Computers have to follow what is being said by whom, the context of the conversation and often some real world facts to understand cultural references.
We on El Reg might aspire to all that, but many humans can keep it up consistently?
Cultural references? I'll get some of them (e.g. Pratchett); others (like things off the telly) are likely to go straight over my head. You (other commentard) will share some but not all of mine.
Real world facts? Fake news! Big grey areas there.
Context? Well, shouldn't the article have put AI expectations into the context of how real-life human intelligence looks? Sorry, weak example, I'm only human.
what is being said by whom? Misattribution is so mainstream it has a whole raft of sub-classes: honest confusion, the generic "they say", deliberate misrepresentation and spin, the strawman, the "Melanie Phillips translation", etc.