From Non-Maleficence to Beneficence: Expanded Ethical Computing in the Era of Large Language Models

As modern society grows increasingly complex, access to essential services such as healthcare, legal aid, tailored education, and psychological support remains heavily gated by socio-economic, neurological, and systemic barriers. This paper explores the transformative potential of Large Language Models (LLMs) and Generative Artificial Intelligence not merely as industrial productivity enhancers, but as vital “social scaffolds” capable of fostering a more inclusive society. Crucially, we propose a paradigm shift in the concept of Ethical Computing—moving from a passive defensive framework of non-maleficence (“do no harm”) to an active mandate of beneficence, where AI systems are explicitly developed to serve marginalized and un(der)served populations. Through this expanded ethical lens, we systematically analyze the democratizing impact of AI across four primary axes of inclusivity: socio-economic (providing zero-cost medical triage and legal translation for undocumented populations), neurospicy (acting as a non-judgmental communicative bridge for individuals with Autism Spectrum Disorder), pedagogical (delivering hyper-personalized executive function support for Special Educational Needs), and psychological (serving as an accessible, first-level triage system for mental health crises). By framing LLMs as a modern social safety net, we outline a clear trajectory for future research, advocating for an “ethical-by-design” development paradigm that explicitly prioritizes equity, accessibility, and the active dismantling of historical barriers for the digitally and socially disenfranchised.