From: jimruttshow8596
The discussion of analog and digital systems holds significant implications for understanding complex systems, particularly in the context of biological cognition and the design of artificial intelligence.
Analog Control in Complex Systems
Top-down control in complex dynamical systems is suggested to be analog in nature, representing a continuous change in system settings rather than discrete states, akin to a dimmer switch over a toggle switch [00:30:09]. This analog control enables processes like homeostasis to maintain both timeliness and sensitivity to local conditions [00:30:24]. The emphasis on analog control by figures like Freeman and George Dyson is noted as potentially insightful [00:30:46].
Comparison of Analog and Digital Processing
In terms of computational efficiency, analog systems are noted to be six orders of magnitude more efficient than digital systems for comparable tasks [00:32:12]. This efficiency is considered a key reason why biological brains, which process vast amounts of information, do not overload [00:32:22].
Key characteristics and relationships include:
- Underlying Nature Digital systems are fundamentally analog at lower levels, with clever engineering techniques used to transition between analog and digital representations [00:32:00].
- Brain Function The brain itself operates as a hybrid system, utilizing both digital elements (like neurons) and analog processing, switching between them as needed [00:32:30].
- Cognitive Representation Higher-level cognitive functions, such as the domain of ideas, concepts, or object ontologies, involve continuously variable classifications, which are characteristic of analog processing. Yet, these are implemented on digital neural circuitry [00:33:09]. This implies that while actions might be carried out digitally, the top-down commands or high-level decisions are analog [00:33:31].
Simulation and Cost
While it is theoretically possible to simulate analog processes to any desired level of detail using digital computation, achieving extremely fine detail is computationally very expensive [00:33:46]. Analog systems, by contrast, offer this precision inherently and “for free” [00:33:59].
Analog Concepts in Human Systems
Even in predominantly digital infrastructures like the internet, the focus shifts to analog-like relationships within social media, such as “what’s connected to what” [00:34:14]. People themselves, in their interactions and the “dance” of their collective behavior, are considered analog, forming “condensated nodes” at the intersections of various constraints [00:34:22].
The early work in artificial neural networks that processed words, particularly recurrent networks, demonstrated that injuring parts of the network below feedback loops could lead to “deep dyslexia” where the network would output semantically related but visually incorrect words (e.g., “band” becoming “orchestra” or “bed” becoming “cut”) [00:37:16]. This suggested that the middle layers of the network created “semantic attractors” that had causal power over the output [00:38:31]. This phenomenon can be seen as an example of an emergent property exerting causal influence within an AI system [00:38:45].