As researchers continue to explore the applications of deep learning, we'll keep coming across tasks that humans find hard but AI finds easy.
A recent paper showing how deep learning techniques can be used to decode pixelation and other face-blurring techniques is a great example of this. Pixelation is a well-established method of protecting individuals' privacy in videos, but if AI can undo it then the technique becomes much less valuable. The really interesting thing is that the researchers didn't need to specify what features they were looking to uncover in advance. The neural networks uncovered all visually hidden information.
The immediate message here is that someone needs to develop a better technique for blurring out faces, that's not vulnerable to neural nets decoding it. But the wider issue is that there will be many applications or problems that were previously very hard to solve, but may now be a lot easier. The trick will be in finding these problems first and capitalising on that.
After being trained on a dataset of 530 individuals, the researchers’ neural net was able to recognize pixelated images at a rate of 57 percent