In Medio Stat Virtus

In Medio Stat Virtus

Wintersemester 2017/2018

 

machine learning, AI, balancing, kinetic installation, Things That Learn

 

We want to teach machines how we and our environment work. Tons of real-time data in machine-readable form are pushed into systems to classify and predict the real world. The kinetic installation 'In Medio Stat Virtus' shows a system that learns to find the balance. The efforts of the machine to find the balance make the success and failure of the system physically tangible. It has to deal with data from the real world and its physical environment. Analysing the environment with almost no loss of information is essential for a perfect execution of the task. The artificial intelligence tries to balance between extremes, something people strive to get closer to a contentment in life.

The system attempts to balance a hanging, transparent cylinder. It contains 21 metallic spheres whose surface reflects the environment. The rolling metal balls represent individual units belonging to a whole. The weight of each individual ball influences the condition of the whole structure. Similar to a puppet, the container is moved up and down continuously by motors on two strings. The machine knows nothing about its task. It learns by experimenting. The system receives positive and negative experience by being rewarded accordingly. So it tries to approach a balanced state. The mission can only be accomplished by the artificial intelligence of the machine if the input data is reliable and the system is rewarded appropriately. Then the balls would stand in the container in a perfect line. Machine learning soon permeates almost all areas of life. Keeping things in balance has always been considered by most philosophers to be the best way to master life. The installation translates the natural concept of keeping balance to artificial intelligence. In a world where decisions are made from machines and we are numbers in large datasets, it can be important to translate our ethical concepts into intelligent machines.

The project was developed within the course "Things That Learn" from Luiz Zanotello.