I hear many people say it's unrealistic to expect all creators behind training data to receive micropayments whenever a generative AI model trained on their work is used. This strikes me as being a very weak argument.
First, it's a straw man: there are many different business models for training data licensing. To name a few: upfront payment, payment upon training, equal revenue share, weighted revenue share based on impact on output, equity-for-data. Making out there's just one proposed solution is intentional misdirection.
Second, it suggests it's not already happening. This is false. There are already a number of gen AI companies who take exactly this approach.
Third, it suggests that payments to a large number of creative contributors is an impossible problem to solve. It's not: there are existing mechanisms in place that do precisely this outside of generative AI.
And fourth, it often comes as part of a suggestion that the responsibility should be on the creative industries to work out how to license their content to gen AI companies. But this has it backwards. It is not incumbent on creators to create business models that work for the gen AI industry. If gen AI companies need training data, it is their job to propose licensing structures that make it worth creators' while.