In 1984 the educational psychologist Benjamin S. Bloom published “The 2 Sigma Problem: The Search for Methods of Group Instruction as Effective as One-to-One Tutoring” in Educational Researcher (volume 13, issue 6, pages 4-16). It reported a striking finding: students who received one-to-one tutoring combined with mastery-learning techniques performed, on average, two standard deviations - two sigma - better than students taught in a conventional classroom of about thirty. Put differently, the average tutored student scored above 98 percent of conventionally taught students.
The paper drew on doctoral dissertation studies by Bloom’s University of Chicago students Joanne Anania and Joseph Arthur Burke, who compared three conditions: conventional instruction, mastery learning in a group, and tutoring. Bloom framed the result as a problem rather than a solution, because individual human tutoring for every student is far too expensive to deliver at scale. The challenge he posed was to find group methods - or new technologies - that could approach the two-sigma effect at affordable cost.
Why business readers should care: the 2-sigma problem is the explicit target of nearly every modern AI tutoring pitch, from Khan Academy’s Khanmigo to adaptive learning startups. Whether large language models actually close any meaningful fraction of that two-sigma gap remains an open empirical question, but the number itself is the yardstick the whole field is measured against.