Solidarity in the AI Era: The Ethical Dilemma of Civil Society

The project “Code of Conduct Democratic AI,” led by the net-policy association D64, together with nearly 20 civil society organizations such as the Arbeiterwohlfahrt, the Bundesarbeitsgemeinschaft für Seniorenorganisationen (BAGSO), and the FrauenComputerZentrumBerlin, published a white paper on “solidarity practice along the use of AI” on Monday. The creators aim to provide NGOs with a practical guide.

This is intended to help with the dilemma faced, particularly by NGOs: solidarity is considered a central value of societal action, but in the context of artificial intelligence (AI), this claim seems hardly fulfillable. A few global companies, especially from the USA and China, dominate the development of large AI models and digital infrastructure, leading to digital spaces characterized by unequal power relations. Consequently, common good-oriented organizations are in a dilemma: They are often forced to use commercial systems whose development conditions they cannot control – for example, regarding poor working conditions, high energy consumption, and monopoly structures.

The use of AI systems presents fundamental challenges that contradict the principle of solidarity, according to the paper. These include invisible, often exploitative digital labor for creating metadata and annotations, immense ecological costs due to energy consumption and resource extraction, the lack of participation of vulnerable groups in decision-making, and the displacement of the common good by supposed efficiency narratives. Furthermore, journalists, authors, or artists whose content OpenAI, Google, Anthropic & Co. use to train their AI models are neither adequately compensated nor recognized.

Conscious renunciation of ChatGPT & Co.

Therefore, according to the initiative, the goal must be to anchor solidarity as a conscious principle in the design, selection, and use of AI. Civil society can play its role as a critical user and designer to open up concrete areas of action through conscious decisions. This applies, for example, to the selection of models and AI apps, competence building, and work on common standards.

Solidary action should be anchored along the entire AI usage cycle, the white paper states. In the development phase, this means insisting on transparency about datasets and algorithms, using diverse training data to reduce bias, and including the needs of marginalized groups. Furthermore, the invisible work of clickworkers must be recognized. Fair conditions should also apply to them.

The authors see the choice of specific AI applications as a lever. Here, they recommend using open-source or semi-open (open weight) models to facilitate verifiability and reduce technological dependencies. At the same time, it is essential to define clear boundaries and non-application areas, for example, in sensitive areas such as personnel selection or resource-intensive image generation. The association Digitale Gesellschaft serves as a role model, having consciously decided against using generative AI like ChatGPT after thorough consideration. This is because the potential benefits do not outweigh the risks of reproducing discrimination and entrenching exploitative supply chains.

AI must not replace humans

In the deployment phase, solidarity must be ensured by empowering employees and jointly reflecting on responsibilities, is another appeal. The introduction of AI requires building data literacy through low-threshold educational offerings. Any efficiency gains should primarily be used to create scope for relationship work, care, or strategic activities. Human work should not be replaced.

An example of joint learning, according to the authors, is the BAGSO project “AI for a good aging,” in which older people can strengthen their skills and pass on their knowledge at special learning centers. In November, an alliance from civil society, under the banner of “democratic AI,” already signed a self-commitment for dealing with the technology.

(mho)