Microsoft Corp. is developing a more powerful form of AI called “superintelligence”with which he hopes to achieve advances in areas such as medicine and materials science.
Mustafa Suleyman, head of Microsoft’s AI group, will lead the MAI Superintelligence Team, which will focus on hypothetical milestones even more ambitious than artificial general intelligence (AGI). This is the often ambiguous term that organizations like OpenAI use to describe systems capable of achieving human-comparable performance. Previously, Microsoft had agreed not to develop IAG as part of its collaboration with OpenAI.
“If Artificial General Intelligence (AGI) is usually considered the point at which an AI can match human performance in all tasks, superintelligence is achieved when it can far exceed that performance,” said Suleyma.n in a blog post on Thursday, where he announced the initiative, which he said will focus on the development of AI personal assistants and advances in healthcare and clean energy. The team will seek to build what it calls Humanistic Superintelligence, avoiding the potential risks associated with the development of powerful automated tools and working for the benefit of people rather than achieving technological milestones.
OpenAI, Meta Platforms Inc. and other companies are increasingly focusing on superintelligence as the new goal of AI development. The term “superintelligence,” like AGI (Artificial General Intelligence), is vague: it is not clear how capable AI must be at certain tasks to cross the threshold from “general” intelligence to “superintelligence.”
The announcement follows a renegotiated agreement between Microsoft and OpenAI that determined the software maker’s participation in the startup and modified some aspects of their relationship. This included the removal of a previous ban on Microsoft’s development of advanced AI tools, which had limited much of the Redmond, Washington-based company’s work to smaller, narrower-scope models than those powering ChatGPT.
Thursday’s announcement formalizes a project that Microsoft had been preparing since March of last year, when it hired Suleyman and licensed the intellectual property of his startup, Inflection AI. With a combination of reorganized Microsoft teams and new hires, Suleyman set about building a new family of Microsoft AI models, which to date remain much smaller than the most powerful products from OpenAI or Google (Alphabet Inc.). In September, Suleyman told employees that Microsoft would make “significant investments” to expand the capacity of those models.
