Scientists have traditionally treated the internal mathematical layers of artificial intelligence \u2014 often described as AI\u2019s \u201cblack boxes\u201d \u2014 as too abstract and complex for humans to fully interpret.<\/p>\n
But a new collaborative project between Linguistics Professor Ga\u0161per Begu\u0161; his wife, Nina Begu\u0161,an Artificial Humanities researcher at UC Berkeley\u2019s Center for Science, Technology, Medicine & Society; and artist collective Metahavenis challenging the idea that AI\u2019s inner workings are beyond human understanding.\u00a0<\/p>\n
Mapping the “black box” \u201cLatent spaces are the hidden layers doing all the work that nobody really understands well,\u201d Ga\u0161per explained. \u201cThey learn as they look at data \u2014 just like humans start with a brain that is not yet trained to speak or move and gradually from connections until we become adult human beings with complex thought, language and self-awareness.\u201d\u00a0<\/p>\n To investigate these hidden layers, the researchers studied speech-generating generative adversarial networks (GANs). GANs are two-part AI systems where one part generates data and another evaluates and refines it until the output appears realistic. By examining the latent spaces within these models, the researchers found that unlike the large language models (LLMs) behind most modern chatbots, GANs develop language more like human infants: by listening, babbling and gradually turning noise into meaningful speech. They refer to this process as \u201cinformative imitation.\u201d<\/p>\n Rethinking humans and machines \u201cThe relationship between machines and humans was never meant to be a relationship between two equivalents, but evidence suggests machines might eventually be on par to humans,\u201d he said. \u201cWe have to navigate that and redefine what it means to be human \u2014 perhaps learning to value biological intelligence for its own sake.\u201d<\/p>\n He added: \u201cFor the first time, something non-human can learn human language perfectly. It used to be that you had to be human to learn language; now, that\u2019s no longer true. The notion of humanity itself is changing.\u201d\u00a0<\/p>\n Using the humanities to understand AI \u201cThis is a crucial moment where the humanities can help us tremendously in the design of AI,\u201d he said. \u201cUnderstanding our similarities and differences is important for building models that are safe and for anticipating what will emerge as models get bigger.\u201d<\/p>\n Nina added that \u201cunderstanding how these models work leads to more informed decisions about AI and human futures.\u201d\u00a0<\/p>\n To make these hidden layers more understandable to the human eye and ear, the researchers turned to the novel \u201cFinnegan\u2019s Wake\u201d by James Joyce. The book is famous for its “stream of consciousness” style and invented words, which the researchers say mirrors the “pre-linguistic” space in our own minds where thoughts exist before they are turned into clear sentences.<\/p>\n Nina, whose work focuses on the emerging field of \u201cArtificial Humanities\u201d<\/a> (as described in her book), argues that literary tools are just as important as code when it comes to understanding these systems.<\/p>\n \u201cLiterary and humanistic methods such as interpretation, comparison and metaphorical or spatial thinking are legitimate tools for navigating interiors of machine learning models and can help us assess and reshape them from without and from within,\u201d she said.<\/p>\n The team extended this interdisciplinary approach by collaborating with Metahaven<\/a>, who engaged closely with Ga\u0161per and Nina’s work and, with\u00a0media artist and coder Riccardo Petrini<\/a>, developed visual and spatial interpretations of how these AI systems work internally. Together, they created immersive interactive artworks that allow people to \u201ctravel\u201d through AI systems.<\/p>\n The process began by translating Ga\u0161per\u2019s first GAN model \u2014\u00a0fiwGAN<\/a> (Featural InfoWaveGAN) \u2014 into an interactive visual landscape, said Metahaven artist Daniel van der Velden. The model appears as a navigable, abstract environment filled with clusters, fragments and blurred image forms. As users move through the model\u2019s \u201clatent space,\u201d these elements emerge and dissolve, representing the concatenations (or sequences of operations) the GAN performs as it imitates and generates language, van der Velden said.<\/p>\n
Drawing on neuroscience,\u00a0linguistics, philosophy, the humanities, art and literature, their Latent Spacecraft<\/a> project shows that the internal \u201clatent spaces\u201d of AI models are not just abstract math, but actual spaces that can be navigated and mapped, much like the human brain. \u201cLatent spaces\u201d refer to the hidden layers inside AI models where learning actually happens.<\/p>\n
These findings\u00a0have broader implications for how we understand the relationship between humans and machines, Ga\u0161per said.<\/p>\n
For\u00a0Ga\u0161per, this moment calls for deeper collaboration across disciplines, particularly in designing safe and responsible AI systems.<\/p>\n