There are two possibilities for the recommendation system:
- It could be part of the blockchain logic. The Steem engine needs to compute a lot of stuff, recommendation could be part of it. Then the cost of running it would need to be covered by witnesses and all the people who are running full nodes. The benefit would be that the it would be paid for in block rewards, so for average user it would seem that it is for free.
or
- It would be an independent 3rd party system. In my opinion it has all the advantages and 1 big disadvantage.
+ the recommendation logic is completely separate from blockchain logic and is very different (computer security, network communication, economics vs. statistics, data analysis, machine learning, deep learning). Why should steem developers be doing it instead of doing what they know the best?
+ recommendation system is a very complicated thing and will need to be continuously improved. Every time an improvement is made there would have to be a hardfork.
+ the computational demands of such a system might be huge. The market needs to say whether the value brought is worth the costs, or not. If we tie it to the blockchain it could drag down Steem as a whole in case it is uneconomical. When separate, the system provider could adjust the service based on how people value it. In case it is uneconomical, it will just stop the service. Maybe someone else will find a cheaper / smarter approach.
- the disadvantage is, that currently there is no business model for developing and running such a system. Sharing revenue from curation rewards is one way to fix this. (Although not a perfect one - people could still receive the recommendation and then upvote it through other means)