Can AI really be biased, and how does that happen?

Yes, AI can absolutely be biased. It's not that the AI has opinions. It's that it learns from data created by people, and that data is full of our own conscious and unconscious biases. Think of it like a chef learning to cook only from a library of 1950s cookbooks. The chef isn't biased against modern, healthy food. They just don't have any recipes for it. The result on the plate will be heavy, bland, and very specific to that era. AI works the same way. If a hiring tool is trained on a company's past 20 years of successful resumes, and that company historically hired mostly men for leadership roles, the AI will 'learn' that being a man is a positive signal for leadership. It might start downgrading resumes that mention women's colleges or have a name statistically associated with women. This isn't a hypothetical. Amazon scrapped an internal recruiting tool in 2018 after discovering it was penalizing resumes that included the word 'women's.' The tool didn't hate women. It just found a pattern in a biased dataset and ran with it. The problem goes deeper than just hiring. Image generators have been shown to produce pictures of CEOs that are overwhelmingly white and male. Language models can associate certain jobs with specific genders or races. The key insight here is that bias is a data problem, not a robot-sentience problem. Fixing it is hard because you have to first spot the bias in the training data, which is often millions of documents. Then you have to carefully correct it without breaking the AI's other useful knowledge. It's like trying to remove all the salt from an already-baked cake.