Artificial intelligence (AI) is revolutionizing entire industries, including libraries. Although this technology offers immense potential to enhance services, library workers and others must consider its environmental impact.
Training and operating AI tools like large language models, for instance, require substantial computational power. Generating one image could use as much energy as charging an average smartphone, according to a 2023 study by AI developer Hugging Face and Carnegie Mellon University in Pittsburgh, the first study of its kind to measure the carbon and energy impact of generative AI models. Data centers worldwide also use massive amounts of water to cool AI servers. A 2023 study from Cornell University estimates that by 2027, the water needed for AI could reach 1.7 trillion gallons—more than half of the water the UK uses in a year.
Libraries and their decision makers should consider the following strategies when possible.
First, libraries should prioritize energy-efficient hardware and cloud providers to reduce energy consumption. This includes choosing servers with high energy efficiency ratings, such as ENERGY STAR certification, and using sustainable features like power-saving modes. Additionally, more efficient large language models result in algorithms that can minimize the number of computations required, further reducing energy consumption. It is essential that libraries carefully consider the specific needs of their users and avoid unnecessary AI applications.
Second, libraries should select or advocate to upper management for data center locations that prioritize renewable energy. Some data centers are powered by sustainable sources such as solar or wind, which are crucial for reducing carbon emissions. What’s more, using local data centers can minimize the distance that data travels, reducing the energy consumed to transfer data.
Libraries should select data centers that prioritize renewable energy.
My library has developed a rubric for assessing AI use. In addition to energy consumption and carbon emissions, here are some other environment-related takeaways.
Resource depletion. Consider the environmental impact of the increased demand for rare-earth elements used in computer chips, including for AI hardware.
Long-term sustainability. Assess the longevity of AI technologies and their potential for innovation to minimize energy and water consumption. Consider questions including: Is there a plan for recycling or repurposing the AI hardware you’re using? Or, how is pollution being mitigated from the use of these systems?
Ethical implications. Evaluate the potential for bias, discrimination, and job displacement that comes with automation.
Beyond the ecological impact, choosing AI tools that are trained on diverse and unbiased data is essential to avoid perpetuating stereotypes and harming groups that are most vulnerable to adverse environmental effects. By contracting with companies that use diverse data sets, libraries can implement AI systems that are fair and equitable. Promoting transparency in AI algorithms can build trust and accountability. By understanding how AI algorithms work, libraries can identify and address potential biases. Protecting user privacy and data security is also crucial to ethical AI implementation. By implementing strong data protection measures, libraries can safeguard sensitive information.
Librarians can play a crucial role in promoting sustainable AI practices. By raising awareness about these tools’ environmental and social impacts, collaborating with information technology professionals to identify energy-efficient solutions, educating users about ethical AI, and supporting research into sustainable AI, librarians can help ensure it is used responsibly and minimize its negative consequences. This will contribute to a healthier and more equitable future.
