Making A Mind

 

 

 

 

 

Making A Mind

Prototype application of Making A Mind, an interactive visualization of the process of decision tree classification, a type of machine learning, and the data flowing through the algorithm. The first prototype uses the Iris data set.

The second example uses synthetic data, to focus on process and structure, and to visualize mistakes in the classification.

 

 

The third example uses data from bank loan assignations.

The motivation for this research stems from the obfuscation problem in machine learning algorithms. Whilst such algorithms are statistically the most accurate forms of prediction available, they still produce errors, and there is no theoretical framework to understand how they work and why they arrive at the decisions they do. This research works toward a design research methodology that intersects critical code studies – the close reading of source code for social and political interpretation – with computational design, in order to explore this problem through a critical visual framework. The questions being addressed are: whilst it has been established that biases and discrimination exist within datasets, what can we learn by temporarily isolating data and studying the structure and process of an algorithm? Can a computational visualization support our understanding of algorithmic process, in the same way that data visualization supported the understanding of complex statistical issues? Specifically which visual design tactics could support an ethical debugging of algorithms? Decision tree classifiers were chosen to begin this research because they are more simple systems, can be reverse engineered, and are already graphic in nature, however the intention is to progress to work on more complex neural networks.

 

7

6

9

5

8

10

1