Broadening the scope: Nine dimensions of Sustainability and AI development

The rapid expansion of artificial intelligence (AI) has brought increasing attention to the material, energetic, and labor conditions underpinning digital technologies across their life cycles. In response, emerging initiatives, labeled 'green computing', 'sustainable AI', and 'AI for the planet', seek to reconcile the resource-intensive nature of AI with environmental sustainability. Within AI ethics and the environmental ethics of technology, the concept of sustainable AI has gained prominence. The central claim of this paper is that these approaches represent significant progress in debating environmental issues and AI, however a deeper and broader examination of sustainability is needed. AI development is challenged an intensifying environmental crisis and evolving geopolitical security frameworks, both of which heighten resource constraints and ethical tensions. The aim of the paper is to expand the reflection on sustainable AI. To do this, it identifies and maps nine key dimensions of sustainability in the context of health-related AI, drawing on a broad conception of sustainability as the preservation of material foundations, social conditions, and viable options for living and action. The dimensions are: (i) environmental and ecological sustainability, (ii) economic sustainability, (iii) societal sustainability, (iv) professional sustainability, (v) data sustainability, (vi) funding sustainability, (vii) sectoral sustainability, (viii) democratic sustainability, (ix) epistemic and institutional sustainability. While the first three dimensions are widely discussed in ethical and policy discourse, the remaining six are largely marginalized or treated only in niche debates, often falling outside the scope of traditional individual ethics. Through a dual analytical strategy—what one may call zoom-in and zoom-out—the paper reveals these dimensions: the zoom-in focuses on granular technical and infrastructural aspects (e.g., data provenance, interoperability), while the zoom-out attends to the broader socio-historical, environmental-historical, political, and institutional contexts and conditions of AI development.