Bias
how you understand bias and avoid reproducing it
AI tools learn from data that carries human bias, and they can repeat it. This card is about understanding where bias enters and how it could shape coverage, sources, or framing. A team pauses here because skewed output can quietly distort the journalism.
Questions to explore
- Where could bias in an AI tool show up in your reporting or framing?
- Whose perspectives might be underrepresented in the data behind the tools you use?
- How would you notice if an AI tool was steering coverage in a skewed direction?
- What kinds of stories or communities are most at risk from biased output?
- How do you check AI suggestions against your own judgment and sourcing?
Expert voices
“Training data carries human and structural bias, and that can lead to algorithmic discrimination.”
“Most AI systems are trained primarily on English-language, Western content. That has real implications for journalism in the Global South and non-English speaking communities.”
“An AI model can check whether a text uses narratives that reinforce stereotypes, like describing a femicide as a crime of passion.”
“AI systems reproduce the prejudices, inequalities, and stereotypes embedded in their training data, and sometimes those imposed by the companies that develop them.”
Things to consider
- A tool can carry bias even when its output looks neutral.
- Bias is easiest to miss when the output matches what you already expected.
- Human judgment is still the main check against skewed AI output.
Pull Bias when it is relevant and set it aside when it is not. Pair it with the other AI Conversations cards, lay them out on a table, and use the questions above to get everyone on the same page. Capture what you discuss on sticky notes or in a shared doc.
More AI Conversations cards
