Learning to See is a series of art works created by Memo Akten, using machine learning algorithms to reflect on how human cognitive biases shape our "seeing" and understanding of the world. The core metaphor of the work is: an artificial neural network looks at the world and tries to understand what it sees, but it can only see through the filter of its existing knowledge. Just like us humans, we see things not as they are, but as we are.
Purpose and Themes The series is part of a broader exploration of cognitive bias, the inability to see the world from another's perspective, and the resulting social and political polarization. The work points out that the images we see in our consciousness are not a mirror of the external world, but a reconstruction based on expectations and prior beliefs. "Seeing" here refers both to low-level perceptual and phenomenological visual experience, and to high-level cognitive meaning-making and the construction of "truth."
Main Forms and Capabilities The series includes several video works (such as True Colors, Mountains, Gloomy Sunday, Supergan, Hello, World!, We are made of star dust, Learning to dream, Studies, etc.) as well as an interactive installation. In the interactive installation, a custom neural network analyzes live camera footage pointed at a table covered with everyday objects, and attempts to reconstruct the image in real time using prior experience (i.e., training data) and knowledge (i.e., what was learned during training). The work also includes tags such as artificial intelligence gallery, NFT, open source, and perception.
Usage and Open-Source Resources The author released an open-source demo as the basis for the series and many works. The demo allows real-time preprocessing of camera input (such as a webcam) and feeds it into a neural network for real-time inference. The source code and example pre-trained models can be found at https://github.com/memo/webcam-pix2pix-tensorflow Download. More information can be found in the 2019 SIGGraph paper and Chapter 5 of the author's doctoral thesis. The project was created during a PhD at Goldsmiths, University of London, and was funded by EPSRC.
Applicable Scenarios The work is suitable for art exhibitions, academic research, AI art discussions, cognitive science reflection, and other scenarios. The work has been presented in multiple exhibitions, performances, and talks, and has been cited in numerous books, papers, media reports, and academic papers. Specific exhibition, publication, and citation information can be found in the detailed list in the original text.
Limitations The materials do not provide specific usage steps, fee models, or pricing information for the work, nor do they provide functional instructions that ordinary users can directly operate. The work is mainly an art project and academic research, and the open-source code is available for technical users to explore on their own.
