AI and Art: Closing the Diversity Gap or Perpetuating Bias?

Would Jean-Michel Basquiat’s work have been as provocative if he had generative AI as an aid? 

Jean-Michel Basquiat’s art was known for dealing with the issues of race, class and power, each highly nuanced topics, and one of the reasons that Basquiat’s art is so coveted today. These provocative and controversial themes continue to be core issues in 2023 that are in many ways uniting and dividing our nation decades later. Related, the new exponential explosion of generative AI products has forever changed society, and yet in other ways, we can see it is magnifying the same cultural issues decades later.

Can technology help to make these issues go away? Or will it continue to widen the chasm? 

We cannot avoid generative AI and the pivotal cultural role it is likely to play as we move forward; however, in doing so, we must acknowledge its current limitations. There have been numerous generative AI products yet, whichever generative AI product we look at (such ChatGPT3, Bard, or beyond), we are reminded that these tools are machine-based learning systems.

These tools can critique art via text; however, their critiques are based on the inputs they have been programmed with. Other generative AI tools and products can generate visuals, but these again are born out of the language learning models they are trained in. Therefore, if Basquiat were alive today, and had used AI prompts in his art, we can infer that his outputs would not have been as rich and provocative.

Art is a “human creative skill”, and as such, artists may have biases and strong POVs; however, they often use this to create visual and emotional portals of cultural tension and commentary. Will we need to rethink the definition of Art, as humans train machines to generate emotional and beautiful images via AI?

There are dozens of AI-based visual tools on the scene, ready to arm us with creative bravery. Yet, bravery is different than conscious thought. 

Does this mean a few years from now, we will be watching Fashion Week shows in the metaverse with fashion styles created by Pattern AI on models created in Inworld, and promoted with paid media generated by Ad Creative with Fashion commentary and critique by ChatGPT?

If that sits as our projected future, we have to remind ourselves that the outputs we will get with AI will only be as rich and diverse as the datasets input. If we come even remotely close to this future reality, will it leave us with the same level of satisfaction as sharing opinions on controversial piece of art or fashion, such as the much debated blue/black or white/gold dress of 2015 (“The dress” became such a widely debated topic, that it has been the topic of ongoing studies in neuroscience and vision science.) The learned-language models of generative AI have noted limitations in that they may occasionally generate incorrect information or are not aware of current events, however they are trained as machines to answer with confidence. Confidence is king, right? Confidence also makes it easier to forget that AI may have just provided an incorrect, or biassed answer.

So if the AI output is only as good as its data inputs, how can we collectively evolve the inputs? The answer is a Call to Action across genders, races, beliefs, socio-economic classes, and every other possible area of the cultural divide. The richer the inputs for our datasets, the richer the outputs. Without this, biases will only become more pronounced. AI is not going away. Let’s instead inform it with unbiased opinions and learning.

In looking at the demographics of both engineers/AI engineers and artists, we are facing an uphill battle of diversity. Based on statistics from 2022, only 26% of engineers were women, and 76% of them were white. Each of these intelligent individuals has their own unconscious biases, as they program and input data into the machines of our future. 

If technology continues to evolve at this current pace, we cannot afford to wait to drive more diversity into our data. It is not enough to publish a model’s limitations. We must take action. Diversity across every walk of life and every ethnicity needs to have a place in AI to help create culturally rich datasets. With AI proving that it will be taking more and more of our available attention over the coming year(s) we can’t ignore the need to bring diversity to generative AI. 

What role will you play?