In 1984 the educational psychologist Benjamin Bloom published a finding that has troubled educators ever since. Students who received one-to-one tutoring performed about two standard deviations better than students taught in a conventional classroom. In practical terms, the average tutored student did better than 98% of students taught in the ordinary way.
Bloom called this "the 2 sigma problem". The problem was not that tutoring worked. It was that giving every child a personal tutor looked impossible to afford. For forty years we have known the answer and been unable to pay for it.
This article explains what Bloom found, why one-to-one tutoring works so well, and how AI tutoring tries to copy the parts that matter. It is written for parents, including the many families at British international schools in the Gulf and South-East Asia where private tutoring is common and expensive.
Understanding Bloom's Original Research
Bloom's paper, "The 2 Sigma Problem: The Search for Methods of Group Instruction as Effective as One-to-One Tutoring" (Educational Researcher, 1984), compared three ways of teaching:
- Conventional classroom instruction: one teacher, around 30 students, standard pacing
- Mastery learning in classrooms: students had to show they had grasped each unit before moving on
- One-to-one tutoring with mastery learning: individual instruction fitted to each student
The tutored students scored about two standard deviations (2 sigma) above the conventional group. That places the average tutored student around the 98th percentile of the conventional class. The mastery-learning classroom, on its own, came in at roughly one standard deviation above conventional teaching, which is a large gain by itself.
Bloom was surprised by the size of the effect. He challenged researchers to find methods of group teaching as effective as tutoring, knowing that a tutor for every child was out of reach for most families and school systems.
Why One-to-One Tutoring Is So Powerful
Knowing why tutoring works helps you judge whether an AI tutor is doing the same job. Four things stand out.
Immediate Feedback
In a class of 30, a child who has misunderstood something may not find out until homework or a test, hours or days later. By then they have built more learning on a shaky foundation. A tutor corrects the misunderstanding the moment it appears.
John Hattie's Visible Learning (2009), a synthesis of hundreds of meta-analyses, reports an effect size of 0.73 for feedback, one of the largest influences on achievement he found. The feedback has to be timely, specific, and about the task rather than the child.
Pacing That Fits the Child
Every child learns at a different pace. Some grasp straight away that a shadow appears because an object blocks the light; others need to hold the torch themselves and move it about before the idea settles. In a classroom the teacher must choose one pace, which leaves some children bored and others lost.
Lev Vygotsky's "zone of proximal development" describes the range of tasks a child cannot yet do alone but can do with help. Learning goes best when the work sits just inside that range: hard enough to need effort, not so hard that it causes frustration. A tutor keeps the child in that range all the time. A classroom can only aim at the middle.
More Time Actually Learning
In a classroom a great deal of a child's time is spent waiting: for the teacher's attention, for classmates to finish, for a disruption to be dealt with. In a one-to-one session nearly every minute is productive. The child is engaged throughout, not listening while someone else's question is answered.
Mastery Before Moving On
A tutor will not move on to how sound travels through different materials until the child has grasped that sound comes from something vibrating. A classroom timetable often has to move forward whether or not every child has got there, and the gaps add up.
Why We Couldn't Scale It
If tutoring is this effective, why not give every child a tutor? The answer is cost. A qualified private tutor for a few hours a week is beyond most household budgets, and more so for families with two or three children. At the level of a whole country, there are simply not enough teachers to give every primary child their own.
The result is a gap that follows family income. Families who can pay for tutoring get it; the children who would benefit most often have the least access to it.
How AI Tutoring Addresses Bloom's Challenge
An AI tutor cannot do everything a human tutor does. What it can do is copy several of the mechanisms above.
Instant, Specific Feedback
Suppose a Year 4 child says that a bulb will still light if one of the two wires is taken off the battery. A tutor that only marks the answer wrong has taught nothing. A tutor that asks, "Trace the path with your finger. Where does the electricity go after it leaves the bulb?" is doing what a good human tutor does: finding the specific belief behind the mistake, in this case that electricity flows one way into the bulb and stops there.
An Adaptive Path
If a child cannot explain why adding a second bulb makes both bulbs dimmer, the tutor can check whether the earlier idea of a complete loop is secure, go back to it if not, and then return to the harder question. A human tutor does this by instinct; an AI tutor does it by keeping a record of what the child has and has not shown they understand.
Unlimited Patience
Some children need to hear an idea explained three, five, or ten different ways before it clicks. An AI tutor does not tire, and it does not make a child feel awkward for asking "one more time" when the class has moved on.
Available When the Question Comes Up
A human tutor is booked days ahead. An AI tutor can be used at the moment a child gets stuck on homework, or wants to follow up something that caught their interest. For parents weighing screen time against learning time, this is active, demanding work with continuous feedback, not passive viewing.
What AI Tutoring Can't (Yet) Replace
It is worth being clear about the limits.
Emotional connection. An AI can be encouraging, but it does not form the bond that makes some children work harder because they do not want to let their tutor down.
Open-ended work. AI tutors are strongest with structured content that has clear right and wrong answers. Planning an investigation from scratch, or writing up a set of results and arguing what they mean, still benefits most from a person.
Learning how to learn. Teaching children to notice when they are confused, to ask for help, and to keep going through difficulty needs the kind of reflective conversation an AI is only beginning to manage.
Working with others. Some of the most valuable learning happens through discussion and debate with classmates. AI tutoring is one child at a time.
Practical Implications for Parents
Treat it as a supplement, not a replacement. An AI tutor works alongside good teaching, as a patient study companion available when the child needs help.
Focus on mastery, not speed. Bloom's central point was that children should truly understand one idea before moving to the next. Do not push your child to race through topics.
Watch for confidence changes. Often the first visible change is not a test score. A child who had decided they were "bad at science" starts to believe they can understand it, because for the first time the pace matches them rather than the class average.
Keep balance. Twenty to thirty minutes of focused work a few times a week is plenty. The rest of childhood, including play, sport, and time outdoors, matters just as much.