By Amuna Wagner

"Any Black person using AI today can confidently attest that it doesn’t actually know them, that its conceptualisation of their reality is a fragmentary, perhaps even violent, picture," writes artist Linda Dounia Rebeiz in her curatorial note to In/Visible, an NFT exhibition exploring (the lack of) Blackness in Artificial Intelligence (AI). 

The showcase features new work by ten Black artists who enter conversations with AI and each other, to critique its shortcomings in understanding them. Their premise is that technology is not created in an apolitical, post-racial vacuum, but rather carries within its fabric the implicit biases of the past which, if left unchallenged, will render the Global Majority silenced in the 4th industrial revolution (4IR). By playing with the representation of Black hair, recreating the essence of spiritual drawings, or imagining a magical realist cityscape of Khartoum, the artists assert and critique Black visibility in and through AI. 

Investigating AI Bias 

Rebeiz is an artist, curator, and writer, born and raised in Senegal. Trained in design and digital programming, she worked as a designer for a decade before trying her hand at selling her art. “It was difficult for me to fit in between the physical and the technological world,” shares Rebeiz. “But NFTs helped me find a community.” Her art has travelled the world, showcasing at Unit London, Art Basel Miami, The Dakar Biennale, and beyond. “As an African woman, it was also difficult to identify with the tech industry and internet culture,” says Rebeiz, “It seemed like something for other people, like I had to change to fit into it.” Thus, her artistic practice, in which she incorporates a variety of technological tools like creative coding, scripting, and machine learning, focuses on bringing her context and experiences into light.

Rebeiz’ intro to AI started with a curiosity about censorship and representation in technology years before the current AI boom. “I watched a documentary and realised that there was a real problem with AI and surveillance,” she recounts. “A lot of Black people were misidentified as criminals, because AI couldn’t tell them apart.” Intrigued, she learned how to train custom generative adversarial neural network (GAN) models – where one network generates images and a second distinguishes the real from the fake – and made a number of crucial observations. “Tech in and of itself is not necessarily a political instrument,” explains Rebeiz, “It’s the question ‘which humans does tech need to exist?’ that makes it political. You need humans to curate the data sets, train the AI, and tweak the model, and these humans are political entities. When AI is racist and misogynistic, it’s because the data used to train it is biassed.”

The averaging effect 

Black artists have raised concerns about AI’s inability to accurately depict Black life and humanity, often distorting faces and bodies, lacking definition of features and the ability to understand cultural references. When Rebeiz gave OpenAI’s image generator, DALL-E 2, a prompt about a building in her hometown, Dakar, it produced a desert-like place with dilapidated buildings – a perfect stereotypical image of the western imagination of Africa. Even though 86.11% of the world's population owns a smartphone in 2023, most internet knowledge is created by North Americans and biassed towards English. Their perspective of this world distorts knowledge production and leaves the Global Majority at a serious risk of erasure. A recent Bloomberg study analysed more than 5,000 images generated by Stability AI and found that racial and gendered stereotyping was even “worse than reality”. 

Rebeiz explains this phenomenon as “the averaging effect of AI.” When a model is being trained, every additional image is a data input of which the algorithm averages out a common denominator. If it is fed 80% eurocentric data and 20% data from the rest of the world, its common denominator is fraud. “Imagine that you’re training AI with eight apples and two oranges. When you ask it to give you the texture of an orange, the answer will mostly be like an apple,” explains Rebeiz. 

She identifies the lack of transparency over what data is used to train most AI tools as one of the main issues and asserts that users should have the ability to contribute their own data in order to be taken into account. “AI mediates a deeply flawed and exclusionary understanding of the world. Against the vastness and richness of the world's artistic expressions, it's criminally limited,” she writes in a series of tweets that break down the deep-rooted historical origins of technological bias. Rebeiz goes on to say, “If we are vulnerable to the legacies of our history, it is not so far-fetched that what we create is also vulnerable to it.” 

Defiantly Visible

Attuning AI to better represent Black life and humanity is a group project in which every contributor’s data makes up a deeply necessary drop in the ocean. Rebeiz believes that Black liberation in the digital world is contingent on the ability to feel represented by technology and the internet, and the ability to create an own, decentralised space. “There’s solidarity in being excluded and calling out the same issues,” she asserts. “Multiple people coming together to train AI is not happening yet, but it’s definitely possible.” 

The exhibition compiles thought-provoking artistic endeavours which, Rebeiz hopes, will encourage Black artists to critically engage with new technological tools, to share their findings and tell their stories autonomously. “Artists make these issues apparent,” she explains. “It's easy to get lost in the technical terminology, but when you see and feel images, it sparks conversations.” 

To Rebeiz, non-participation is not an option, because these new technologies are not going anywhere. “If we don't hold AI accountable for what we want it to do, someone is going to make these decisions for us,” she says. “There’s something political about being a Black artist from the Global South and using it even though it doesn't know you. That’s all we’re allowed to do at this point.” Africa is an interesting case as the continent lags behind technologically, which leaves its citizens vulnerable. At the same time, a proactive approach could give way to possibilities of better futures in which Africans learn from the journeys and challenges of other countries and find ways to represent themselves authentically. Citing Arturo Escobar’s Pluriversal Politics, Rebeiz proposes a world in which multiple futures can coexist, both online and offline. “Let’s not give ourselves limited ideals of what we can be.”