Current computer vision systems produce a large number of images, many of which are never seen by humans. Yet, in order to understand, critique and shape the impact of machines that can see in ways that exceed human vision capability, humans will need to learn to see like machines, to understand their abstractions and their categorizations of things in the world. The title of the work refers to the measurement of perfect human vision (20/20), which is contrasted with a yet unquantifiable measure of seeing (specifically in a cultural context) for a computer vision system - represented by the variable "X." Audience members are invited to experience the process of seeing in a complex neural network based computer vision system and determine the value of "X" for themselves. Based on an advanced computer vision system developed by Dr. Eugenio Culurciello and Alfredo Canziani in the School of Biomedical Engineering at Purdue University, 20/X explores visuality, representation and knowledge in the age of intelligent seeing machines. Presented as either a freestanding observation station or an immersive viewing apparatus, the installation provocatively asks the question: do we need to acquire new literacy skills in an emerging visual culture of computer vision?
20/X reveals the process of machines looking at the world, leading up to but not including the final classification and categorization of objects in a live video feed. The interactive interface allows users to trace the levels of abstraction employed by the algorithm—from coarse and geometry driven in the beginning to more specific and detail-oriented in the end at which point distinctive patterns, areas and objects that "excite" the computer vision system can be identified. The authors are interested in these processes before the final classification precisely because they reveal the logic and strategies of the machine vision system, thus helping humans to "see like a machine" rather than just providing visitors with the almost magical technological feat of naming the objects it sees in an image. While this is important for a different set of applications, the authors believe that an experience of the system in this more ambiguous form is most insightful to anyone who is interested in learning to understand how machines see the world around us and anyone who is interested in asking critical questions about this process or shaping it.