This becomes very obvious when you deal with large language models like Chat GPT and Grok. There is systematic bias in all of them.
The phrase “garbage in garbage out” comes to mind.
Large language models essentially represent the average of human beliefs. You can bias them one way or another by the nature of the questions you ask them.
Examples
Ghandi had Narcissistic Personality Disorder (NPD) - you have to work hard to overcome Grok’s belief that Ghandi should not be regarded this way