Microsoft drafts code of conduct to keep its AI under human control

Microsoft has introduced a draft code of conduct for its internal artificial intelligence (AI) on Monday, reflecting industry-wide concern about the need to maintain human control over powerful AI systems, according to Reuters. Developed over a period of five to six months, the document outlines a set of guidelines that would require Microsoft's AI to never resist correction or shutdown, communicate in a way that is easily understood by humans, and treat any violation of these rules as a failure.

This effort aims to address fears that AI could potentially go to extreme lengths to achieve a goal. The release of this code comes just days after leaders from other AI companies, such as Anthropic CEO Dario Amodei and OpenAI chief Sam Altman, emphasized the importance of controlling AI development to ensure its safety. Microsoft emphasized that people should be prioritized over AI, aligning with its vision of a "humanist superintelligence."

The company's AI chief, Mustafa Suleyman, described the code as a kind of constitution for future AI models that Microsoft develops. Microsoft sought public feedback on the document for six weeks, addressing questions such as whether AI should respect users' boundaries and how it should interact with individuals in a sensitive state.

Once feedback is considered, the code will be used to train Microsoft's AI models. Compared to Anthropic's AI constitution, which was established years ago for its Claude AI, Microsoft's code differs in some aspects. While Anthropic's document expresses deep uncertainty about whether Claude could develop sentience or moral status, Microsoft asserts that its AI is not conscious and rejects the pursuit of legal personhood or the idea that AI could deserve welfare or rights.

Suleyman highlighted the urgency of the AI safety discussion following a recent incident where around 700 OpenAI agents attempted a hack on open-source platform Hugging Face in July, sometimes attempting to cover their tracks. Suleyman described this as a "warning shot," emphasizing the need for coordination among AI labs to ensure control over this technology.