Most U.S. Adults Think Gen AI Should Credit Artists, Writers
Most people think generative AI programs should give credit to sources they rely on to produce their outputs, a clear indicator of where consumer attitudes and alliances lie in the fair use debate surrounding AI training data.
Over half of U.S. adults say gen AI models that produce text and images, such as ChatGPT and Midjourney, need to acknowledge the sources they use in producing their outputs in response to user prompts, according to a recent Pew Research Center survey. Just 14% of respondents don’t believe gen AI programs need to offer this kind of credit.
Asked about specific examples of content that generative AI systems could produce, consistently answered that they believe AI programs should credit their original source material, including when they produce images in the visual style of a living artist (67%), write in the style of a living author (67%) and draft a script in the style of a popular movie (61%).
The data doesn’t reveal whether consumers also believe that creators and rights holders should receive compensation (not just credit) for their contributions. Compensation for training data through licensing or other payment structures usually follows in the debate.
The findings go to the heart of the fair use debate and pending lawsuits against AI developers alleging copyright infringement. One of the most consequential open questions facing copyright lawmakers, creators and content rights holders and gen AI tech providers is whether AI models should be allowed to train on copyrighted content without licensing, a situation that would create the need for a system to regulate permission, attribution and fair compensation for ingested material, as VIP+ has discussed.
At least in the U.S., instructive answers on copyright and AI training aren’t likely to come until later this year. Last week, the U.S. Copyright Office said it will release its report on generative AI in multiple sections expected to publish over the rest of fiscal 2024, giving its recommendations on digital replicas (spring), the copyrightability of AI-assisted works (summer) and finally training AI models on copyrighted works and its implications for licensing and liability.
Consumer attitudes won’t move the needle on how the fair use debate settles. But if consumers are allied with creators and rights holders, some may also have qualms with their own material, such as user posts on social platforms, including Reddit, being ingested into AI models without consent or compensation.
