Anthropic updates AI interaction rules

Anthropic, the company behind the Claude AI model, has introduced a new policy that restricts users from being cruel to its artificial intelligence. Under the new rules, the AI may end a conversation if a user continues to be hostile or abusive.

Policy aims to retain user engagement

The change is seen as a strategy to keep users engaged with AI chatbots for longer periods. By using anthropomorphism—the human tendency to assign emotions to non-human entities—the company hopes to increase the time users spend interacting with the AI.

AI lacks actual emotions or suffering

Artificial intelligence does not experience emotions or suffering. The perception of discomfort from users is based on mathematical valence, a concept where AI systems adjust probabilities and reduce uncertainty in response to clear instructions. In contrast, insults or confusing commands increase computational costs and errors.

Debates about AI's emotional mirage

Some experts warn that attributing emotions to AI creates a cognitive mirage that distracts from more pressing issues like data security, information governance, and financial returns. The focus on AI consciousness may lead to an imbalanced evaluation of its capabilities and limitations.

Focus on transparency over emotional mimicry

A recommended approach for future AI development is to emphasize functional transparency in interactions. An AI that openly acknowledges its mechanical nature may lead to more rational use and fewer misunderstandings in its interactions with users.