Lets say IP law is extended or reinterpreted to include that machine learning from book or papers or articles or comments requires a special license, even assuming the memorization problem is solved. This is what anti ai seems to be arguing.

This would then result in some kind of “deal”. Producers of AI models are required to pay some kind of overall license fee or percentage into some kind of public fund. Even Bernie Sanders suggested something like that. The problems I see:

1. The AI corporations can afford this and it will not really impact them in any way. The prices for AI rise a little. Free access is reduced.
2. Open Weight models may no longer be used freely. You could still pirate a Chinese one but while capitalists have access to any potential benefits in replacing labor with AI, for ordinary people it becomes an additional form of rent.
   Also in combination with advances in robotics, these AI models could do a lot to allow people to become “independent” by just telling your $6000 robot (actual price for a humanoid robot in china today) to plant some potatoes and vegetables there there and there, then clean the house etc. Or DIY build your own out of plywood, servos and a smartphone once technology advances.
3. The government through that fund gains an active interest in protecting that source of incoming and increase AI use, even if it does replace workers.
   This happened with the tobacco funds when vaping came around. Governments had leveraged the future payouts of these funds with banks for short term payouts and would have been in big trouble if vaping actually reduced smoking significantly.