Back in 1968, Philip K Dick’s science fiction novel Do Androids Dream of Electric Sheep? asked whether intelligent robots could have genuine emotions. Technology is not quite there yet, but the collaborations at Google’s Artists and Machine Intelligence (AMI) program are making potentially revolutionary discoveries about the links between human and machine minds. AMI resident artist Refik Anadol has turned these tools to the archives at SALT Galata in his installation Archive Dreaming, which Dick might have titled, “Do Algorithms Dream of Alternative Histories?”
The SALT Galata archives include around 1.7 million documents, from fine art and portrait photographs to financial documents and architectural plans, ranging from the late-Ottoman era to the present day. This vast collection of data was inputted into the artificial intelligence (AI), which analyzed the visual qualities of each item and then assessed the relationships of similarity and difference between all 1.7 million items. The algorithm, called t-distributed stochastic neighbor embedding (t-SNE), then arranged the data according to these relationships, which can be visualized in either 2-D (a scrollable spectrum) or 3-D (a cloud-like cluster) forms.
Anadol’s installation at SALT Galata is in a circular room with a mirrored ceiling and floor, producing a recursive effect of infinite space both above and below. This represents the metaphorical channel of data flowing from the archives to the touchscreen system in the center of the room. With this touchscreen, visitors can travel through the archive as a 2-D or 3-D space. You can also zoom in to a single item, zoom out to the entire archive as a ball of dots and strings, or view the data at any scale in between to group the items by various degrees of similarity. Being able to see the entire archive at a glance, and the connections between any one document and all the others in the archive, is a major advantage of this system.
When not engaged by a human, the AI spontaneously “dreams” with the archive data that it has processed and possibly forms new connections and groupings that might never occur to a human mind.
Software expert Blaise Agüera y Arcas has noted that as AI is capable of recognizing classes of objects or faces by their common features, so AI can turn those classes into output, effectively “hallucinating” those objects onto neutral space. This is the principle behind Google’s Deep Dream software, which hallucinates faces and animals onto submitted images to create surreal results. Arcas has questioned the effect this artificial ability will have on human art in future—while the camera created a revolution by replicating the human eye, this software is beginning to replicate the functions of the human brain.
This hints at exciting possibilities in the context of the SALT Galata archives. In a talk with Vasıf Kortun, artist Anadol explained how we might use artificial intelligence to create alternative histories: “The method we use here is creating a game of opposites between two networks: a differentiator that tries to learn how to differentiate real documents from fake documents, and a producer that tries to create fake documents for the differentiator to recognize. Once we’ve trained the producer, we can use the same statistical rules to create new images and create documents that look real but actually never existed … like documents that might come from another version of history.”