Methods of Learning

Aug 16th, 2016

The field of artificial intelligence (AI), computer systems that can learn and act based on their findings, is gaining in popularity lately. The upsurge in interest is due to improvements to the usage of neural networks, statistical models created after the image of neurons in biological brains. A neural network consists of artificial neurons, little processing units that take on a specialisation within the network. Together these artificial neurons can accomplish difficult tasks, like recognising faces in images or predicting stock markets.

These days neural networks are used for a variety of tasks, but mainly to assist or interpret us. Amazon, for example, uses neural networks to predict what products we like based on our previous searches and purchases. This allows users to find new products that are to their interest, and it results in major profits for Amazon. Neural networks play a subservient role within the technology that we use every day.

A more fruitful relationship between humans and AI, whether synthesised through neural networks or a different technique, could emerge through collaboration. Whereas humans might solve a problem in one particular way, an AI might solve the same problem in a totally different, although not necessarily wrong way. Through collaboration both parties might get a glimpse of each others reasoning and come up with the best solution to a particular problem.

Methods of Learning illustrates this collaboration. The lasers on the left side of a canvas are oscillating according to a sine function. When humans learn and apply knowledge, they often use tools to do so. The sine function is an example of one of those tools and can be traced back to trigonometric functions used by astronomers in India more than fifteen-hundred years ago. These functions allowed the astronomers to get a better grip on the workings of our universe.

The lasers on the right side of the canvas are controlled by a neural network that is figuring out how cosinus works, which it does by looking over cosinus data over and over again. Every laser on the right side is a neuron in this neural network, giving insight into what the network is actually doing. The canvas on the left shows a neural network in its infant stages, the canvas in the middle shows the same neural network in a later stage, the canvas on the right shows the same neural network in its final stages.

When tracing the intersection point of both types of intelligence, a circle slowly starts to take shape.

Methods of Learning was on display at GOGBOT Blink youngblood Award, best of graduates 2016, where it received an honourable mention. The work was also on display at InScience festival 2016.