University of Sheffield: Generative AI Principles for Productivity
Source: University of Sheffield IT Services — Generative AI Principles for Productivity Published: Current Publisher: University of Sheffield
Overview
The University of Sheffield has established principles for the responsible and effective use of AI tools to enhance productivity across all administrative and professional services, including research support and teaching-related support tasks.
These principles reflect the University’s commitment to a positive but cautious approach to AI adoption, emphasizing that the use of GenAI is optional and supplementary, not mandatory. Staff retain the right to decline AI use if they have ethical concerns, believe human judgment is required, or prefer traditional methods. The principles are designed to enable appropriate, responsible, and ethical use while respecting professional judgment.
Core Principles
1. Human Oversight and Accountability
Guiding principle: AI tools are intended to compliment, not replace, human judgment and decision-making.
AI use cases suitable for support:
- Calendar scheduling
- Email and document drafting
- Meeting summaries
- Document summarization
Critical responsibility: Users remain ultimately responsible and accountable for the outcomes produced with AI assistance. This cannot be delegated.
2. Fairness and Bias Mitigation
Key awareness: Users must recognize potential biases in AI outputs and take steps to mitigate them.
Actions:
- Examine data used with AI tools for diversity and representation
- Ensure data is ethically sourced
- Be aware of systemic biases that may be reflected in AI training data and outputs
3. Transparency and Explainability
Expectation: Where appropriate, users should:
- Understand how AI tools generate their outputs
- Be able to explain the rationale behind decisions informed by AI
- Apply critical evaluation to AI-generated content
4. Privacy and Data Protection
Critical requirement: All use of AI must comply with GDPR and the University’s data protection policies.
Safeguards:
- Sensitive or confidential information must NOT be entered into public or unregulated AI models
- Users must exercise caution and ensure data is anonymised
- Personal identifiers must be removed where necessary
- Verify the tool’s data handling practices before use
5. Sustainability
Institutional commitment: The University is targeting a 30% reduction in emissions associated with IT equipment and services by 2030 (baseline 2018–19), as part of its Environmental Sustainability Strategy.
User responsibility: Use GenAI consciously and only where it adds clear value. The evidence base on carbon footprint is still evolving and varies significantly by task, tool, and usage pattern.
Accuracy and Verification
Fact-Checking and Validation
Critical rule: Outputs generated by AI tools should never be accepted without critical evaluation.
Priority areas for verification:
- Factual claims
- Research findings
- Data interpretation
- Official communications
Contextual Understanding
AI tools may lack nuanced understanding of organizational or domain-specific context. Users must apply their professional expertise to ensure AI-generated content is:
- Appropriate for the specific situation
- Relevant to organizational context
- Aligned with institutional policies and tone
Avoiding Misinformation and Disinformation
User responsibility: Ensure that AI-generated content does not contribute to the spread of misinformation or disinformation.
Learning and Development
The University encourages staff to:
- Stay informed about advancements in AI technologies and best practices
- Develop new competencies using AI tools
- Use AI to enhance existing skills and foster innovation and efficiency
Training available: The University provides resources and training to help staff understand and effectively utilize AI tools for productivity. Professional services staff currently have access to:
- Gemini for Professional Services via MyDevelopment
Data Protection and Safe Use of AI Tools
Data Security
Users must adhere to the University’s IT security policies when using AI tools, especially concerning the handling of University data.
Software and Tool Approval
Institutional AI tool: Google Gemini is the University’s institutionally-supported GenAI tool for productivity work.
Access: All staff have access to Google Gemini via the Student hub (requires Student hub login credentials).
Use: Gemini should be used to support productivity activities.
Other tools: Where tools other than Google Gemini are required for specific use cases, they must be requested through the New IT Solution Request Process (available via Staff hub). This ensures compliance with data protection and information security policies before approval.
Intellectual Property and Copyright
Considerations:
- Users must respect intellectual property rights and copyright laws
- Content generated or analysed by AI must not infringe on existing rights
- Users should acknowledge that there are potential copyright implications in the use of GenAI tools
- This area is currently partially understood and contested — staff should stay informed as legal positions develop
- Be aware of the risk that AI may reuse original content without acknowledging its creators
Implications for this project
The productivity principles provide context for how tools in the catalogue should be positioned for staff and administrative support (as distinct from research-facing tools):
| Principle | Catalogue relevance |
|---|---|
| Human oversight | Tools flagged with human-oversight-required status; emphasis on augmentation use cases over automation |
| Fairness and bias | bias-risk-notes field carries weight for tools used in decision-making or resource allocation |
| Transparency | output-explainability important for tools used in official communications or policy decisions |
| Privacy and data protection | gdpr-compliant, data-training-policy essential for tools handling institutional data |
| Sustainability | Carbon footprint and model training efficiency noted for tools in high-volume use (email, scheduling) |
| Tool approval | Google Gemini is the default approved tool; all other tools must be vetted via IT approval process |
| Copyright | copyright-implications flag recommended in tool metadata as the legal landscape continues to evolve |
Key Takeaway for Tool Procurement
The productivity principles establish a permissive but governed approach: GenAI use is encouraged where it adds clear value, but all tools must be vetted for data protection and information security compliance. This positions Google Gemini as the low-friction entry point for staff experimentation, whilst maintaining a formal approval gate for other tools.
Related notes
- cmi-strategy-2026 — the University’s wider Centre for Machine Intelligence strategy these productivity principles sit within
- uos-genai-principles-research — the research-specific counterpart to these productivity principles