The majority of the users will use the big Ai´s from Google / Microsoft / Adobe etc. - only a minority will run their own open source LLM.

The majority could still download all movies from pirate bay - still they choose Netflix and pay for it.

I am pretty sure something like this will develop as well, the big content farms like NYT will demand compensation for using their content.

For me, the comparison with Netflix and Pirate Bay doesn't hold since it's obvious that Pirate Bay is infringing copyright.

This means that the following:

raises the question: Why not? Afaict, it is pretty hard to proof that an AI is just "rewriting" instead of just taking in information and producing "new content".

so basically, you would publish the content under a license which states that you can only reuse it if you have paid a ln invoice?

I don't want to discourage you, but that whole copyright issue is really complex and hard to enforce. In Europe, there were several attempts to force Google to pay publishers for the use of certain content and all of them have failed as far as I know.

https://github.com/indra-labs/indra is the very WIP repository, there is a draft white paper in there in the docs. The current state of the work is I have got all the protocol working via a simulated network, right now in the process of connecting it up to libp2p's DHT for sharing peer and hidden service information. In a few weeks time we may have a minimal testnet working, maybe even started on the interface to LND's payment subscription and wallet management APIs.

Details of how this would apply to monetising creator and publisher fee splits would be something else. I don't know much about the prisms, they essentially are a way where you pay into an address and it is automatically split to two or more other addresses. I know with Taproot it will be easy to do this on-chain but I'm more interested in how to make it so content creators can be automatically paid when their content is delivered out of the fees... This is what prisms do, on LN.

The content monetisation problem can be solved using a relay network because its normal operation is loading session balances with AMP payments and spending them via sending packets that route through them. It is just a small extra layer to interface between the session spends and dispatching a royalty fee in accordance a metadata format which specifies this.

It's definitely something I want to develop more once the main Indra network is running. I'm sure you can see how it becomes a primary mechanism of payment for data in general but specifically "intellectual property" in a performance kinda way. Easy to see how nice the whole thing can be for the user to not have to think about paywalls or subscriptions, just set their client to keep reliable relay sessions topped up and the fee rate on the service.

Game theory in Indra uses a minimal trust model, you pay ahead for traffic, but you can pay very small amounts, so loss is minimised. Once you have good relays set up you can move general traffic around, and the relays that deliver specific content types have fee rates for the data for the service, and the idea would be that for any given piece of content, again, somewhat trust-based, the relay is supposed to account the data volume for each rights claim and forward it as soon as practical to the LN address specified.

Such schemes are not really specifically within the scope of Indra, but charging for relaying traffic and charging for delivery of content can be the very same thing, and greatly reduce the cognitive burden, once you wrap your head around the idea of treating your indra session balances as credit for accessing content, and relating it to people's familiarity with prepaid balance account systems.