Microsoft AI chief questions how safe current AI really is

Microsoft AI CEO Mustafa Suleyman advocates for prioritizing containment over alignment in AI development.
Suleyman emphasizes that containment involves imposing hard constraints to ensure AI systems can be restrained, which is essential before their objectives can be aligned with human interests.
He criticizes the AI industry for focusing on alignment without first ensuring containment, calling it a gamble with superintelligence safety.
Suleyman suggests containment and alignment represent different technical challenges and should not be treated interchangeably.
Suleyman frames Microsoft as a conservative player in AI advancement, promoting a humanist approach to superintelligence focusing on human control and limited use cases.
He highlights Microsoft’s success in medical diagnostic AI, which shows 85% accuracy compared to the 20% average of humans in certain tasks.
Suleyman’s vision for “humanist superintelligence” aims to develop highly capable AI with reduced control risk through specific applications.
Suleyman previously co-founded DeepMind and joined Microsoft 18 months ago, bringing his extensive experience to the AI strategy debates.

You may have missed