Making Waves in AI
Reservoirs of Venice
Reservoirs of Venice is an art installation that uses water columns instead of transistors to predict the time of day based on human activity. This approach challenges the usual way artificial neural networks are built, offering a unique physical interpretation.
Typically, artificial neural networks run as mathematical models on digital computers. While digital computers rely on transistors, alternative computing elements like vacuum tubes or mechanical gears have also been used historically. Similarly, neural networks can be physically constructed using structures designed to interact like layers in a model, transforming inputs into predictions.
Though practical applications may be limited, the Reservoirs of Venice installation by Dietmar Offenhuber and Orkan Telhan offers a fresh way to educate, entertain, and inspire new thinking. It features columns of water that interact, using these interactions to predict the time of day based on observed human activity near Venice’s canals.
The input comes from webcams monitoring the canals. Data about human activity is extracted and converted into water disturbances generated by wave makers in four water reservoir columns. Each column’s disturbance influences the next, creating a chain of interactions across the reservoirs.
The disturbance from the final water column is then processed by a traditional single-layer digital neural network. This network interprets the physical data to determine the time of day, learning from numerous examples to improve its predictions.
Obviously, the water columns could be replaced by additional layers in a digital model, simplifying construction and speeding up computations. However, using water provides a tangible way to visualize processes inside a neural network. It serves as a reminder that current methods are just one option and encourages exploring different approaches for smarter, more energy-efficient AI solutions in the future.
Your membership also unlocks:
