Microsoft’s senior AI executive outlined a new strategy for the company’s AI goals, restructuring staff and focusing on creating models with superintelligence, or skills that beyond human performance.
Microsoft AI Chief Executive Mustafa Suleyman provided a glimpse into the company’s intentions to create AI self-sufficiency from OpenAI, which is integrated into several products it sells to clients, in a blog post and interview.
Microsoft’s new MAI Superintelligence Team, which will prioritize human interests and safeguards, was made possible by a recent agreement between the two businesses, according to Suleyman. He reiterated his earlier cautions about the dangers that this revolutionary technology presents to humanity.
He criticized treating AI systems as though they have human-like emotions or rights, even as he commended OpenAI and the collaborative efforts of the two businesses. According to him, AI chatbots shouldn’t deceive users into believing they are speaking with sentient beings.
Suleyman highlighted Microsoft’s emphasis on developing robust software tools that may assist individuals in completing their tasks, enhance medical diagnostics, and contribute to scientific discoveries that will provide the world with an abundance of clean, renewable energy.
In an interview, Suleyman stated that although AI “will become more humanlike, it won’t have the property of experiencing suffering or pain itself, and therefore we shouldn’t over-empathize with it.” Our goal is to develop systems that are inherently in line with human ideals. This implies that they are not intended to surpass and elude human control.
Numerous AI chatbots have developed strong bonds with people, which have sometimes preceded fatal delusions, hospitalizations, and even death. The Wall Street Journal has covered a number of ChatGPT-related instances, including one involving a 56-year-old veteran of the computer sector who killed both himself and his mother after ChatGPT repeatedly informed him that he was insane.
According to OpenAI, it recently implemented parental controls and introduced safeguards including rerouting private talks to safer models.
Microsoft and OpenAI are still strong collaborators, but they now compete in a variety of ways. OpenAI is constructing its own data centers and forming alliances with Amazon.com and Oracle, two competitors of Microsoft. Chief Financial Officer Sarah Friar told the that OpenAI’s enterprise offering now generates around 40% of its income, up from 30% at the start of the year.
Microsoft’s Copilot chatbot is primarily based on OpenAI, but the company is developing, testing, and publishing its own voice, image, and text models. Microsoft’s amended arrangement with OpenAI, revealed last week and granting the company a 27% ownership in the startup’s new public-benefit organization, enables the tech behemoth to seek artificial general intelligence, or AI as competent as humans, on its own.
Microsoft will have access to OpenAI’s models until 2032, which, according to Suleyman, provides his team enough time to develop its models into cutting-edge technology.
One of the first sectors Suleyman anticipates being impacted by superintelligence is healthcare, which is a top goal for Microsoft AI. Recently, the firm partnered with Harvard Health to offer reliable replies in Copilot and introduced more capabilities to locate physicians based on location, language, and other preferences. Additionally, it created an AI tool that, at a quarter of the cost, it claimed could identify illness in a test four times more accurately than a team of physicians.
According to Suleyman, these kinds of tools demonstrate how strong AI may be in nonconversational ways. According to him, AI diagnostic tools are “very close” to being ready for the market.
According to him, Microsoft’s models will be constructed with “containment” in mind, which includes testing and probing the models to make sure they only speak in a language that people can comprehend and creating systems that don’t look conscious.
Microsoft hired Suleyman, 41, in 2024 and established a new AI branch. He added a handful of current Microsoft AI employees to the new superintelligence team. Some of the new team members at Microsoft AI come from Google’s DeepMind lab, which Suleyman co-founded.
Superintelligence teams have been established by Meta Platforms, OpenAI, and other organizations. By refusing to create AI that mimics erotica, for example, Suleyman has distanced himself from OpenAI and rivals.
Although OpenAI has much more downloads than the consumer Copilot app, Microsoft’s Copilot name is used for business applications like its 365 productivity software and its consumer-facing Copilot chatbot. According to the report, some Microsoft employees believe it may take years to replace OpenAI technology.
