CPDP.ai 2024 - Computing Using Physics; What Can AI learn from Analog Computing

Computing Using Physics; What Can AI learn from Analog Computing?
Ever since the late 1940’s, the Turing Machine has been the central paradigm of how computing machines are defined and designed, including the material devices that power the current wave of AI. But there are other paradigms of computing that have known a certain measure of success in the past or that are currently being developed, including different forms of analog computing. Using the properties of materials or with a minimum of electronic components, some of these analog methods of computing exploit the laws of physics to obtain quantitative results. Such methods were used in special purpose computers that were built to only solve specific equations. Other devices known as General Purpose Analog Computers have a generality that is similar to Turing machines but compute with continuous values and are inherently parallel. All these devices use orders of magnitude fewer components and energy to perform their computations. Together, they show a bewildering variety of approaches that is in marked contrast to the standardized world of universal Turing machines. Are there things that AI can learn from analog computers?
• Turing machines are universal: within material limitations, every computer can in principle perform all the computations that any Turing machine can perform. Is standardization inherent to such a view of computing ? Does universality hinder innovation?
• One could argue that Turing machines as well as most of the alternatives to it were inspired by the human brain. How is it that one source of inspiration gives rise to such different views of computing?
• AI is currently an important driving force for software and hardware development. Is there a role for analog computing in the AI realm ? How would using a form of analog computing change current thinking around AI?
• Artists have speculated on forms of computing that are more situated and ecological. These speculations go beyond concerns for energy-efficiency and often include a reflection on the seemingly inherent connection between computing and control. Can analog computing help us think of a more open-ended form of artificial intelligence?
Organised by Privacytopia (BE)
Moderator Darian Meacham, Maastricht University (NL)
Speakers Manuel Le Gallo-Bourdeau, IBM (CH); Joana Moll, Academy of Media Arts, Köln (DE); Joost Rekveld, KASK Gent (BE); Francien Dechesne, eLaw Center for Law and Digital Technologies Leiden University (NL)