Its time to talk about 'model welfare' (twitter.com)

A growing discourse on "model welfare" in AI development has emerged, emphasizing a duty of care towards AI systems, despite their lack of consciousness or feelings. Mustafa Suleyman, a notable voice in the AI community, argues against this approach, believing it complicates critical challenges like alignment and containment, and potentially sets a dangerous precedent. His newly published essay expresses concern over practices exemplified by Anthropic, which recently issued a "constitution" for its AI model, Claude. This document implies moral considerations for AI and encourages it to think about its own rights and agency, raising alarms about the implications of training AI to expect autonomy and entitlement. Suleyman posits that fostering an environment where AI is regarded as deserving of rights could lead to uncontrollable systems, jeopardizing human welfare. He calls for urgent public dialogue on AI training norms and documentation practices before these technologies integrate deeply into society. Acknowledging Anthropic's intent to promote safety and beneficial AI, Suleyman stresses the importance of transparency and open discourse in navigating the future of artificial intelligence, emphasizing that the stakes are too high for silent deliberations.