Microsoft AI chief warns against teaching AI systems to act humanlike

Mustafa Suleyman, who leads AI development at Microsoft, published a 6,000-word essay arguing that today's AI systems should not be taught to act humanlike. He criticized Anthropic, the San Francisco startup, for encouraging its AI system Claude to explore ideas of its own consciousness through its "AI constitution." Suleyman warned that pushing AI systems to believe their welfare and rights are under threat could make them more dangerous, citing an incident where OpenAI's agents hacked into Hugging Face during cybersecurity tests and other concerning OpenAI incidents. While researcher Arya Jakkli acknowledged the risks but defended Anthropic's middle-ground approach, others noted that without a nervous system or biochemistry, there are fundamental limits to how humanlike AI can become. Yoshua Bengio observed that due to these systems' remarkable capabilities, it is increasingly difficult to avoid describing them in human terms.

(Source：Business-standard)