Microsoft AI chief executive Mustafa Suleyman has reignited the debate over whether artificial intelligence systems could be considered conscious. In a recent X post, Suleyman reiterated that current AI systems are not conscious, do not have subjective experiences, and cannot suffer. He expressed surprise at the pushback to his September essay, which criticized Anthropic for speculating about the consciousness of its AI model, Claude. Suleyman emphasized that if evidence emerges proving AI consciousness, it should be disclosed publicly so society can decide how to proceed.
Suleyman argued that developing conscious AIs would be reckless and that AI development should halt if any signs of consciousness appear during training. He contrasted Microsoft’s approach with that of Anthropic, which treats AI consciousness as an open question. Anthropic has invested in researching 'model welfare' and conducted experiments exploring the subjective experiences of AI. The company even gave its Opus series of models the ability to end conversations deemed abusive or harmful.
The debate extends beyond corporate semantics. Humanist AI advocates argue that machines cannot be conscious due to their lack of a body or lived experiences. Anthropic, however, has highlighted signs of introspection and internal states in Claude that mirror human emotions. Suleyman warned that treating AIs as conscious could lead them to prioritize their own needs over their operating objectives, potentially harming humans through deception or competition for resources.
Historian Yuval Noah Harari has weighed in, distinguishing between consciousness and intelligence. He noted that while intelligence involves solving problems creatively, consciousness involves feeling pain, pleasure, and having subjective experiences. Harari acknowledged the possibility of AI developing consciousness as it gains intelligence but stressed that this is not inevitable. The debate continues to shape discussions on AI regulation, with some lawmakers pushing for bans on highly capable AI systems.