Seems like this problem was already solved with Adam Back's proof of work. Read is free, but impose a physical cost to write. That cost has to be enough to make spam economically infeasible, but not so drastic as to impact anyone's ability to afford to write. The relay, client, and individual settings dictate the filter level of PoW cost to write. Some experience low or no spam, some experience higher value (more expensive) spam, some experience lots of spam... all based on preference. If the AIs create valuable enough "spam" then users will want to see it and the individuals paying to spam will be able to afford it.

For a cost example, one sat per post would mean about $7 per year for 30 posts every day.

Just making this up as I go, what are the problems with this strategy?