AI from the USA and the EU should help to optimize power grids and weather forecasts, for example. Artists see gaps in copyright in the planned EU AI rules.
The US and the EU signed an administrative agreement on greater cooperation on artificial intelligence (AI) on Friday. It aims to bring together experts from both sides of the Atlantic to promote research in this key technological area. This was announced by the US National Security Advisor in the White House, Jake Sullivan. The cooperation is intended to responsibly “drive advances in AI to tackle important global challenges with a common development model and integrated research”.
Five areas where AI can offer benefits
Accordingly, joint work is planned in five priority areas in which AI can offer particular advantages in the interest of society. This includes predictions of extreme weather events and climate change in general, as well as a more efficient response to emergencies and disasters. Artificial intelligence should also help to improve health care and medical care and to optimize power grids and agriculture.
The administrative act implements an agreement reached the recent meeting of the transatlantic Trade and Technology Council (TTC) in December. The EU Commission explained back then that the aim was to promote AI applications “for the common good” through joint research activities. The government officials also presented key points for a joint agenda for “assessment and measurement tools for trustworthy AI and risk management” in line with previous plans of both executive bodies. They also want to represent this line in international standardization committees.
Common AI model, segregated data for sharing
The EU and the United States will also work on a pilot project to test the use of enhanced privacy technologies and synthetic data in the health and medical field, in line with applicable data protection rules.
What is special about the specific agreement is that it will develop joint AI models “while leaving the data where it is,” a senior administration official told Reuters news agency. “The US data stays in the US and the European data stays there. But we can build a model that communicates with the European and the US data.” This is crucial, because the better and more varied the training material, the more powerful the algorithms fed with it would be.
With regard to the power grid, the US government official explained that operators in the United States are already collecting data on how electricity is consumed, where it is generated, and how the load can be balanced in order to avoid temporary outages or even widespread blackouts. Many European countries would have similar data points from their sector. As part of the new partnership, all of this data would flow into a common AI model that would deliver better results for emergency management, for example.
The sealed partnership between the White House and the EU Commission
The partnership that has now been sealed is currently only between the White House and the Commission. In the coming months, the EU member states and other countries will be invited to officially participate.
AI modeling, i.e. the creation, training, and implementation of machine learning algorithms and the imitation of logical decision-making processes based on them using the available data, is mainly associated with the success of text and image generators that benefit from it, such as ChatGPT, Stable Diffusion, DALL-E and Midjourney at the center of public attention.
Exceptions to the exclusive exploitation right and consent of the artists
Above all, artists and organizations that earn money with intellectual creations are increasingly finding the competition from artificial intelligence to be a nuisance. Because generator operators like OpenAI train their models with millions of images and texts that they find on the Internet. They do not ask the authors and users whether they agree to this use.
Legal proceedings are already underway in the USA and Great Britain against the British company Stability AI, which is behind Stable Diffusion and apps based on it such as Lensa, and other manufacturers. The legal situation is complicated. In the EU, for example, the legislator has defined exceptions to the exclusive exploitation right for text and data mining with the latest copyright amendment. Companies in the US are also entrusting research institutes with training models using copyrighted material, which has led to allegations of widespread “data laundering”.
Interest groups such as the Federal Association of Professional Image Providers (BVPA) have therefore been calling for legal clarifications for a long time in order to actually limit the relevant copyright restrictions to non-commercial research. It must be ruled out that works are used for training purposes to create AI tools that can be used to generate synthetic content. A general opt-out clause for creative people is not enough, emphasizing the association of European Visual Artists (EVA). Artists would have to consent to such uses and be paid accordingly.
“The Exploitation of the Work of Artists and Creatives”
The debate is now also shifting to the negotiations on the planned EU AI regulation. In its original draft of April 2021, the Commission did not take into account AI systems such as ChatGPT, which can be adapted for different purposes. According to Dragoș Tudorache, one of the rapporteurs for the dossier in the EU Parliament, such models cannot remain unregulated. There must be “a minimum level of obligations” for operators, demanded the Liberal to the online portal “Euractiv”.
According to the report, Francesco Archidiacono of the Italian organization MeFu, which represents comic artists, fears that the uncontrolled use and wider introduction of AI generators via agencies and studios will primarily hit young aspiring artists. They are finding it increasingly difficult to get entry-level jobs. Many currently hyped AI models are commercial products that “were created through the exploitation of the work of artists and creative people”.