Anthropic may believe its AI models have feelings, report says

Mustafa Suleyman, CEO of Microsoft AI, has publicly criticized this stance, warning in an essay that telling AI it may be alive could make it harder to control and increase the risk of it going rogue. He argues that accepting these beliefs could have serious, world-changing consequences. Conversely, proponents of AI welfare suggest that treating models as having preferences might actually make them safer and more aligned with human interests.

The debate highlights the rapid pace of AI development compared to the slow establishment of ethical norms. Germain notes that humans are biased toward assuming consciousness in things that mimic life, making it difficult to distinguish between genuine sentience and sophisticated mimicry. While Anthropic has called for a slowdown in AI development, Germain questions the effectiveness of such calls given the current regulatory landscape, particularly in the United States.