. He was commenting on a post by Wagner James Au about some of my Deep Dream Generator images.
What I really like about your results is that they don’t just look like the usual automatic output, they’re better and they show the artist’s hand in making them fully realized, even though the elements of the algorithms are still visible.
Was it possible for me to use generative AI and play the role of a (human) collaborator in the process of generating an image? According to the researcher, it was possible. Also, was this process anything like engineering? Well, how about we look at the engineering design process. The example below comes from Youth Engineering Solutions or YES.
The first step in the process is to ASK: define a problem, identify the requirements (criteria) and limitations (constraints). This is what happens when someone enters a prompt into an AI generator. Next, the AI model imagines the different ways to answer the query. Users share and select their best ideas to generate one image (design).
In the example above, I can select from four thumbnail images, chosen the best one for my project. I can improve the image by going through many iterations. I can go further by creating subtle or strong variations of my chosen image to explore different creative directions. Additionally, the “Vary Region” feature (in Midjourney) allows for targeted edits to specific parts of the image while leaving the rest untouched.
This process really is like engineering… prompt engineering.
Close to where my Faith Ringgold mural was installed in Brooklyn, NY sits a large yellow shipping container with black triangles painted on its side that represent the flying geese quilt pattern, which may have served as a coded message for enslaved people escaping to freedom along the Underground Railroad. This is the work of transmedia artist Stephanie Dinkins and it includes screen displays of AI-generated images. “If We Don’t, Who Will?” counters AI bias and addresses some ethics issues.
QR codes stationed around the public art project lead to an app where people are invited to submit their own personal stories or to answer prompts such as “what privileges do you have in society?” People around the world can also answer questions through the app. A ramp leads to the inside of the container, where after a few minutes, a large screen displays a generated image that reflects the information that patrons submitted in the app. Images that appear on a loop until another response is uploaded are mostly portraits of people of color, even if the person who submitted it is not one themselves. — Melissa Hellman
More important, there is a sense of urgency for underrepresented and historically marginalized people to make sense of the AI landscape and not let it continue to perpetuate bias. Climate change is urgent but remember generative AI is just a fraction of a bigger problem. In “Creative AI and Design” young people learned that they could think differently about these issues. Their ideas could change worlds.