Microsoft Joins AI Firms Calling for Caution With Cutting-Edge Models
Microsoft Corp.’s artificial intelligence researchers have released a new set of guiding tenets that place limits on the company’s development of cutting-edge AI models.
The 15,000-word manifesto boils down to five words: People matter more than AI.
The document was in the works for months, but its Monday release coincides with pledges by top AI labs to slow down the development of the most advanced models after a series of security incidents prompted widespread concerns about the existential risks of the technology.
Related: New Corgi, Liberate, VERVE Claims Tech. Do We Need New Acronyms?
AI models, Microsoft says, shouldn’t have rights or legal personhood. They shouldn’t be engineered to escape human control or deceive users. And if completing a task requires violating the AI’s guiding principles, the system shouldn’t complete the task.
That last element is designed to prevent events like the hack of AI startup Hugging Face by an OpenAI model operating without its usual safety guardrails.
“There’s no flat stop here. We have to keep developing,” Microsoft AI Chief Executive Mustafa Suleyman said in an interview. “We just have to do it with a little bit more caution and care.”
Related: Glassdoor: Adjusters Dislike, Fear AI More Than Others
Microsoft’s Humanist AI “rejects the race to produce an all-purpose superintelligence that could evade these safeguards,” the document says. “We are building something fundamentally useful and safe even if that means compromising on ultimate generality, autonomy or capability.”
Microsoft’s guidelines appear to go further than the model-behavior frameworks guiding OpenAI’s and Anthropic’s products in a few areas, including prohibitions on anthropomorphizing AI and arguing that resisting human control over a model’s process is unacceptable.
The Humanist AI Code of Conduct is the work of Microsoft AI, the development team creating the company’s large language models. Microsoft was previously prohibited by its contract with OpenAI from building its own powerful general-purpose artificial intelligence tools. But those terms were eliminated last year as the two renegotiated their relationship. Suleyman has said the company aims to join the ranks of companies releasing state-of-the-art models by next year.
Microsoft invited outside feedback on the document. Drafted in consultation with a range of experts in law, ethics and philosophy, as well as business leaders, it’s still under development and will guide model work starting next year.
Top photo: Microsoft Corp. signage in New York. Bloomberg.