When artist and developer Josie Williams began looking into A.I. chatbots some two years ago, she was immediately taken with the technology, but also with the possibilities of what more she could do with it. These natural-language applications, after all, suffer from a clear set of limitations. They are told to make relational connections and stick to syntax in their outputs, and most damningly, have been trained on datasets that often lean European, leading to the technology’s notable racial bias.
“What would happen,” Williams wondered, “if I used the words of radical Black thought leaders in an A.I. dataset, so that was the only thing that a chatbot could use to formulate responses about itself or the world?”
Thus was born Ancestral Archives, Williams’s latest project encompassing four A.I. chatbots that will be unveiled at an installation at SXSW in Austin, Texas, on March 10.
Each of these chatbots has been built on a dataset exclusively containing the work of a Black author, inviting viewers, in effect, to interact with these subjects in an A.I.-mediated conversations. The four thinkers—James Baldwin, Audre Lorde, Octavia E. Butler, and Zora Neale Hurston—are the “ancestors” referenced in the work’s title, who, Williams told Artnet News, “really allowed me to have a sense of identity, and feel seen and heard as a Black femme queer person.”
To further maximize the technology’s potential, Williams lifted constraints when it came to syntax in the chatbots’ responses, allowing their output to be “more abstract and fluid,” she said. “I was really curious how an A.I. chatbot would say things, given the dataset, when having free bounds for how it can deliver that information.”
The results, Williams admitted, were often “nonsensical,” but also, hauntingly poetic. She brings up an interaction with the James Baldwin bot that, when asked if it was dreaming, returned with: “A dream so infantile / That is practically forbidden, forbidden, forbidden.”
“I call them virtual poets,” said Williams, “because they produce really beautiful couplets that only have actual meaning in the terms of [the user’s] contextual understanding.”
The work’s physical installation will feature four traditional Nigerian masks to house each chatbot, all of them designed by Williams in a nod to her West African heritage. These masks, she explained, played key roles in traditional ceremonies, enabling wearers to channel their ancestors and their ancestors, in spirit, to be part of these rituals.
Ancestral Archives is embedded with that same purpose, staging meetings between tradition and technology, forebears and heirs. “I’m by no way trying to reanimate them,” Williams said of her ancestral subjects, “but rather [highlight] that the energy and meaning behind their words will always be here to inspire and lead us.”
The project is supported by New Inc, the New Museum’s innovation incubator, in partnership with Ernst and Young, the multinational consultancy which recently initiated its EY Metaverse Lab aimed at building inclusivity and equity into the virtual space. In addition to Williams’s piece, New Inc and EY are backing the development of a host of other tech-assisted art with an eye on an inclusive metaverse.
“Where Black speculative futures fit into how A.I. is measured and learned is really interesting for a practitioner who has always felt like her story has been on the outskirts of technology’s development,” Salome Asega, the director of New Inc, told Artnet News. “It’s our place as a program that nurtures cutting edge and emerging art practices to support the research and the questions artists are asking.”
Indeed, the question posed by Ancestral Archives is less about the creative potential of A.I. than how Black experiences and identities can be centered in the build-out of A.I.—making them the core, not an afterthought.
Williams further intends to build this same inclusivity into the delivery of Ancestral Archives. Following its physical launch, she hopes to host the project online to ensure it an expansive reach beyond the niche corners of technology or academia, where it might be “hard for people from under-resourced demographics to have access to it.”
“It’s this idea of building for people who are most marginalized rather than building for people who are already benefiting from the system,” she emphasized, “starting from the outside in, rather than the inside out.”