I think two main things need to happen: increased transparency from AI companies, and limits on use of training data.

In regards to transparency, a lot of current AI companies hide information about how their models are designed, produced, weighted and use. This causes, in my opinion, many of the worst effects of current AI. Lack of transparency around training methods mean we don’t know how much power AI training uses. Lack of transparency in training data makes it easier for the companies to hide their piracy. Lack of transparency in weighting and use means that many of the big AI companies can abuse their position to push agendas, such as Elon Musk’s manipulation of Grok, and the CCP’s use of DeepSeek. Hell, if issues like these were more visible, its entirely possible AI companies wouldn’t have as much investment, and thus power as they do now.

In terms of limits on training data, I think a lot of the backlash to it is over-exaggerated. AI basically takes sources and averages them. While there is little creativity, the work is derivative and bland, not a direct copy. That said, if the works used for training were pirated, as many were, there obviously needs to be action taken. Similarly, there needs to be some way for artists to protect or sell their work. From my understanding, they technically have the legal means to do so, but as it stands, enforcement is effectively impossible and non-existant.