A behaviourist said classrooms reinforce too little and too late

The classroom note

On 24 October 1958, the journal Science published a paper by B. F. Skinner titled, plainly, 'Teaching Machines.' Its opening complaint is about ordinary classroom practice: a teacher moving a class through a lesson 'usually knows that he is moving too slowly for some students and too fast for others,' so fast students are held back and slow ones are, in Skinner's words, 'unnecessarily punished by criticism and failure.' Four years earlier, in the Harvard Educational Review, Skinner made the same argument at length under the title 'The science of learning and the art of teaching' — a paper this 1958 article cites directly, giving the citation as volume 24, issue 2, 1954.

What the evidence says

The 1958 paper is a technical description, not an experiment with a control group; it argues from theory and describes a device Skinner built, rather than reporting a trial. Its central claim is that a machine can act 'like a private tutor,' reinforcing correct responses, using immediate feedback 'not only to shape his behavior most efficiently but to maintain it in strength.' Material is broken into small 'frames,' each requiring a written response, with hints built in so most answers are correct; a wrong answer is corrected on the spot, not accumulated onto a graded paper. Skinner estimates fifteen minutes a day 'should suffice' for a programmed course. A companion piece, 'New methods and new aims in teaching' (1964), restates the argument and adds that 'machines are inevitable' given rising demand — not to regiment students, Skinner argues, but to let each proceed individually.

The implementation question

Neither paper states what it would cost a school to build or buy such a device, who writes its 'frames,' or how much teacher training it needs. Skinner notes, in passing, that producing programmed material is demanding: frames for a physics course or a language vocabulary list are described as an ongoing project by named assistants, not a one-off task. The implicit cost is authorship — someone must break a subject into small, mostly-answerable steps — a labour requirement 'immediate feedback' alone does not disclose.

What holds and what fails

What holds is the mechanism Skinner is precise about: a response confirmed or corrected before the next step changes behaviour differently than one graded and returned later. Largely absent from both papers is any sustained treatment of motivation beyond reinforcement — curiosity, contact with a teacher, or a student's own goals barely appear. That silence is notable given how much of today's argument about engagement with AI tutors turns on exactly those questions; it is an editorial gap, not one either paper claims to fill.

• Does a tool's feedback loop match Skinner's mechanism, or is 'personalised' doing different work?
• Who authors the sequence of steps, and is that labour costed in a vendor's pitch?
• What does the source say, if anything, about motivation beyond correct-response reinforcement?

Skinner's 1958 paper is a primary account of an argument underpinning most automated tutoring since, including generative tutors; read against what it omits, it is as useful for its silences as for its claims.