How to detect and manage bias in AI
Using AI responsibly requires more than acknowledging that biases exist: it is necessary to learn how to identify them, understand their consequences within the university, and apply concrete strategies to reduce their impact on teaching, learning, and research.
SIMPLE STRATEGIES TO IDENTIFY BIAS IN AI RESPONSES
NOTICE whether examples repeatedly represent the same profile—gender, culture, age, language, or socioeconomic background.
REQUEST the same answer from different perspectives, for example:
“Rephrase this explanation incorporating cultural and gender diversity.”
CHECK whether the AI assumes non-universal conditions such as access to technology, type of family, mobility, or economic level.
LOOK FOR stereotypes in professional, social, or academic roles, even when not explicitly mentioned.
COMPARE several AI responses to see whether patterns repeat, which can signal the presence of bias.
RISKS OF BIAS IN THE UNIVERSITY CONTEXT
TEACHING
Explanations may be partial, examples may lack diversity, and representations may be stereotyped, reducing the quality of educational materials.
Explanations may be partial, examples may lack diversity, and representations may be stereotyped, reducing the quality of educational materials.
LEARNING
Students may not feel represented in the examples, hindering comprehension and decreasing inclusivity.
Students may not feel represented in the examples, hindering comprehension and decreasing inclusivity.
RESEARCH
If model biases are not detected, conclusions may be unreliable or may reproduce existing inequalities.
If model biases are not detected, conclusions may be unreliable or may reproduce existing inequalities.
EQUITY
There is a risk of reinforcing inequalities linked to gender, origin, language, socioeconomic status, or cultural background.
There is a risk of reinforcing inequalities linked to gender, origin, language, socioeconomic status, or cultural background.
HOW TO REDUCE THE IMPACT OF BIAS
CRITICALLY REVIEW EVERY RESPONSE
Remember that AI should never be used as the sole source of information.
Remember that AI should never be used as the sole source of information.
REQUEST MULTIPLE VERSIONS OF THE SAME ANSWER
Do you compare versions to identify discrepancies and potential implicit biases?
Do you compare versions to identify discrepancies and potential implicit biases?
ADD INCLUSIVE CONTEXT TO THE PROMPT
Do you include instructions such as “Include cultural, linguistic, and gender diversity”?
Do you include instructions such as “Include cultural, linguistic, and gender diversity”?
EXPLICITLY ASK TO AVOID STEREOTYPES
Do you tell the AI: “Generate the response without relying on stereotyped roles or associations”?
Do you tell the AI: “Generate the response without relying on stereotyped roles or associations”?
CROSS-CHECK INFORMATION WITH RELIABLE SOURCES
Do you verify content using reliable sources, especially in research or evaluable academic activities?
Do you verify content using reliable sources, especially in research or evaluable academic activities?
ACKNOWLEDGE THE MODEL’S LIMITATIONS
Do you keep in mind that no AI can guarantee total impartiality and that critical judgement remains essential?
Do you keep in mind that no AI can guarantee total impartiality and that critical judgement remains essential?
AI reflects both available knowledge and its inherent biases. Responsible use consists of recognising these limitations, contextualising each response, and maintaining a constant critical review.