It's hard to convert this to technical requirement -- this is largely a moral requirement on someone running the service. That's the second best option -- first would be this being technicall not possible.

The only way I can think of to deter this is cost per action. So accounts start with some virtual (or real) currency which spends on actions like connecting or adding to the graph. This will at-least minimize if not stop the behavior.

Any other ways of implementing non-creepiness?

I don't think a basic social network starts out creepy, and quite a lot of effort goes in to implementing the features that enable it to become creepy. Non-creepiness is not a requirement, it is a constraint, such as 'no retroactive default boundary-weakening' and 'do not send email that dishonestly misrepresents the intentions and views of your members, and particularly not of specific members by name.' It is not that hard.

> It's hard to convert this to technical requirement -- this is largely a moral requirement on someone running the service. That's the second best option -- first would be this being technicall not possible.

It's hard to imagine this idea getting traction, but one way to implement this would be to require new features to be approved by (say) a majority of active users. This could act as a brake on roll-out of creepy features (if people read before approving …), but it might also slow or retard genuine advancements (although "no to constant interface re-design" is a feature in my book).