Bias
AI bias
Bias in AI models
AI bias
Bias in AI models
Written by Martin Ågren - May 27, 2026
AI bias is:
Consistent distortions/preferences/prejudices in an AI model’s output.
Different AI models can have different types of bias (be biased in different ways) depending on which country/organization/person is behind the AI model.
Keep in mind that AI bias does NOT only apply to text from language models. AI bias can exist in both language models and image models (and other types of AI models).
Cause of AI bias: Training data (that an AI model has been trained on) can contain certain dominant perspectives or prejudices. Consider this: The training data typically used for LLMs is dominated by text in English. This can lead to certain perspectives or opinions becoming dominant in the training data, which can then be reflected in the AI model’s output.
Maintain a critical mindset toward output from AI. Ask yourself whether the output may be biased as a consequence of possible AI bias. Are enough different perspectives represented in the output? Also reflect on who is behind the AI model, meaning which country/company/person, and what impact that could have on the output from the AI model. You can also compare output between different AI tools/AI models to try to get more nuanced output.
You need to be aware that AI can produce biased responses that affect the organization’s work.
Encourage healthy critical thinking regarding AI output and emphasize that AI is a support tool, not an objective truth. When making important decisions, AI responses should always be supplemented with human judgment and other sources/AI models.
For your AI policy: Clarify that AI output should be reviewed with a focus on AI bias. Specify in which situations this type of review is especially important.