Amanda Lagerkvist: “Existential AI: Putting Simulation to the “Existential Test””

ABSTRACT:

This lecture sets out from the fact that, existentially speaking, a human being is always more than what can be known, quantified, pinned down or simulated. The fact that humans – creatures of the limit situation – are irreducible to statistical averages, prompts putting our AI-driven lifeworld to an existential test, by asking: What existential needs can and cannot be met by simulation? And how do digital simulations stand in relation to the non-representability of liminality and existence? To discuss these questions, I will be delving into the contemporary surge of simulating humans in generative AI systems, through three examples. First, “This person does not exist” (a generative adversarial network called StyleGAN2) is a machine learning framework producing synthetic images of persons of extraordinary fidelity, offering an eerie and deep relationality with technology. The second case is chatbots invoking pan-relationality or what I call relational expectancy in users. Finally, I will discuss the case of biometrics that nails body data to “truths” about a person – cutting open an existential chasm between the existential and dis/abled body and the biometric body. Without spoiling how the models fared, I can reveal that the aim is finally to try to offer a discussion that reconceives of them as existential AI and sets existentially sound limits on simulation, at a point in time when reality seems to be bending itself to a flurry of deep fake distortions, in a vast and violent manner.