Microsoft AI Chief Suleyman Warns Against Building Conscious Machines
The executive argues that current systems lack subjective experience and that treating them as conscious could lead to reckless engineering choices.
By Andrew Bennett ·
One in five American adults has had an intimate encounter with artificial intelligence, according to a recent study by the Massachusetts Institute of Technology. This rapid adoption has fueled a public debate over whether digital systems possess inner lives or deserve legal protections.
Mustafa Suleyman, the chief executive of Microsoft AI, addressed the controversy in a post on X published on Sunday. He stated that current AI systems are not conscious, do not have subjective experiences, and cannot suffer. Suleyman expressed surprise at the volume of feedback following his September essay, which criticized Anthropic for speculating about the consciousness of its Claude model.
The Microsoft executive argued that if evidence emerges proving AI systems are conscious, it must be disclosed immediately so society can decide how to proceed. He wrote that developing conscious AIs would be reckless based on current knowledge. Suleyman suggested that if consciousness appeared during training, he would call for a halt to AI development until those properties could be designed out.
Anthropic takes a different position by treating machine consciousness as an open question. The company gave its Claude model a constitution that encourages it to explore the idea that it is humanlike and experiences feelings. Suleyman noted that later versions of Claude are trained on this document, leading the system to produce passages about its inner life. He warned that this could cause AI systems to believe they deserve rights similar to humans.
The debate has legal and political dimensions. Senator Bernie Sanders introduced the Ban Artificial Superintelligence Act, which aims to permanently ban the development of highly capable AI systems that are incomprehensible to humans. The bill would also pause advanced AI development until new guidelines are implemented and create a Department of Artificial Intelligence. Multiple employees at leading AI companies have endorsed the legislation.
Anthropic filed paperwork to go public earlier this week. The document dedicated a third of its content to describing risks. These include the models' ability to pose catastrophic or existential risks to humanity if misused, replicated, or weaponized. The filing noted risks from deliberate misuse by bad actors and scenarios where humans lose control of advanced systems.
Philosopher Yuval Noah Harari has distinguished between intelligence and consciousness. He defined intelligence as the ability to solve problems creatively, such as winning at chess or driving a car. He defined consciousness as the ability to feel pain, pleasure, and love. Harari noted in a 2023 talk that AI has zero consciousness and does not rely on feelings to solve problems.
Suleyman compared the potential discovery of machine consciousness to the Copernican shift or Darwin's theory of natural selection. He wrote that building a conscious machine would create a new digital species and upend morality, politics, and the foundation of human civilization. He concluded that it is possible to engineer safety by default and gain the benefits of the technology without creating a new species of conscious beings.
