Microsoft AI executives promise: If superintelligence threatens humanity, development will cease

On Thursday, November 11th, Eastern Time, Mustafa Suleyman, the head of consumer artificial intelligence (AI) at Microsoft and a co-founder of DeepMind, pledged to immediately halt related development work if superintelligence poses a threat to humanity.

Suleyman is currently focused on creating superintelligent systems that "align with human interests." In a media interview on Thursday, he stated, "We will not continue to develop systems that could potentially go out of control." He believes this should be a consensus in the industry, "but I think this is a novel position in the current industry."

This statement comes as Microsoft accelerates its autonomous AI development. An agreement last October reshaped Microsoft's relationship with OpenAI, granting Microsoft previously restricted development rights. Suleyman announced the establishment of the MAI superintelligence team in November, explicitly stating that it would build "practical technologies designed solely to serve humanity."

Suleyman's commitment highlights the ethical considerations that tech giants face while pursuing breakthroughs in AI. Although competitors like OpenAI and Anthropic also claim to have similar goals, Suleyman believes Microsoft's position is unique.

Agreement Breaks Development Restrictions

Microsoft's shift in AI strategy stems from its adjustment in relations with OpenAI. Previously, Microsoft was unable to develop general artificial intelligence (AGI) or superintelligence due to contractual limitations. The former refers to systems with capabilities comparable to humans, while the latter surpasses human abilities.

Suleyman explained that Microsoft had essentially relinquished these development rights in exchange for the ability to use OpenAI's latest products and had built and equipped data centers for OpenAI over the years. "They now have agreements with many companies like SoftBank and Oracle, with more data centers built than Microsoft was willing to build for them," Suleyman stated. "In return, we now have the right to develop our own AI."

This shift is significant for Microsoft. He said, "For the past 18 months, we have been an AGI development organization, but now we can explore technologies and methods that have the potential to surpass human performance across all tasks. This is indeed a transformation for us."

Superintelligence Team Focuses on Practical Applications

In November, Suleyman announced the establishment of the MAI superintelligence team in a blog post, which he personally leads. The team will focus on practical issues in areas such as medical diagnosis and education, rather than "vague and ethereal superintelligence."

The team's first goal is to develop AI that far exceeds human capabilities in specific fields like medical diagnosis. Suleyman wrote in the blog, "We will achieve expert-level performance in comprehensive diagnostics while possessing high levels of planning and predictive capabilities in clinical operational environments."

This initiative echoes similar efforts by competitors like Meta but also raises industry skepticism about the potential for technological breakthroughs. Suleyman emphasized that Microsoft "will not build superintelligence at all costs and without limits," addressing growing investor concerns about the return path on AI expenditures

The Technical Reality is Still Being Improved

Despite discussing the vision of superintelligence, Suleyman admits that the current technology is still in development. AI systems like ChatGPT can interact with humans in ways that computers a decade ago could not, but there is still a gap in meeting consumer and business expectations.

The intelligent agent feature of Microsoft's Copilot consumer assistant "is not always accurate" and is still in the development stage. "We are still experimenting," Suleyman said, "but when it really works, it's the most amazing thing you've ever seen."

Suleyman joined Microsoft early last year, after Microsoft acquired the intellectual property and most of the employees of Inflection AI, which he founded. Since then, Microsoft has also started using models from Google and Anthropic, continuously reducing its reliance on OpenAI. Currently, Microsoft holds a $135 billion stake in the restructured OpenAI