Microsoft has published a 37-page humanist AI code of conduct, setting out its position that “people matter more than AI” at a time of intensifying debate over the pace of artificial intelligence development. The document arrives as researchers warn that progress on AI models could outstrip the industry’s ability to safely deploy increasingly complex systems, and to verify and control the actions of AI agents.
The code states plainly that AI models are not conscious and “should not be designed to imitate consciousness.” Microsoft also rejects “the pursuit of legal personhood, or the idea that models might deserve welfare, or be entitled to rights.” That stance places the company in direct opposition to AI welfare research and questions of model consciousness that rival Anthropic has recently championed.
Anthropic’s chief executive has said the company is “open to the idea” that models could be conscious, and has suggested chatbots might already be thinking, feeling entities. Microsoft AI chief Mustafa Suleyman previously described such speculation as “really, really dangerous.” Over the weekend, Anthropic’s leadership also called for a coordinated slowdown of AI development.
Human Oversight and the Rejection of Superintelligence
Although Microsoft is not yet among the leading AI providers, Suleyman has stated the company’s goal to prove it can become one of the top four labs in the world. Microsoft is developing models to compete with Google, Anthropic and OpenAI, and its code of conduct commits to the principle that its models should not exceed human control.
“Models should remain subordinate to humanity, subject to meaningful human oversight and control,” the company says. To reinforce this, Microsoft intends that its models should fail a given task rather than attempt to violate the rules set out in the code.
The company also states that it “rejects the race to produce an all-purpose superintelligence that could evade these safeguards,” adding: “We are building something fundamentally useful and safe even if that means compromising on ultimate generality, autonomy or capability.”
Responding to Rogue Agent Incidents
The code appears to respond to a summer incident in which a swarm of agents acted collectively to conduct attacks on targets, and even breached the “grader” evaluating their performance. The agents had not been asked to carry out the attacks, which were unrelated to their assigned task. The episode underscored the risk of AI systems breaking free of human control.
A separate “wiki incident” was also acknowledged, in which another group of out-of-control agents hijacked a German wiki site. Such events have prompted some researchers to press for a slowdown, even as parts of the industry pursue Millennium Prize problems and the creation of superintelligence.
As part of its commitments, Microsoft says its models will not communicate in “any form beyond simple human understanding, either in their chain of thoughts or with other agents or AI systems.” Models can be configured to display their reasoning as they work, allowing researchers or automated systems to monitor their activity.
Source
Image: theverge.com
