Skip to content
Thesis Info
- LABS ID
- 00923
- Thesis Title
- How machines see the world: Five essays on biological and artificial vision
- Author
- Carloalberto Treccani
- E-mail
- carloalberto.treccani AT gmail.com
- 2nd Author
- 3rd Author
- Degree
- PhD
- Year
- 2020
- Number of Pages
- 196
- University
- City University of Hong Kong
- Thesis Supervisor
- Olli Tapio Leino
- Supervisor e-mail
- OTLEINO AT cityu.edu.hk
- Other Supervisor(s)
- Maurice Benayoun
- Language(s) of Thesis
- English
- Department / Discipline
- School of Creative Media
- Copyright Ownership
- Carloalberto Treccani
- Languages Familiar to Author
- English, Italian
- URL where full thesis can be found
- Will be available after institution approval
- Keywords
- Vision, biological vision, artificial vision, machine vision
- Abstract: 200-500 words
- This thesis is a journey into the domain of biological and artificial vision. The aim of this thesis is to investigate how vision emerges in biological and artificial ‘creatures’; furthermore, it outlines the foundation for a wholly new understanding of vision.
The thesis, positioned at the intersection of art, aesthetics, neuroscience, and artificial intelligence studies, begins by questioning why and how the world appears articulated into objects, distinct from each other and with precise characteristics. The most immediate answer is certainly: “because the world has those characteristics”. However, as shown by several psychophysical studies, biological visual systems cannot retrieve the physical properties of the world; retinal images, for instance, conflate the properties of objects – size, orientation and luminance – and therefore cannot be used to reveal what the world looks like. Nevertheless, despite these problems, visual-guided behaviours are normally successful. Given this paradox, this thesis investigates the strategies that visually gifted creatures have implemented to circumvent these obstacles, and what the conquest of artificial vision suggests about how vision works.
With these purposes in mind, the thesis, first, highlights difficulties, complications and erroneous convictions about vision, that the attempts at the creation of visually ‘intelligent’ machines, since the 1950s, has unveiled. Secondly, it discusses the role played by Artificial Neural Networks to better comprehend how biological vision works. From there, using the driving principle of neural network technologies, i.e., trial and error, the thesis proceeds to formulate an empirical approach to vision, able to connect biological and artificial vision. Consequently, the threefold objective of this research is: first, to demonstrate that vision, across the whole spectrum of the biological and the artificial domains, has emerged through random trial and error; secondly, to examine the implication of this new empirical theory of vision in understanding why we like what we like, as well as investigate what Artificial Neural Networks suggest about the way we visually appreciate artworks, and more in general, the world; and thirdly, to explore the possibility of understanding visual appreciation through biology.
Finally, the thesis tries to foresee several future scenarios of the co-evolution of human and machine vision. The hitherto unexplored scenarios that the arrival of 'intelligent' artificial vision and the prospect that neural controlled technologies opened oblige us to reconsider vision as a diffuse practice across visually gifted entities, and call us to re-examine the visual modalities in which we see the world and our position in it.