How much bias is actually built into AI tools?

Hey everyone,\

I’ve been using AI more for research, writing, and general decision support, and one thing I keep wondering is how much bias is already baked into these systems before we even touch them.

I’m not just talking about political bias either. I mean bias in which sources are trusted, which perspectives are prioritised, how certain groups are described, what gets filtered out, and even how “confidence” is presented.

A few times now, I’ve asked AI about the same topic in different ways and noticed the tone or direction of the answer shift a lot depending on wording. That made me wonder whether the model is actually being objective, or just very good at sounding objective.

For people who understand this stuff better:

- What kinds of bias are most common in modern AI?

- Is the bias mostly due to the training data, fine-tuning, moderation layers, or all of the above?

- Have you seen examples where bias materially changed an outcome?

- And how do you personally fact-check AI when the answer sounds polished but may still be skewed?

Interested in both technical and real-world examples. Feels like this is one of the biggest issues in AI, but also one of the hardest to clearly see.

Thanks, Cathy.