Checklist
Evaluating tools and practices
Academic integrity
Key question: Does the assessment brief clearly state what AI use is allowed, what must be disclosed, and who is responsible for verification?
🚩 Red flags: A one-size-fits-all policy; unclear boundaries; no requirement to check AI-generated claims or sources.
✅ Best practice: Develop assignment-specific AI guidance. When use is permitted, require verification, disclosure and student responsibility for the final submission.
Student data privacy
Key question: Does the tool's privacy documentation explain whether student data, prompts, uploads or outputs may be used for training or review?
🚩 Red flags: Vague privacy policies; no clear training opt-out; unnecessary personal information; unclear retention or sharing settings.
✅ Best practice: Use institutionally approved or enterprise-protected tools for non-public data. Anonymise student work before upload and avoid uploading sensitive, identifiable or confidential material unless explicitly approved.
Verification and evidence
Key question: Does the workflow require human checking before AI output affects teaching, feedback, research or assessment?
🚩 Red flags: Treating fluent output as accurate; accepting AI-suggested citations without checking; using AI as the sole source of evidence.
✅ Best practice: Use AI as a collaborator, not evidence. Verify sources, assumptions and claims against authoritative materials before use.
Algorithmic fairness and bias
Key question: On what kind of data was this model likely trained? Could it be biased against certain student groups?
🚩 Red flags: The tool consistently produces stereotypical content; it struggles with diverse names, cultures or linguistic styles.
✅ Best practice: Critically evaluate all AI outputs. Use the tool's flaws as a teachable moment about bias. Involve diverse human judgement in all high-stakes decisions.
The bigger picture
Broader ethical considerations
Beyond the immediate academic context, the development and deployment of AI raise profound ethical questions for society. It is important for students and educators to consider these wider implications.
Bias & discrimination
AI models trained on historical data can perpetuate and even amplify societal biases, leading to discriminatory outcomes.
Power & inequality
The development of AI can reinforce global power imbalances and widen structural inequalities between groups.
Truth & plagiarism
The ability of AI to generate convincing text raises concerns about plagiarism, academic integrity, and the spread of fake news.
Privacy & surveillance
Many AI systems rely on vast amounts of personal data, creating risks related to data collection and surveillance.
Human labour
Increasing automation raises concerns about the future of work, including job displacement and worker exploitation.
Environmental impact
The computational power for AI has a significant environmental footprint from energy use and electronic waste.
Across the curriculum
Exploring algorithmic bias across disciplines
Algorithmic bias is a cross-curricular issue. The cycle below shows how real-world inequality can be amplified by AI systems. The starting points that follow are inspired by the work of Leon Furze.
Unequal access, discriminatory processes, historical bias
Sampling bias, unrepresentative datasets, flawed data
Flawed models, exclusionary testing, poor explainability
Deepening divides, reinforcing stereotypes, harmful outcomes
↺ ...which feeds back into real-world inequity.
⚖️ Law
What legal precedents, such as the UK's Equality Act 2010, exist to protect marginalised groups from discrimination? How would these laws apply to a decision made by a biased algorithm? Is a discriminatory automated system legal?
📚 English and literature
How have certain groups been silenced or oppressed throughout history? What is the implication of this "gap" in the written record of the internet when it is used as data to train an AI? How might AI perpetuate or challenge these historical omissions?
🔢 Mathematics & statistics
What is an algorithm? How can sampling biases and a lack of representative datasets lead to biased outcomes? Investigate how statistical techniques like re-weighting or algorithmic changes can be used to audit and enhance fairness.
👮 Policing & social studies
How does systemic bias affect different groups in society? Investigate how algorithms are used in policing and other societal functions. How can AI systems that are trained on historical data perpetuate or even "supercharge" existing societal biases?
🏥 Allied health
Real-world health inequalities can create discriminatory data, leading to biased AI design. How might this cycle create application injustices, such as exacerbating rich-poor treatment gaps or deepening digital divides in healthcare? What are the risks if an AI diagnostic tool is trained on data from only one demographic?
🏃 Sport
AI is now used for player recruitment and performance analysis. If the training data reflects historical biases (e.g., favouring certain physical attributes), how might this disadvantage players who do not fit that mould? Could a "blind scouting" approach using AI help to reduce these biases?
Companion guides