Bloom's 2 Sigma Problem
One-on-one tutoring can substantially outperform conventional classroom instruction, but its cost has historically prevented it from scaling. AI may loosen that constraint by making personalized tutoring cheaper to provide.
In Bloom’s study, students taught one-on-one improved by two standard deviations—an advantage so large that the precise estimate remains disputed.
E1Personalization collides with scale
The result points to a structural difference between tutoring and classroom instruction: one tutor serves one learner, while a classroom teacher must divide attention across many. The apparent educational gain therefore comes attached to an economic bottleneck. Personalized instruction may work unusually well, yet remain unavailable to most students because supplying a human tutor to each one is expensive. AI matters here not because it proves that software teaches as well as a person, but because it may reduce the cost of providing individualized help.
E1Where it shows up
Bloom’s tutored students
The reported two-standard-deviation improvement is the striking case behind the problem: an unusually effective form of instruction existed, but conventional classrooms could not reproduce its one-to-one structure at scale.
E1A benchmark, not a guarantee
The exact size of Bloom’s effect is disputed, and the finding does not establish that every tutor—or an AI tutor—will produce the same result. The principle identifies a promising gap between personalized and mass instruction; it does not prove that any particular technology can close it.
E1Episodes that teach this
-
AI Is Making Students Smarter (and Dumber) | Should AI be Used in Education? Future IQ
· explained at 4:16
207 views
we have technology which can provide one-on-one tutoring at scale and that is AI.