well it not really a good analogy because pigs are very robust and have existed a long time and 'AI'(LLMs) are very brittle fragile embryos that can die from a sneeze.

a more relevant analogy for llms is a bullshiter that don't know shit about anything but like talk about everything and they're good at that (talking like they know), I'm sure we all at least know or knew in the past someone like that.

at a low enough percentage these bullshiters can thrive and get praise from other people but when there a lot of them and not enough of the real deal, well everyone is in trouble including the bullshiters since they no longer have who to mimic.

The actual question seems to be if humans are more resilient to errors in their training data. I would assume so.

As a relative statement, I'm not sure if AI or humans are better or worse (or even "it depends on what you're measuring"), but in an absolute sense humans are also pretty suggestible and easy to influence, and it's only countervailing influences pushing in other directions that turn it into noise.

This space is where the Venn diagram of marketing and genocides overlap.