How to Confuse Machine Learning

This is a post that was shared in LinkedIn and thought of exploring this with my pet tool Orange Data Mining tool.

First, separate these images into single images randomly. Since there is a small number of images, thought of employing clustering over classification. 

Following is the very simple Orange flow and it can be found at GitHub as well. 

As you could see, it is a very simple Orange workflow. Inception v3 was used as the embedded and Cosine distance was used as the distance calculation. Then Hierarchical Clustering was used. it is as very clear clusters which can be seen in the following figure.

Then the cluster distribution as below. 

So, you can't confuse Machine Learning as you think.