On Thursday, November 11 (Eastern Time), Mustafa Suleyman, Microsoft's consumer AI chief and co-founder of DeepMind, pledged to immediately stop development if superintelligence poses a threat to humanity.
Suleyman, who is currently focused on building superintelligent systems "aligned with human interests," stated in a media interview: "We won’t continue developing systems that risk spiraling out of control." He acknowledged this should be an industry consensus but noted, "I think this is still a novel position in the field today."
This declaration comes as Microsoft accelerates its independent AI development. A restructuring of its partnership with OpenAI in October 2023 granted Microsoft previously restricted development rights. In November, Suleyman announced the formation of the MAI Superintelligence team, explicitly tasked with creating "practical technology designed solely to serve humanity."
Suleyman’s commitment highlights the ethical dilemmas tech giants face while pursuing AI breakthroughs. Though rivals like OpenAI and Anthropic claim similar goals, he asserts Microsoft’s stance is distinct.
**Agreement Breakthrough Lifts Development Restrictions**
Microsoft’s strategic shift stems from its revised OpenAI partnership. Previously, contractual terms barred Microsoft from developing artificial general intelligence (AGI) or superintelligence—systems matching or surpassing human capabilities.
Suleyman explained that Microsoft had effectively surrendered these rights in exchange for access to OpenAI’s latest products and years of data center support. "They now have deals with SoftBank, Oracle, and others to build more data centers than Microsoft was willing to provide," he said. "In return, we’ve regained the right to develop our own AI."
This change is pivotal for Microsoft. "For 18 months, we’ve operated as an AGI development entity," Suleyman noted. "Now, we can explore technologies and methods with the potential to outperform humans across all tasks. This marks a real transition for us."
**Superintelligence Team Prioritizes Real-World Applications**
In November, Suleyman unveiled the MAI Superintelligence team in a blog post, which he personally leads. The group will tackle concrete challenges in healthcare diagnostics and education rather than pursuing "nebulous, abstract superintelligence."
Its first goal is creating AI that vastly exceeds human expertise in specialized fields like medical diagnosis. Suleyman wrote: "We’ll achieve expert-level performance in comprehensive diagnostics while excelling in clinical planning and prediction."
While echoing efforts by competitors like Meta, the move has sparked industry skepticism about technological feasibility. Suleyman stressed Microsoft "won’t build superintelligence at all costs or without limits," addressing growing investor concerns over AI spending returns.
**Technical Realities Remain a Work in Progress**
Despite his visionary rhetoric, Suleyman admitted current technology is still evolving. While AI systems like ChatGPT interact in ways unimaginable a decade ago, they fall short of consumer and business expectations.
Microsoft’s Copilot consumer assistant, for instance, has intelligent agent features that "aren’t always accurate" and remain under development. "We’re still experimenting," Suleyman said. "But when it works, it’s the most magical thing you’ve seen."
Suleyman joined Microsoft in early 2023 after the company acquired Inflection AI’s intellectual property and most of its staff. Microsoft has since incorporated models from Google and Anthropic, further reducing reliance on OpenAI. It currently holds a $135 billion stake in OpenAI post-restructuring.
Comments