Circular Legibility: From Data Extraction to Data Care, Designing Sustainable AI Systems

As organisations race to build increasingly data-hungry AI systems, environmental and ethical costs often remain invisible. This session introduces Circular Legibility as a framework for evaluating AI across its full data lifecycle, from extraction and processing to retention, deletion, and long-term stewardship, and brings this into dialogue with the technical concept of the AI Memory Wall.

In computing, the AI memory wall describes a performance bottleneck where increasingly powerful AI processors outpace the ability of memory systems to supply data, leaving hardware under-utilised. This session extends that definition into an ESG and governance context, positioning the AI memory wall as both a technical constraint and an ethical environmental threshold, the point at which additional data produces diminishing performance value while exponentially increasing ecological and infrastructural costs.

Through case comparisons (e.g., 100GB vs 30TB storage footprints), visual mapping, and organisational scenarios, the presentation examines AI memory as a material infrastructure with tangible impacts: energy consumption, cooling water use, hardware waste, and rare-earth mineral extraction. Participants are invited to rethink the assumption that “more data equals better intelligence,” and instead consider how sustainable storage, responsible deletion, and data minimalism can become core ESG practices rather than afterthoughts.

The session ultimately asks: What would it mean for organisations to treat data not as an infinite resource to be extracted but as a finite ecological responsibility to be cared for?

Key Takeways:

-AI Memory as Infrastructure: Recognising that digital storage has physical and environmental consequences. Which include carbon emissions, water consumption, and hardware supply-chain impacts.

-The AI Memory Wall Dual Lens: Understanding the term both as a hardware performance bottleneck and as an ethical-environmental threshold where data accumulation exceeds meaningful utility.

-Circular Legibility Framework: A practical model for assessing AI systems across their entire data lifecycle.

-Data Minimalism vs Data Maximalism: Concrete comparison of storage scale and how excessive data retention increases environmental footprint without proportional decision-making gains.

-Embedding Environmental Care in AI Ethics: Moving beyond abstract principles towards measurable ESG actions tied to storage policies, retention limits, and reversible data practices.