It won’t be that simple but it’ll have some nasty effects: for example, media sites which are funded to support political viewpoints with a lot of money behind

It won’t be that simple but it’ll have some nasty effects: for example, media sites which are funded to support political viewpoints with a lot of money behind them are less likely to have paywalls because they can rely on sponsorship. That’s likely to make the AI summaries even less trustworthy because training will only be so successful at correcting for the selection bias of training data which is freely available or cheaply licensed. Most people know to some extent that this needs to be controlled for on really big issues like climate change but there are tons of smaller issues where someone working on an ML system won’t have any idea of the scale of the problem because they don’t know about issues which are largely only known within a particular regional, business, linguistic, etc. community.