Chapter 5: Full-analogue Photonic AI for Embracing the Uncertainty of the Environment

Traditional ML methods are time- and energy-consuming. The digital (discrete, sampling) representation of natural signals destroys its spectra. It is accompanied by a growing number of parameters for achieving high-level accuracy. Modern computers with high-performance chips for AI require many thousands of times more energy than the human brain. The size of an atom sets a limit on the reduction of computer chips and their performance growth.

AI helps humans understand the environment by immersing them in a hybrid reality and supporting purposeful human–computer interaction through convergent conditions for AI’s inference explanations (Lepskiy, 2018; Perko, 2020; Raikov et al., 2024). However, AI’s digital character and replacing the natural continuous signal with its digital values distort the computer’s reflection of reality. The digitalisation of the continuous signal is accompanied by an exponential increase in ML parameters and energy consumption.