Big data AI’s environmental toll prompts debate over small data future

Small Data AI alone cannot counter the environmental and social costs of Big Data, the study concludes. A wider sociotechnical transformation is necessary, involving political, cultural, and economic change. Efficiency gains risk triggering rebound effects, where lowered costs simply accelerate wider adoption and higher overall consumption. Without shifting incentives, the voracity of Big Data AI is likely to persist.

Artificial intelligence faces mounting scrutiny for its environmental and social toll, with new research warning that the dominant Big Data paradigm is pushing the technology toward unsustainable extremes. The analysis argues that a transition toward “Small Data AI” could mark a more sustainable path forward if supported by broader social and political change.

The study, titled Small data AI as a more sustainable next-wave AI? Exploring a potential sociotechnical shift away from the voracity of big data and published in AI & Society, builds a comprehensive taxonomy of artificial intelligence’s environmental footprint and explores whether emerging small data methods can reduce the strain on global resources.

How voracious is big data AI?

The research frames Big Data AI as fundamentally voracious. Beyond the well-known “three Vs” of volume, variety, and velocity, the study highlights a fourth V, voracity, to describe the system’s insatiable hunger for data, energy, water, and natural resources. This voracity extends beyond technical requirements to encompass economic and cultural dynamics shaped by surveillance capitalism and growth-driven business models.

The study underscores that Big Data AI’s lifecycle intensifies environmental costs at every stage. Manufacturing chips demands extensive mining of rare earths and large quantities of water. Training large-scale models consumes enormous energy and cooling resources. Deployment of everyday tools, from search engines to conversational systems, adds to global power use. Finally, rapid obsolescence and limited recycling produce mounting electronic waste, much of which ends up exported to low-income countries.

The taxonomy points out interdependence. Large datasets necessitate greater computing power, which in turn drives demand for advanced chips, resource extraction, and increased waste. This chain effect highlights why incremental efficiency gains may not be enough to curb the trajectory of Big Data AI.

Can small data AI reduce the burden?

The study positions Small Data AI as a possible alternative. Defined not as a singular method but a set of approaches, Small Data includes manual and community-based data collection, targeted datasets for specialized applications, and machine learning techniques designed to perform under scarcity, such as few-shot learning, transfer learning, and active learning.

The potential benefits are significant. Smaller, carefully curated datasets could limit energy-intensive training. Hybrid approaches that combine symbolic reasoning, expert knowledge, and statistical learning could reduce dependence on massive datasets while retaining accuracy. In healthcare, for example, personalized or localized data can generate effective results without requiring global-scale repositories.

Yet the study warns against overstating the technical solution. Popular methods like transfer learning and data augmentation, while framed as efficient, can reproduce the resource demands of Big Data if deployed at scale without rethinking underlying goals. True sustainability, the author notes, depends not only on technical adjustments but also on addressing structural issues that drive the appetite for more data and larger models.

What would a sociotechnical shift require?

Small Data AI alone cannot counter the environmental and social costs of Big Data, the study concludes. A wider sociotechnical transformation is necessary, involving political, cultural, and economic change. Efficiency gains risk triggering rebound effects, where lowered costs simply accelerate wider adoption and higher overall consumption. Without shifting incentives, the voracity of Big Data AI is likely to persist.

Policy intervention is therefore critical. Regulation and public funding could encourage alternative AI ecosystems that prioritize sustainability over scale. The European Union, with its focus on digital sovereignty and regulatory oversight, is highlighted as a region that could foster Small Data development by reducing dependence on Big Tech’s growth-driven model. Encouraging smaller, more localized AI systems could also align with broader goals of inclusiveness, transparency, and reduced geopolitical risk.

The study stresses that a move toward Small Data AI does not represent a regression to earlier symbolic AI, but rather the potential rise of hybrid models blending symbolic reasoning, human expertise, and machine learning. These approaches may support a more diverse and resilient AI ecosystem, less dependent on global data extraction and more aligned with sustainable development.