Problem

Artificial intelligence (AI) is rapidly evolving into a general-purpose technology with profound yet largely uncertain impacts on environmental sustainability and and human well-being. Existing research addressing these impacts is often fragmented and siloed, tending to focus on specific issues or disciplinary perspectives, and frequently examining environmental or social dimensions in isolation. This limited integration and breadth create a significant gap in understanding the full scope of AI's consequences. There is a pressing need for a comprehensive analysis that maps the entire landscape of AI's influences, identifies key insights, highlights underexplored areas, and guides future research and policy development. This paper aims to address this by systematically reviewing the literature on AI’s environmental and well-being impacts, the methods used for assessment, and the critical research gaps that emerge.

Method

This study conducted a systematic literature review spanning publications from 2010 to 2024, adhering to the PRISMA 2020 framework. Searches were performed across major academic databases including Scopus, arXiv, and NBER Working Papers, complemented by backward and forward citation tracking to ensure comprehensive coverage across various scientific disciplines. Inclusion criteria mandated studies be published in English, within the specified timeframe, from peer-reviewed sources or reputable grey literature, with full text availability, and directly addressing the research questions. A blind screening process was employed for both title/abstract and full-text stages. Data were systematically extracted into a spreadsheet using an iteratively developed coding framework. For studies deemed less relevant, data extraction was initially aided by an LLM (Microsoft Copilot) and subsequently manually verified for accuracy. The analysis incorporated both quantitative mapping of impact types, scopes, methods, analytical scales, and sentiment, as well as qualitative synthesis of recurring findings, research gaps, and a proposed future research agenda.

Results

The review analyzed 1,291 relevant studies, with a significant increase in research activity in both environmental and well-being domains since 2018. A notable divergence exists between these two literature streams. Environmental studies, comprising 523 papers, typically present a highly optimistic outlook, with 83% suggesting positive impacts. These studies are predominantly quantitative, micro-level, and narrowly focused on energy consumption and CO2 emissions, accounting for 72% of the environmental sample, while broader indicators like water, material, and biodiversity impacts receive minimal attention. Critically, only 11% of environmental studies address systemic impacts, overlooking potential wider consequences such as rebound effects. Conversely, well-being research, consisting of 768 papers, exhibits a more balanced sentiment (44% positive, 46% negative) but is largely conceptual or theoretical (54%), with limited empirical evidence. These studies tend to adopt a macro-level perspective, frequently examining social cohesion, inequality, and employment. While AI is often seen positively for income and health impacts, it is viewed negatively for existential risks (65%), inequality (57%), social cohesion (51%), and employment (48%). Upstream supply chain impacts on well-being, such as precarious microwork, are rarely investigated and generally carry a negative sentiment. The paper also expands Kaack et al.'s framework to categorize well-being impacts into computing-related, application-level, and systemic, revealing that computing-related social impacts (e.g., from supply chains) are significantly understudied (1%).

Implications

Based on these findings, future environmental research on AI must broaden its scope beyond energy use and CO2 emissions to comprehensively include water, material consumption, land use, non-carbon pollutants, and biodiversity across the entire AI life-cycle. This requires employing life-cycle assessment and environmentally-extended input-output analysis to account for global supply chain impacts and potential burden shifts. Furthermore, there is a critical need to quantify systemic impacts, particularly rebound effects, using scenario-based approaches. For social and well-being impacts, a significant shift towards more empirical investigation, including qualitative and participatory methods, is essential. Research should extend to cover upstream supply-chain impacts, such as data worker conditions, local community burdens, and ecologically unequal exchange. Greater attention must also be paid to subjective dimensions of work, cognitive capabilities, and overall subjective well-being, as these are currently neglected but crucial for understanding AI's deeper human implications. The review underscores the necessity of an integrated approach, considering environmental and well-being impacts simultaneously, as they are often intertwined. Ultimately, guiding AI development towards environmental sustainability and human flourishing requires a more holistic, inclusive, and anticipatory research agenda that acknowledges AI as a socio-technical system shaped by deliberate choices.