An interesting article on Nature.com about how AI Models collapse when trained on recursively generated data.
Link: https://www.nature.com/articles/s41586-024-07566-y
Now, reassuring to us that they still need a diversity of inputs, hence the "I" part of "AI" is not yet complete. In human terms think of it as the conspiracy oriented wing-nut who gets his/her news from a single source and begins to parrot the most God-Awful theories, and even (god forbid) create his own.
What would be interesting is to create robots - in the likes of Boston Dynamics, have them "experience" the world - touch, sight, temperature, sound, taste, etc - and then see what they can tell us - so much of our own perception is based upon our own personal, cultural and even biological biases. What truths underpin the universe that are apparent when we strip those aside? And then monitor their interactions with each other and see how they learn/assimilate other robots "truths".
Reading between the lines there are some curious experiments being done, I'm sure, as we speak.