My understanding is that these are planning tasks which are required to make a future technical decision and precise scheduling.
For example, we might have a future user story: Implement an auto-upgrade service for our endpoint client. To do that, we need to have a design and technology choice in place, thus we might have a knowledge acquisition task called: Prepare high-level architecture for auto-upgrade. If we do not have existing auto-upgrade experience in the team, we might have even an earlier task called Research and summarise existing auto-upgrade techniques. The idea is that each task in the path allows us to have proper and reliable estimates for the next one.
Some PBI to get implemented need additional decisions and steps made before. During sprints it is easy to focus only on incoming sprint, which may be to short horizon for proper planning and decision making.