"Know for a fact that there's no machine observer"

Using a hashtag, for a lot of people, isn't a deliberate indicator to effect a specific machine state at present time; it's an indicator that machine action may take place, not that it necessarily will. So, for example, people have been using non-functional hashtags on Facebook for a long time. Why would they do such a thing? They do so because the utility of the hashtag is dependent on the hashtag already being in widespread use. If you use it now, it may become useful later.

So it's not about saying "this text is definitely observed" or "this text is definitely not observed", it's about enabling the possibility under observation, either present observation or future observation.

Anyway, I think one of the articles linked to in your article (thanks for the link, btw), kinda misses the point:

    “Sarah Palin for President??!? #Iwouldratherhaveamoose”

    This usage totally subverts the original purpose of the hashtag, since the likelihood of anyone searching the term “Iwouldratherhaveamoose” is next to zero.

I would say the likelihood of anyone searching the term #Iwouldratherhaveamoose is not next to zero; it's probably closer to one. Firstly, the speaker. The speaker may have no prior knowledge of the hashtag, but having thought of it, knows that someone else had thought of it. After sending that tweet, what are the odds that the speaker's very next action will be to search #Iwouldratherhaveamoose, to see who else had thought of that tag and what they were doing. Secondly, the (human) audience. Part of this is to encourage someone else to make a new statement ending in #Iwouldratherhaveamoose; so the original speaker is incentivized to search this term sometime again in the future to see if their action has spawned a reaction.