I honestly don't think neural networks ingesting terabytes of data are the future of AI. In principle, these are all engineering problems that can be solved by

I honestly don't think neural networks ingesting terabytes of data are the future of AI. In principle, these are all engineering problems that can be solved by advances in few-shot learning. Once you can train a model for a task with only a couple examples, the ethics/legality of scraping IP become irrelevant. I feel like if model weights are made to be public domain that would make sense. The only reason it costs any money to use them is because of infrastructure, so any argument to recoup via licensing fees for the weights themselves doesn't make sense. You can charge people to use your infrastructure, but not for what could essentially be their own work. As for students and cheating - unnerving as it is, I think we have to accept that there's a sort of linguistic calculator in their toolboxes now. Whatever the solution ends up being, I don't hope that it's restrictive. AI, just like any other technology, can be integrated into our lives in a beneficial way.