No ban planned: Creative Commons is working on licenses for AI training
Creative Commons wants to ensure that those responsible for data sets can signal what they expect in return when an AI model is trained with it.
(Image: Jirsak/Shutterstock.com)
Creative Commons, the organization responsible for the licenses of the same name, is working on labels to indicate the conditions under which data sets can be used for training AI models. This emerges from a blog entry in which the "CC Signals" are presented. The declared aim is to preserve the open internet by clearly regulating whether and how content can be used for AI training. Otherwise, there is a threat of a future in which data will simply continue to be used and those responsible will try to prevent this technically. The result would be a network of demarcated areas that would be protected by paywalls, for example.
Much criticism on the project page
As Creative Commons explains, four CC signals have been developed. All of them stipulate that an acknowledgement ("credit") of those responsible is due for the use of the data set, i.e. it is a kind of attribution. It should also be possible to stipulate that a financial contribution is due for use. Creative Commons provides for two forms. One is money for those responsible or money for the ecosystem to which it belongs. The amount should be determined in good faith. Finally, those responsible should also be able to stipulate that their data can exclusively be used for the training of AIs that are open. A complete ban on AI training is not planned.
Videos by heise
The aim of the project is to develop a global tool, just as the Creative Commons licenses have long since become. No new property rights should be created, but rather a kind of "manners for machines", writes the organization. It recognizes that we are dependent on AI companies adhering to this. Of course, one could be cynical and have doubts about this, but there are many reasons to expect the opposite. The signals will also be based on technical standards developed by the Internet Engineering Task Force. The project has its own page on GitHub, where the organization is asking for feedback. So far, there has been almost exclusive and sometimes harsh criticism.
(mho)
