Skeptics call Meta's AI glasses "gawking apparatus" – study shows consequences for AI language models

Legal assessment in Germany

In Germany, the focus is primarily on the Federal Network Agency (Bundesnetzagentur), which, according to MDR reports, has provided guidance on the AI glasses and the data protection regulation TDDDG. The authority points to existing rules according to which devices with built-in cameras may not be operated without further ado in certain situations. The state government of Saxony has also taken up the issue and is discussing how to deal with the recordings in public spaces.

Parallel to this discussion, a study by Google, the University of Chicago and other universities has investigated how the training of AI chatbots affects them when they are drilled to not claim consciousness for themselves. According to the findings, the training not only changes statements about one's own "inner life." The models also begin to attribute less inner life to animals, plants, the sea, the wind and devices. In addition, they express belief in God, an afterlife or the supernatural less frequently.

Effects on AI language models

The study classifies the change as an indirect effect: when a model learns to be cautious with attributions of consciousness about itself, it transfers this caution to other entities. This is relevant for the debate on AI safety, because it shows that seemingly technical training decisions can have far-reaching linguistic and ideological consequences. The Karlsruhe Center for AI Updates (ZIT) has also published similar observations.

In another business chapter, the AI company Anthropic says it is preparing an IPO that could surpass SpaceX's previous record. Industry observers see the move as a sign of growing acceptance of AI companies in the capital markets. The exact terms are still open.

Market developments and AI agents

At the same time, so-called AI agents are driving the discussion about responsibility. These programs increasingly act autonomously on the internet and raise questions of liability and oversight. The findings from the language model studies are feeding into the ongoing debates about regulation. Critics are calling for training guidelines to be made more transparent, while advocates point to the benefits for safety.

Overall, the complex of topics shows how closely data protection, AI safety and market developments are intertwined. While some see the glasses as a "gawking apparatus," others see them as a useful tool. The coming months will show whether regulators such as the Federal Network Agency will enact new rules or whether the industry will voluntarily adapt its standards.