in Projects

Hill Climbing

Software bot, 2021

Hill Climbing is a bot walking through Google StreetView that always goes up, following the direction of the steepest incline. As a software bot, the work performs autonomously, creating an open-ended visual meditation on human and machine agency, artificial intelligence and optimization, and the landscapes and operational images of the measured globe.

The bot implements the “hill climbing” optimization technique (more precisely: steepest ascent hill climbing with random restarts), that is widely used in machine learning and artificial intelligence. It constitutes the process of incrementally finding better solutions to a given search problem by following the ascent of a search space. The work illustrates the fact that all contemporary machine learning techniques can be understood as methods of mathematical optimization. It visually connects the idea of AI research as a heroic quest of finding artificial general intelligence (AGI) to the lone wanderer of romanticism confronted with the sublime beauty of nature. It tries to demystify artificial intelligence and machine agency by illustrating how they are grounded in the mundane reality of mathematically navigating error landscapes. In taking these landscapes literally, the bot appropriates the space of Google’s corporate image of the world, including its distinct aesthetics, absurdities and limitations. This spatial representation is connected to publicly available elevation data stemming from a scan of the Earth’s surface conducted during a Space Shuttle mission.