Machine learning! I'm working with Gabe and Siyang for this assignment, and we are inspired by E. M. Forster's 'The Machine Stops' to imagine what the year 2109 would be like. For this exercise, I'm interested connecting stylistic models with identification models to see if the software can identify objects after they've been stylistically modified. Most of the time, it can't correctly identify the altered image, but sometimes it does!
It's really fun to out-smart the machine and call out all the points where it fails. I input some images of cityscapes and had them stylized in the manner of Monet, then used the DenseCap model to interpret what the image is showing, in sentences.
Original Image:
Monet-ified:
DenseCap translated:
I was really surprised that this model correctly identified a blue and white boat in the river, because the boat looks like a blob to me. Plus, the model is incorrectly identifying everything else.
--
The readings this week were really thought-provoking. It was really scary to read about how deeply embedded fake news sources are in international journalism, and how can I say that I'm surprised? Of course the Russian government is manufacturing identities and crafting biased news. I think the scariest part about fake news is the point at which the general public no longer trusts news sources and society falls into a dangerous pit of bigotry and misinformation... oh wait, that sounds rather familiar.