How do you see me?

Year
Country
United States
Publication Type
Technologies referenced
Topics
Sentiment
Description (in English)

InĀ How do you see me?Ā I utilized adversarial processes, algorithms that are designed to deceive machine learning systems, to generate self portraits that are recognized as my face, although they look nothing like me, or like any human face.

How do you see me? is an attempt to get closer to the other that is watching me, watching us all the time.
we live in a world in which we are constantly be looked at, studied, analyzed. Cameras are everywhere.
these systems know a tremendous amount about me - but what do I know about them?
How do you see me? is my way of looking back and trying to learn how this alien intelligence that is so attentive to my every move is structured, internally.

To do this I broke the problem into two pieces: face detection and recognition of my face. The images below show the generative process (using artificial evolution) to create images which, over hundreds of generations, come to be either detected as a face, or recognized as my face, by facial recognition algorithms.

What does it mean?
-In the first algorithmic exploration [indicated by green squares] we discover a template of what defines ā€œa faceā€ to a face detection system. and we see that it is a white oval shape.
-In the second [indicated by blue circles] we gain a glimpse inside the black box of a neural network trained to extract key features from faces. We can see a bit of the internal representation structure by seeing images which neighbor my face in feature space.Ā 

In machine learning, a model is an artifact produced through a training process in which a learning algorithm is exposed to data. Within the context of AI, the model makes an implicit claim to represent its subject. We have to be cautious of this tendency to see models as objective representations and rather see them as framing identity through a particular lens of experience (experience of the data).
And then we have something much less intuitive to grasp, and that is the multiplicity of the subject. Because the feature vector, the abstract representation of "my face," has so many neighbors that are so vastly different, so alien to human knowledge. And this shows us also that we should exercise extreme caution in handing control over to intelligences, to automated systems, that we cannot intuitively relate to or understand.

Source:Ā https://www.deweyhagborg.com/projects/how-do-you-see-me

Situation machine vision is used in

Authored by

UUID
a5a7db7e-1618-4cc3-90ae-381223970df7