AI in libraries: sustainability, responsibility, and a practical path forward

Artificial intelligence is reshaping research, discovery, and service delivery across public and academic libraries. At the same time, it has raised legitimate concerns, particularly about environmental impact, governance, and long-term stewardship of public resources.

Rather than focusing on whether AI should or should not be adopted, the more useful question is how libraries can adopt it responsibly.

How can libraries adopt AI in ways that strengthen service while remaining accountable to environmental, ethical, and institutional responsibilities?

That requires clarity about both impact and control.

Sustainability In Library Practice

Libraries are mission-driven institutions. They are accountable not only for the services they provide, but for how those services are delivered. Public funding, academic oversight, and professional ethics require careful stewardship of financial, technological, and environmental resources.

Recent analyses show that data centers now account for a growing share of national electricity consumption. In the United States, estimates suggest data centers have increased from roughly 2 percent to more than 4 percent of national electricity use in recent years, with projections rising as AI workloads expand. Water use for cooling large facilities has also increased.

These figures matter. But they require context.

Libraries are not operating hyperscale AI models. Most library use cases involve targeted applications: search enhancement, metadata assistance, accessibility tools, or workflow support. The environmental footprint of AI depends heavily on model size, frequency of use, infrastructure efficiency, and deployment design.

Looking at these variables together provides a clearer picture of AI’s environmental profile.

What Actually Drives AI’s Environmental Impact

AI systems consume energy primarily in two phases:

• • Training large models
  • Running inference (everyday usage)

Training frontier-scale models is computationally intensive and energy-heavy. In contrast, many day-to-day applications rely on pre-trained models and comparatively smaller inference tasks.

The environmental profile of AI use varies significantly depending on:

• • Whether new models are trained or existing ones are used
  • The size of the model selected
  • How efficiently prompts are structured
  • Whether repeated outputs are cached
  • The energy mix of the hosting infrastructure

Not all AI deployment looks the same. A continuously running global recommendation engine is fundamentally different from a scoped, opt-in feature embedded in a library platform.

For library leaders, the relevant question is not whether AI has an environmental impact, it does . It is whether implementation choices minimize unnecessary computational load while delivering meaningful benefit.

Sustainability As Part Of Responsible AI Governance

Environmental impact does not stand alone. UNESCO’s Recommendation on the Ethics of Artificial Intelligence explicitly includes sustainability and environmental protection alongside human rights, transparency, and accountability.

For libraries, this aligns with existing commitments: intellectual freedom, privacy, equitable access, and prudent use of public funds.

Responsible AI adoption therefore involves asking:

• • Does this application meaningfully improve service or reduce administrative burden?
  • Is the computational cost proportional to the benefit?
  • Can the system be audited, configured, and governed locally?
  • Is human oversight preserved?

If those questions cannot be answered clearly, implementation should pause.

Where AI Can Strengthen Sustainability

AI can also contribute to operational sustainability when applied deliberately.

Examples include:

• • Improving discovery across digital collections without duplicating infrastructure
  • Assisting with accessibility features such as summarization or transcription
  • Supporting staff with metadata enhancement at scale
  • Reducing repetitive administrative workflows

In these contexts, AI can streamline processes without replacing professional judgment. The goal is not automation for its own sake. It is capacity reallocation. Protecting staff time for research support, instruction, community engagement, and policy oversight.

However, efficiency gains must be measured against environmental and governance considerations. Sustainability is not achieved through speed alone.

The Responsibility Of Technology Partners

Libraries rarely control the underlying AI infrastructure. Vendor design choices matter.

Technology providers have a responsibility to reduce unnecessary computational intensity, provide transparency about how AI features operate, and enable meaningful configuration and oversight.

At Clarivate, sustainability considerations are embedded in how AI capabilities are designed and deployed across academic and library solutions.

Within our Academic AI platform, we prioritize:

• • Using pre-trained models rather than training new large-scale models from scratch
  • Selecting lightweight models appropriate to the task
  • Optimizing prompts and output length to minimize unnecessary token generation
  • Caching responses for repeated use to avoid redundant large language model calls
  • Applying text compression before processing documents to reduce data volume

These measures do not eliminate environmental impact. They are intended to reduce avoidable computational load while maintaining quality and transparency.

Our AI capabilities are deployed on optimized cloud infrastructure from providers with established sustainability commitments. We also maintain centralized AI governance to ensure oversight, accountability, and consistent standards across implementations.

More broadly, Clarivate reports greenhouse gas emissions annually in alignment with the GHG Protocol and has committed to achieving net zero emissions before 2040. Sustainability is integrated into infrastructure, governance, and long-term strategy, with oversight and reporting mechanisms in place to ensure accountability.

Libraries should expect this level of transparency and operational discipline from all technology partners.

Practical Criteria For Leaders

Library directors and senior administrators evaluating AI initiatives may consider:

• • Is the proposed use case specific and purpose-driven?
  • Are sustainability considerations part of procurement discussions?
  • Can the vendor describe how computational load is minimized?
  • Is human oversight built into the workflow?
  • Is adoption phased, with review and evaluation points?

Sustainability should be treated as a design constraint, not an afterthought.

Engagement As Stewardship

Libraries contribute most by engaging thoughtfully with AI, helping shape standards, demanding transparency, and modeling responsible use.

This includes publishing clear AI usage statements, participating in professional dialogue, and collaborating with vendors and policymakers to refine best practices.

While environmental impact cannot be fully eliminated, it can be managed through proportionality, transparency, and governance.

AI will continue to evolve. The institutions that maintain public trust will be those that integrate new capabilities without compromising core values.

Sustainability, responsibility, and technological advancement are not mutually exclusive. But alignment requires deliberate choices about model selection, infrastructure, oversight, and scope.

Libraries are well positioned to make those choices thoughtfully.

The future of AI in libraries will be shaped not by how quickly tools are adopted, but by how carefully they are governed.