University-Wide AI Guidance

In principle…

RESPONSIBILITY and ACCOUNTABILITY - Human judgment, decision-making, and values should be prioritized to ensure responsible choices about the adoption and use of AI tools. The individuals who use AI to accomplish tasks are ultimately responsible for all of its actions, outputs, and impacts.

In practice…

• Use AI only in appropriate use cases; consider cost, resource consumption and institutional priorities as part of the decision to use AI
• Develop AI tools carefully and judiciously, respecting human values, rights, and dignity
• Avoid the use of explicitly prohibited AI tools
• Use AI to support your work, not to make decisions on your behalf
• Understand what actions an AI agent or workflow can take, what data it can access, and related monitoring and controls
• Review and validate AI-generated output before acting on it; rely only on AI outputs you can understand, verify, endorse, and defend
• Maintain meaningful human involvement in decisions, actions, and communications that could affect people, university operations, research, academic outcomes, or compliance obligations
• Subject AI use to continuous assessment and improvement; recognize and respond to the rapid advancements and challenges of AI technology

In principle…

TRANSPARENCY and INTEGRITY -Transparency and integrity in AI use play a critical role in building trust, improving outcomes, and identifying bias and errors. They also promote accountability and help people appropriately use and evaluate AI-supported work.

In practice…

• Disclose the extent to which work is aided by AI; be honest about whose judgment, position, or conclusion is represented
• Strive to understand the data and methodologies employed by AI tools 
• Verify facts, citations, references, and other information sourced and produced by AI tools
• Recognize that AI tools may generate inaccurate, misleading, or fabricated information and should not be considered authoritative sources
• Strive to establish AI collaborations or partnerships (vendors, other universities, etc.) only with organizations whose AI values align with UVM’s

In principle…

ACCESSIBILITY, INCLUSIVITY, and BIAS - The output of AI tools may be biased and value-laden. The values represented in our AI use should align with the values that we are committed to holding, both as individuals and as a community. AI should be used in ways that promote access, fairness, dignity, and equity for all members of the UVM community.

In practice…

• Verify that the values represented in AI systems work align with the values that we hold as individuals and as a community
• Avoid the use of AI tools in which we cannot easily identify the values or potential biases in the output
• Evaluate AI output and actions for potential bias and exclusionary or discriminatory content
• Ensure AI generated output and content meet university accessibility standards and requirements
• Avoid relying solely on AI output and actions in situations that may impact opportunities, service, or outcomes for individuals or groups
• Use AI to enhance access to university resources; recognize that community members have unequal access to AI tools
• Ensure inclusiveness, accessibility and recognize diverse needs and abilities when selecting and designing AI tools and experiences

In principle…

SAFETY and COMPLIANCE - AI use should reflect the highest regard for and conformity with the guiding principles and our shared community values. AI use should prioritize safety and security, and should protect people, data, research, and university resources. 

In practice…

• Observe existing University policies, procedures, standards and guidelines related to privacy, information security, data governance, data stewardship, data compliance, records management, accessibility, procurement, research compliance, and other applicable requirements, which continue to apply when AI tools are used
• Ensure appropriate human responsibility and review, especially for high-impact uses
• Use only institutionally approved and supported AI tools; ensure that UVM’s instance of the AI tool is being used
• Ensure compliance with data privacy and information security policies prior to using AI tools
• Follow university policies, data governance, contractual obligations, and applicable laws when using AI tools
• Consider downstream uses and disclosures and ensure that these also comply with applicable laws and university policies
• Evaluate AI systems that can take actions autonomously for appropriate safeguards, monitoring, and escalation paths before deployment
• Avoid deploying AI systems that can make irreversible changes without appropriate logging, backup, monitoring, or recovery mechanisms

Before using an AI tool for a given task or workflow, consider these questions:

• Is AI the appropriate tool for this task?
• Am I prepared to be held accountable for the AI generated outcome?
• Do I understand the methodologies the system will employ?
• Can I verify the accuracy and quality of the output?
• Am I maintaining meaningful human oversight?
• Will the AI system have access to sensitive, confidential, or regulated information?
• Do I have the right and am I authorized to input the data or information into the AI system and is this use of data within the scope of my job?
• If I am conducting human subject research, has the IRB approved my use of AI tools?
• Will this AI system influence decisions that affect people?
• Does this use require disclosure, approval, or additional review?
• Does the environmental or operational cost justify the benefit?

AI tools are costly to develop and deploy; these costs can be financial, social, and environmental. AI systems rely on extensive computing infrastructure that consumes energy, water, land, and other resources, often disproportionately affecting communities that are geographically distant from those who benefit from AI tools. As a university committed to sustainability, planetary health, and the common good, UVM will strive to be a leader in the responsible adoption of AI. We will be attentive to the broader social structures and impacts of the technology that we develop and use. We will evaluate AI not only by what it enables us to accomplish, but by how responsibly it uses resources, how equitably its impacts are distributed, and how well it advances the well-being of people, communities, and the environment.