Say Alice and Bob like to share media over a communication channel. Previously, I formalized this as a communication game between two predictive coders. Alice consumes media from and Bob from and then they send each other media from the distributions and and have the following objectives, The first term is prediction error for received media and the second term is prediction error for preferences. The basic takeaway is that both needs to be sufficiently cooperative (social) in order to avoid a situation where one of the players is sent media outside their support.
One seemingly technical detail of media sharing is that, in practice, Alice and Bob will need to model one another's preferences using a finite number of samples. For high-dimensional sample spaces, this is going to be problematic. However, if media can be clustered into categories then it would be possible to project the information gained about a person's preferences onto a larger set of unseen/unlabeled media.
Say there is a network of players where each edge represents a communication channel between two players. I will assume that each of these communication channels are independent and equally viable. Now Alice and Bob can share on one channel while Charlie and Dan share on another. Moreover, each player has a media consumption pattern. More specifically, they obtain media from somewhere. They need media in order to send things to their communication partners. In fact, the throughput of a particular player is bounded by the rate at which they can aquire media. This is a demand.
Say there is an additional player, Com, on the network. The player is special in that they are the frame or edges of the network. Their goal is to maximize the amount of media being shared on the network. However, since everyone is a predictive coder this is equivalent to making it easy for players to predict what other people like being sent. Such a player could be provided by a media supplier which would aggregate media being shared on the network. What does the aggregator get out of this? Data. The more data is being sent through the network the easier it is to predict preferences. This is no different than coopeartive objective of the players.
So how does Com achieve it's objective? Here's on approach. The larger the difference between Alice and Bob's preferences the harder it is for the player's to share media. However, if Com is watching to see which media Alice and Bob share with each other they can use this information to relabel all of the media being shared on the network in terms of whether or not Alice/Bob would like it. Using this information, whenever Alice sends media to Bob that he likes Com can use this information to supply Alice with predictions about more things to send Bob. Com receives feedback based on whether or not Bob responds/likes what is sent to him.
Com reduces the amount of cooperation required for communication because now Alice doesn't have to fully understand whether or not Bob will enjoy a particular piece of media. Instead she can rely on Com to give her hints about Bob's preferences. Alice and Com work together to do this. In the limit, Alice only has to predict that if she sends certain types of media through the Com channel she'll receive back media she enjoys. This works because the principle for Bob, and every other player on the network, is the same. Ultimately, Com catalyzes cooperative interaction over it's network. This could be a sense in which the network augments the media to a social media.