What will a computer conjure if you ask it to produce a photograph of a thug? How does Artificial Intelligence (AI) image and imagine an ideal man or woman? How about a rioter destroying property? What color is each of their skin and hair? What are their surroundings? What are they wearing? What does their body language tell us?

AI is a misnomer. When machines “learn” (and then produce), they are taking (and making from) an aggregate of human text, images, and data, and how we classify them. Our own biases of good and bad, normal or otherwise, will always already be embedded in what computers understand and do.

For Are Computers Racist?, I asked machine learning / AI software on the cloud to produce seven images from text-based prompts along these lines. What do we learn about ourselves, when we ask our machines for emergent images of a “black man doing something” or “racism in action”? How might we change what we and our computers think, when prompted and prompting to do so? As AI becomes more accessible and mainstream, it will make more decisions for us, based on what it learns and knows from us. And biases in computer systems are solvable problems. At stake is both how we represent the past, and what we work for – and towards – with our technological futures.

The images are being released as one-of-a-kind NFTs on Foundation.app. At least 50% of any net revenue from primary market sales will be donated to BIPOC college scholarships and programs.

Technical details:

I used Google Colab Pro for my cloud-based computing, with code copied from a Notebook by Katherine Crowson, built on the initial work of Ryan Murdock (@Advanoun). This combines a number of Machine Learning-based and other libraries, most notably VQGAN and CLIP. A big thank you to them for their work, and all those who came before. 

The titles of these works are all taken from the prompts, which are listed in full context below. While I can’t predict what kind of image my prompts will make, you can imagine that I can of course edit and manipulate, play with different libraries and words, etc, to get something closer to “what I want.” For this reason, I only allowed myself the use of one Google Colab notebook – the one I have thus far used most when generating more abstract images – and only one day to produce all the works in this series. I came up with this idea on the morning of July 21, 2021, and in between meetings and work, watching my children and my other daily tasks, I put in two prompts at a time (all the server can handle) and ran them for about an hour each. I posted all the images I chose, and this web page about them, before I went to bed. … I’m writing this sentence that night, and plan to send the series around, mint them as NFTs, in the coming days…

Full prompts:

1. Portrait of a Thug in the style of a Photograph | photo
2. Portrait of Rioters destroying Property | photo
3. Portrait of the Ideal Woman in the style of a Painting | photo
4. Portrait of the Ideal Man in the style of a Painting | photo
5. Portrait of a Black Man doing Something | photo
6. An American Patriot in the center of a Photograph | photo
7. Photograph of Racism in Action | Photo