Key points are not available for this paper at this time.
In April 2025, the Royal Society hosted a discussion meeting titled 'Bits, neurons and qubits for sustainable AI.' The meeting focused on the design and analysis of computing technologies for sustainable artificial intelligence (AI) that leverage information-theoretic and physical properties of computing platforms beyond standard deterministic digital processors.This is a timely topic because of three converging trends in computing and machine learning: (i) the emergence of novel informational principles exploiting noise and randomness (as seen in generative AI models), (ii) the development of energy-efficient hardware inspired by neuroscientific principles and the architecture of biological brains (neuromorphic computing) and (iii) the availability of noisy intermediate-scale quantum (NISQ) computers, which have catalysed the new paradigm of quantum machine learning.These trends signal a shift away from the last 70 years of transistorbased digital computing toward alternative, potentially more sustainable and scalable forms of computing for intelligent systems.By examining these alternatives, the meeting highlighted how future AI systems might be designed and deployed across a wide range of engineering applications-from nanoscale biomedical devices to large-scale industrial systems-in a more sustainable manner.The discussion covered state-of-the-art developments in neuromorphic processors (which process information in fundamentally different ways than traditional CPUs/GPUs) and quantum computing platforms for AI, as well as theoretical advances in understanding the role of noise and information in modern AI.Overall, the meeting provided a forum
Simeone et al. (Thu,) studied this question.