Filling the governance vacuum related to the use of information commons for AI training

More research is needed in this context to understand the relationship between research institutions and commercial entities, particularly in light of the excep -tions for non-commercial and research uses.

Specific AI system regulations, such as the European AI Act or the AI Bill of Rights in the United States (both of which have yet to be adopted), may include rules that govern the development and management of AI training datasets. Other laws are also applicable. The Digital Services Act, recently adopted in the European Union, requires platforms t... See more

dealing with bad actors and reducing harm is not the same as in -centivizing good actors to establish best practices and good dataset governance.

Licensing rules, technical measures, and regulations all interact in ways that alto -gether create a governance structure for AI training datasets. For example, stat -utory limitations on high-risk AI systems seek to impose behavioral constraints similar to those imposed by RAIL-type licenses

cases of obvious harm should be distinguished from those in which there is a sense of potential harm, or the issue is one of the unexpected uses

By 2014, there were almost 400 million CC-licensed photos on Flickr. That year, researchers from Yahoo Labs, Lawrence Livermore National Laboratory, Snapchat and In-Q-Tel used a quarter of all these photos to create YFCC100M, a dataset of 100 million photographs of people created for computer vision applications. Until today, this dataset remains o... See more

In one scenario, these conditions would be personalized by each entity sharing their works through tools that allow them to express motivations for sharing. This path is currently being explored by the new responsible AI licenses introduced in 2022, and their creators envision a proliferation of licensing conditions chosen by licensors 5

This awareness is well expressed by Creative Commons, in the organization’s 2020-2025 strategy is based on a vision of better sharing 1 : “In order to protect what we have achieved so far and to create the world we want to see, we must expand our focus beyond copyright licensing, because content sharing cannot be decoupled from economic or ethical ... See more

Another proposed solution would entail technical means of tracking uses of openly shared works, which could benefit from advances in web technologies. These tools would make downstream (re)uses of content traceable and, therefore, legible to licensors, creating opportunities to interact with users and potentially enforce additional norms

It is time to find ways to manage both the open resources and the AI solutions built on top of them in a way that is more sustain -able and reduces harm