I think people already do something like this, using facial expressions and body language.

Certainly. Those mechanisms weren't working terribly reliably in a conversation that involved 60 people, which is precisely why I'd been looking for ways to augment the normal mechanisms.