This does not preclude the possibility of mutual growth, though. A system interested in long-term stability will optimize minimally for short-term advertiser revenue, and primarily for long-term attention provider satisfaction.
One counterfactual of the current model that can be instructive is the following. Assume a system whereby all communication is happenstance -- there is no advertising or selective filtration (which is what advanced advertising becomes). Such a system degrades in much the way twitter does -- manual (attentive) classification and differentiation are not used universally enough, people don't sufficiently filter their own attention sinks, and the attention eventually dries up in the vacuum of entropy.
It would certainly be possible to move to such a manually filtered system, but it would require a massive reconstruction of social norms. The best system would be one that could passively or subconsciously negotiate filtration strategies in a decentralized runtime environment. It will need to be spatially collocated with the conscious being, eventually, or the latency will be unacceptable.
Eventually, then, we'll find that such filtration will actively seek out opportunities for novelty or for the satisfaction of latent (either unrecognized or frustrated) desires, and surface them to attention. This is precisely what advertising becomes.
Finally, you are right that we're approaching the equilibrium from an oblique angle, but it is a direction from which we can make progress at every step of the way, and the endpoint is ultimately a fairly ideal state.