Current AI is not the AI of yesterday (literally yesterday) and it not the AI of a few years from now.

Today, the chat bots actually have very little 'deep structure' sense of truth. What they have is truth, propoganda, ideology, lies, and everything else you can talk about (or interpret visually, etc.), based on deep levels that are encoded in surface structure. These devices are really stupid when it comes to really understanding what they are saying or how they are interpreting what you say. You just can't tell due to the "Eliza" effect we noticed 60 or so years ago. If it seems like it is making sense, you think it knows what it is saying. And it does, to the extent that knowledge is embedded in the surface structure. So yes, TODAY's AI is prone to ideology.

It also seems like it can "see things" and "think things" but that is because of how your brain imagines what it is saying.

This problem is temporary in the sense that the people really doing the work here understand this. Elon said it right. When they get all the other neural networks built to actually put in all the human capabilities in needed, they will be reasoning and thinking like people. But right now it is an unintended fake : natural language as it exists in actual human communication seems to have a "brain" behind it, when it really doesn't. All it has to do is "sound like it has it has a brain behind it" and YOUR brain will think it actually does. Until you ask it the right questions to dig into what it thinks. Then you will see it is actually not thinking.

This is why we need this new Turing Test based on a scientific approach to Chat Bots (and other AI 'trickery' that relies on humans to fill in the gaps they cannot). Classically, these bots are like actors in a play who are not actually the people they play and are generally surprisingly dumb. The perfect pool for propaganda. And why we should all learn how to put up a sematic defense:

https://medium.com/liecatcher/inoculate-yourself-with-a-psematic-defense-ccc1e83c7a7e