The Challenge and Opportunity of AI and Sustainability
The explosive rise of artificial intelligence (AI) technology has led to tantalizing potential for sustainability solutions as well as enormous challenges for environmental, social, and economic governance. Intentional, responsible, safe, and sustainable AI use is profoundly important, and it also aligns with UC values and policy. The UC Responsible AI Principles includes “shared benefit and prosperity,” meaning that “AI-enabled tools should be inclusive and promote equitable benefits (e.g., social, economic, environmental) for all." The UC Policy on Sustainable Practices also mandates that procurement, energy consumption, and campus data center operations align with the university's strict net-zero carbon, clean energy, and ethical supply chain standards. This webpage provides tips and resources to help readers navigate these overlapping issues.
If you’ve discovered a way to use AI more efficiently or have feedback on these tips and resources, please share your thoughts in our survey.
Tips for Greener AI Use
Artificial intelligence and the data centers that support it require significant energy, water, and critical minerals. The following checklist offers ways to minimize our collective environmental footprint in accordance with UC Berkeley values.
- Be Intentional: is AI necessary for your task? Consider disabling AI services that are unneeded or offer minimal value.
- Consolidate Prompts: One well-structured prompt is more energy-efficient than a long "chat" session. This reduces the number of energy-intensive server calls.
- Opt for Text: Avoid AI image/video generation unless essential. One AI image can consume as much energy as a full smartphone charge.
- Batch High-Impact Tasks: If you must generate images or large data analyses, do them in one session rather than intermittently throughout the day to allow server resources to spin down.
- Minimize "Idle" AI: Turn off automated AI meeting bots or summary tools when they are unnecessary for accessibility or record-keeping. Each automated summary adds to the campus carbon footprint.
- Apply Data Minimization: Only input the data needed for the task. Processing excess data increases the "compute load" and associated carbon footprint.
- Leverage Models Wisely: Test "small" or "efficient" models to determine whether they meet your needs for routine or simple text tasks before using more advanced models. Efficient models are often labeled with names like “Flash,” “Light,” “Fast”, and “Micro” while advanced models are labeled with names like “Ultra,” “Thinking,” or “Pro”.
- Use Berkeley-Licensed AI Tools: In addition to meeting security requirements, licensed tools are procured with consideration of "Green Spend" targets required by policy.
- Prevent E-Waste: Do not upgrade hardware (laptops/servers) solely for "AI-ready" features unless the existing equipment is at the end of its functional life, supporting UC's Zero Waste goals. When possible, update hardware elements rather than discarding the entire device.
