AI for Good
Responsible AI in everyday decisions.
Focus on the people, systems and institutions affected by AI—and make ownership, safety and accountability visible.
Responsible AI Principles
Leadership questions that make responsibility concrete.
Sustainable AI
Four practical guides for lower-impact, better-governed AI.
Responsible AI data-center procurement
Assess vendors on facility-level transparency, environmental performance, community engagement and AI governance.
Read the procurement guide → Model choiceThe environmental impact of LLMs vs. SLMs
Compare models by successful work, utilization, energy and water—and consider hybrid routing to the smallest effective model.
Compare LLMs and SLMs → OperationsReduce AI’s environmental impact
Upskill users, avoid unnecessary AI, right-size models, meter consumption and make sustainability part of procurement.
Explore six practical measures → ReportingStrengthen ESG reporting with AI
Use AI to collect and clean fragmented data, detect inconsistencies and support compliant disclosures with human verification.
Explore the reporting guide →How Responsible AI Works
Practical questions for institutions.
What turns responsible-AI principles into practice?
Name the owner, classify the consequence, define acceptable evidence and build review into the workflow. Policies then need data controls, testing, documentation, monitoring, escalation and a route to pause or retire the system.
Should every AI system receive the same level of governance?
No. Controls should be proportionate to consequence, reach and reversibility. A simple classification process directs scarce review effort to the systems that can do the most harm.
What makes human review meaningful?
The reviewer needs time, authority, relevant information and a realistic ability to disagree with the system. High-impact uses also need a documented route for escalation and appeal.
How does sustainability enter AI governance?
Through procurement, architecture and model choice. Teams should consider energy, water, hardware, data-center location and expected utilization alongside accuracy and cost.
How can AI improve inclusion without creating new barriers?
Involve affected people early, test across languages and accessibility needs, and examine who is missing from the data and design process.
What should buyers ask an AI vendor?
Ask how the system is evaluated, what data it uses, where data and outputs travel, how incidents are reported and how the customer can exit.
Who should be involved beyond the technology team?
Business owners, legal, policy, risk, data, technology, HR, procurement and representatives of affected groups all reveal different consequences and responsibilities.
What should happen when AI affects a person’s rights or opportunity?
Provide notice, preserve the basis for the decision, offer meaningful human review and create a route to challenge or correct the outcome.