Editorial:
From Silicon to Society:
Efficiency, Security, and Responsible
Integration of Artificial Intelligence

The progress of artificial intelligence (AI) is no longer determined by algorithms and models alone. It depends equally on the hardware on which AI workloads are executed, on the security of models once they are deployed, and on the ability of organizations to integrate AI responsibly into everyday practice. This special issue brings together contributions that address these challenges at different levels - from post-Moore computing substrates and the confidentiality of trained neural networks to AI-supported clinical workflows, the use of generative AI in higher education, and the data-driven operation of electric power grids on the path to decarbonization.

Although the papers belong to distinct research fields, they share a common perspective: rather than evaluating an isolated component, they examine how technology behaves as part of a larger system. They show how total energy and data movement determine the efficiency of AI hardware, how the outputs of a deployed model can reveal the model itself, how physics-based models and machine learning can be combined to manage uncertainty in the power grid, and how data, analytics, institutional policies, and human judgment must be coordinated for AI to deliver real value in hospitals and universities. By combining formal, experimental, and empirical approaches, the issue aims to stimulate dialogue between the hardware, security, and application communities.

Even though the papers in this issue address different domains and levels of abstraction, they converge on a common conclusion: the impact of AI and other computational methods is determined not only by the capabilities of models and algorithms, but also by the efficiency with which they are executed, the protection of the knowledge they embody, their grounding in the physics of the infrastructure they help to operate, and the workflows, policies, and competencies that govern their use. We hope that readers will find these contributions useful and that they will inspire further research across disciplinary boundaries. We sincerely thank all the authors for their contributions and the reviewers for their careful and constructive evaluations.