Artist - Rui Hu
Simulation Scripting - Rui Hu
Deep Learning and Data Analysis - Susan Yingshu Chen
In 2007, the Shanghai Stock Exchange Index climbed to an all-time high. The next year, months before the celebrated Beijing Olympic, natural and economic disasters hit, among them the most severe stock market meltdown, seeing the index down from well over six thousand points to just a fifth of it. Globally, financial market crisis also made history, with its root in the housing mortgage system in the United States.
Also in 2007, Nvidia launched CUDA, enabling more efficient and accessible general purpose computing on the GPU. Traditionally used to process images rooted in military and entertainment, GPUs have since made its way to fields such as AI, scientific computing, and finance, with applications like weather and stock market prediction, due to such high efficiency in calculating matrix math.
"Matrix Model and Uberbau" started from thinking about this history. With the promise of deep learning, would we have been able to predict and prevent past financial disasters? Or perhaps, as Keynes has suggested a long time ago, the stock market was more of a blind guessing game of other players' likely behavior rather than a precise math. Forecast might work only as long as nothing unexpected emerges from human's collective behavior.
Though driven by AI, the work is not "generated." Instead, the image and sound refers to moments in history, while the game-engine-based simulation features various scripted events driven by "prediction" results of the 2007-08 stock market using long short-term memory neural network on the first-gen GPU with CUDA support. Through the work, I try to explore how matrix math has become the base of various aspects in contemporary society, the idea of "model" as both in graphics and in science and economics, the group psychology involved in stock trading, human's basic needs and luxurious desires, financial capitalism, state capitalism, and how the perfect model might only be the one that embraces its fundamental impossibility to be correct.
The work features a space appears to be both a sports stadium and a construction site. Various scripted events compose the real-time simulated image, which are driven by deep learning prediction results of the 2007-08 stock market data trained on a GeForce 8800 GTX, one of the first consumer GPUs with CUDA support. Images show men and women in suites running on the field tackling off-scale office items, only to be hit and knocked over again and again, all monitored by hovering searchlight-equipped GPU-drones. Around the space is an imprisoning set of scaffolds and bar-chart-like buildings.
As the scene being observed through automatically switched camera views, we also heard various kinds of audio based on the viewing location. There is a Hong Kong stand-up comedian telling gags about stock trading and housing market, and various readings of academic text about successfully predicting past financial crisis or linking weather with stock market performance.
Many other elements serve as metaphors or one-line jokes: videos of a young David Bowie performing on stage, occasional appearance of an oversized pig person dressed in cardboards, a construction crane, advertisement images, bubbles, butterflies, sickles, torches, fogs, and a pair of giant galactic bull and bear, crouching over the entire scene, engaging in an eternal fight, against a night sky with occasional sunrise-turned sunset and a never-ending thunderstorm.
Screen recording excerpt: