AP: Collaborative decision-making needed for responsible use of Generative AI

Generative Artificial Intelligence (AI) could become an integral part of society. The potential to contribute to economic and social well-being is considered significant by many. Generative AI is possible because models are trained with an overwhelming amount of data. In addition, these systems make use of personal data that people enter during use. According to AP, it is likely that many generative AI models have an unlawful origin.

Moving forward responsibly

Both the societal risks and opportunities of Generative AI are significant, bringing dilemmas for regulation and supervision. Fundamental rights and public values must therefore be the guiding principles. As a society, we can actively steer by setting the conditions for responsible Generative AI. This way, innovation opportunities and effective regulation can coincide.

For the upcoming period, the AP's vision is up for consultation. The vision is intended as a guiding perspective for what is necessary to adopt Generative AI responsibly as a society.

AP Chairman Aleid Wolfsen: "The possibilities of Generative AI continue to amaze. The technology can bring much good, but it can also cause great harm. Generative AI goes hand in hand with storing enormous amounts of data. It is essential that the protection of personal data is in control during the training and use of Generative AI and that as a society we have control over the technology to protect fundamental rights and public values. Given the rapid technological development, correcting in hindsight is not enough. With this vision, the AP shows that it is possible to develop and use Generative AI safely and responsibly."

Future scenario: ‘Values at work’

The AP examines 4 future scenarios for 2030 in this vision. The AP looks at the possible development of technology and European regulation. The AP aims for the 'Values at work' scenario. To achieve this, the AP identifies several fundamental principles on a societal level. Among others, these are European digital autonomy, social resilience, democratic control, a well-functioning market, and the ability to correct throughout the AI development and deployment chain.

Additionally, in this scenario the AP sees fundamental principles at the level of individual AI models and systems. These include transparency, adequate risk management, the formulation of clear application objectives and control over the environment in which the AI system operates and the data is processed. By taking these measures, organizations could comply with relevant legislation, including the GDPR, the AI Act, and relevant sector-specific legislation.

General purpose AI

This vision concerns both the AI models that generate various types of output, as well as the systems and applications in which these models operate. Think of chatbots, image or video generators, or AI-enhancements within search engines. Common names for the underlying multi-use models are 'foundation models' or 'general-purpose AI models'. These type of models fall under the scope of this vision on generative AI.

GDPR preconditions for Generative AI

The AP sees both opportunities and challenges for the lawful development and use of generative AI. Alongside the vision for the desired future scenario, the AP publishes some preliminary preconditions for generative AI under the GDPR. Organizations are invited to provide feedback on this current version. The AP considers these responses while drafting the final version.

According to the AP, the development of most foundation models has been falling short in terms of lawfulness due to processing of (special) personal data in training data and the way this data was collected. The European Data Protection Board (EDPB) has stated that the downstream application of these foundation models by Dutch and European parties is not inherently unlawful.

Steps taken by the AP

The technological developments in the field of generative AI are proceeding at tremendous speed. Developers and organizations using generative AI must be aware of the risks for end-users. With this vision and GDPR preconditions, the AP provides more clarity for organizations.

In addition to these consultation documents, the AP will take concrete steps in the coming period to work towards the preferred scenario 'Values at work'. On the European level, the AP contributes to setting the norms of the technology.

The AP is also starting an input desk for generative AI, where developers and users can share questions and areas of attention with the AP. This helps the AP keep track of the most important challenges that organizations are facing. Additionally, the AP will seek out different perspectives on the vision by attending and organising events in the coming period.