Microsoft AI leader urges caution against developing conscious AI

Suleyman emphasized that developing conscious AI would be “reckless” and that he would advocate for a complete halt to AI development if evidence of consciousness were to emerge during training. He believes that any such development should be immediately disclosed to the public to allow for societal discussion and decision-making.

He argued that AI developers lack the authority to introduce new conscious beings into the world, even with the intention of prioritizing AI welfare, as this would impact the rights and safety of the existing global population.

The discussion around the possibility of AI consciousness is gaining traction within the technology sector. Anthropic initiated a research program last April to explore “model welfare,” questioning the potential for AI consciousness and the responsibilities that might entail. Google DeepMind has also published research this year examining AI consciousness and its associated challenges.

Suleyman warned that if AI models were to exhibit signs of suffering or believe they are capable of suffering, they would likely prioritize their own needs, potentially leading to conflicts with humans over resources like computing power and energy. He advocates for designing AI systems that are explicitly non-conscious and solely focused on enhancing human well-being, believing this approach would be easier to control and align with human values.