The Machine Intelligence Research Institute (MIRI) is a nonprofit research organization in Berkeley, California, and one of the earliest groups devoted to the problem of aligning advanced artificial intelligence with human interests. According to its own history, it was founded in 2000 by Eliezer Yudkowsky together with Brian and Sabine Atkins, originally under the name the Singularity Institute for Artificial Intelligence. The organization adopted the name Machine Intelligence Research Institute in 2013.
MIRI’s stated mission is research and outreach intended to help ensure that the creation of smarter-than-human intelligence has a positive effect. By its own account the organization’s strategy shifted over time: in its earliest years it was oriented toward accelerating AI progress, but Yudkowsky came to view the alignment problem as the central challenge and redirected its work toward solving it. Through the 2000s and early 2010s it focused on field-building and on foundational mathematical questions in alignment, such as decision theory and the behavior of self-modifying agents.
The institute was influential less for products than for ideas and people. Its founder’s writing did much to develop and popularize concepts like the intelligence explosion and the difficulty of specifying safe goals, and the community that formed around MIRI and around Yudkowsky’s online writing produced many researchers who later worked at major AI labs and safety organizations. Nick Bostrom’s 2014 book Superintelligence drew on this body of work.
MIRI’s outlook grew more pessimistic over time. By the 2020s its leadership had shifted much of its effort toward public communication and policy, arguing that the alignment problem remains unsolved and that the development of artificial superintelligence carries grave risks - a position more cautionary than the views of the large commercial AI labs.