| home page | visualizing debt | critique by design | final project I | final project II | final project III |
Hey that’s mine! Art and ownership in the age of artificial intelligence
For Part III, I tried to clarify my story. Some of the details get more technical than I anticipated, and as I learned more about AI and art, the more layered this story became. I reordered a few charts in the introduction and added a case study about an artist to flesh out my intended story arc from Part II.
In general, I spent a lot of time focusing on making dense information clear and engaging the audience so they were willing to dive into technical information they may not be familiar with.
I ended up completed getting rid of the first data visualization. It was repetitive with the second visualization, and the December spike was just confusing for audiences. Tying overall Google trends of all the major AI art generators made the same point, and was more engaging after a color palette makeover (see below).
I changed my bar chart which showed how many AI generated artworks were accurately identified by those surveyed into an infographic. I think it gets the point across much more quickly, although I am not sold on the title. I went back to two of my user research participants ot workshop it and ultimatly landed on “4 out of 5 AI-generated images were misidentified as artist made”, but I still think it’s a bit confusing.
I added one more frame to my neural network graphic to understoce how many images datasets need to be trained on to work well. This helped narratively connect this section to the next (well Stable Diffusion has been trained on BILLIONS of images) and also helped to highlight that this training cannot be undone, and the training is vast.
I didn’t ask for user feedback on my pie chart, but after creating the tree map I realized that the pie chart would make more sense as a segmented bar. The tree map, which is a square, represents a subset of the bar. I thought that visualizing the two as rectangles would help make that connction visually.
I updated the tree map to include what percentage each domain made up of the total dataset. I hope that this helps those who are not unfamiliar with this type of data visualization gain confidence - why yes, the bigger rectangles do mean more images! I also integrated the legend into the map itself to clarify the types of domains for viewers. Of all the visualizations, this one took the longest.
In Part II, I identified three personas: artists, software developers, and business owners. After I recieved feedback from my peers that they wanted the data story to relate back to the experience of artists, I began reading more about their perspective. For artists, addressing the challenge of AI art generators can seem quite impossible - neural networks cannot even be untrained on their work! I made the calls to action for this group more concrete and actionable (although I still wish there were more resources I could find to offer to them.) What I really learned by empathizing with this persona was that a lot can be done for them if the technology is used more responsibly.
Because of this insight, I decided to consolidate the other personas as just “user” which would encompass business users as well as my class peers who could be considered primary demographic users of this technology as CMU masters students. I turned my attention to helping this audience understand how misusing this technology harms artists and ways they could adapt their behaviour to use the technology more ethically. I included calls to action that would inform the average user generating images for personal use, as well as business owerns hoping to generate images for commercial use. To help users sympathize with artists, I also included a case study with a quote from an artist that I hope will make the impact of this technology feel more personal and tangible.
Incorporating my mood board helped to make this data story feel more cohesive as different threads are introduced - business and economic impact, how the technology works, the legal side - and helped to create a common langauge across data visualizations. I incorporated TA feedback from Part II to choose a color for each company. Using all cool colors for this gave space for a neon orange to highlight data points.
I initially intended for the AI art to emotionally manipulate the audience - look at this cool AI art! Too bad it’s existence has some major ethical implications… Instead, too much abstract AI art made the data story feel less human (go figure). I went back and tried to incorporate more human imagery. Even AI genearted portraits felt more engaging than the abstract art, although with another round of user feedback, I would probably test this theory.
It may be worth noting the time stamp of this data visualization for future reference: March 2, 2023. I introduce my data story by noting that AI art generators have only recently become popular, as large and impressive generators are more accessible for public consumption. These generators have existed long enough that in just the last week, the US Copyright Office has begun releasing statements regarding whether AI generated art can actually be copywritten. I added a note to my story that the legal gray area may be more concrete in the coming years. Just the day before I presented my visualization, I also found a lawsuit that a few artists filed against StabilityAI (noted in my calls to action). As this portfolio may stand as a snapshot in time, I look forward to looking back at the challenges addressed in my story to see how society responded to these moral and ethical dilemas surround AI and art ownership.