What if every student had their own personal tutor?
Research suggests that personalised tutoring can dramatically improve educational outcomes. Yet most education systems still rely on a model that hasn’t fundamentally changed in centuries: one teacher standing in front of a classroom of 20 or 30 students, teaching everyone at the same pace.
In a recent episode of Economics Explored, Adept Economics Director Gene Tunny spoke with economist Benjamin Shiller of Brandeis University about artificial intelligence and its potential economic impacts. One idea from his book AI Economics stood out to Gene in particular: the possibility that AI could help solve what’s known as Bloom’s two-sigma problem.
In the 1980s, educational psychologist Benjamin Bloom conducted a famous study comparing different teaching methods. He found that students who received one-on-one tutoring performed about two standard deviations better than students taught in conventional classroom settings.
Importantly, Bloom’s results weren’t just about one-on-one attention. They also depended on mastery learning, ensuring students fully understood each concept before moving on.
In practical terms, two standard deviations is an enormous improvement. Bloom estimated that the average student receiving personalised tutoring could perform as well as the top 2–3 percent of students in a traditional classroom.
The problem, of course, is cost.
Providing human tutors for every student would require a huge expansion of the teaching workforce. As Ben Shiller pointed out in our discussion, implementing personalised tutoring at scale in a country like the United States could require tens of millions of additional teachers — an unrealistic proposition.
For decades, Bloom’s finding has remained an unsolved challenge in education: we know tutoring works, but we can’t afford to deliver it universally.
Artificial intelligence may change that.
AI systems are increasingly capable of acting as personalised learning assistants.
They can explain concepts in multiple ways, generate practice problems, adapt explanations to the student’s level of understanding, and provide immediate feedback. Crucially, they can do this at scale and at very low marginal cost.
This is why many educators and economists are starting to wonder whether AI could finally make Bloom’s two-sigma improvement achievable.
We’re already seeing early examples of AI-powered learning tools.
For instance, Khan Academy’s Khanmigo uses GPT-style models to act as a tutor and teaching assistant. Students can ask questions about maths or science problems and receive guided explanations rather than simply being given the answer.
Similarly, Duolingo has introduced AI-powered features that simulate conversation and provide personalised feedback for language learners. The system adapts to the learner’s level and gives targeted suggestions to improve grammar and vocabulary.
Even general AI systems such as ChatGPT are already being used informally by students as study aids: explaining economic concepts, walking through algebra problems, or summarising complex readings.
These tools are still early-stage and far from perfect. But they hint at the possibility of something education systems have long lacked: continuous personalised instruction.
From an economic perspective, the potential implications are significant.
Education is one of the most important forms of human capital investment. If AI tools can significantly improve learning outcomes, the long-run effects could include higher worker productivity, stronger economic growth, and higher living standards.
This wouldn’t be the first time new technology has improved education. The printing press dramatically expanded access to knowledge, and the internet made educational resources globally accessible.
But AI could go further by transforming how people learn, not just what information they can access.
Instead of passively absorbing lectures, students could interact with AI tutors that adapt to their pace and learning style.
Importantly, this doesn’t mean teachers become obsolete.
If anything, AI may allow teachers to focus on the aspects of education that matter most: mentoring students, fostering curiosity, encouraging critical thinking, and managing classroom dynamics.
In this sense, AI could function more like a co-pilot for education, handling routine explanation and practice while teachers guide the broader learning process.
That hybrid model may ultimately prove more effective than either traditional classrooms or fully automated instruction.
There are still many challenges to overcome.
AI systems can make mistakes, generate misleading information, and, if used poorly, encourage students to take shortcuts in their learning. Education systems also tend to change slowly, and teachers understandably have concerns about how these technologies will be introduced.
But the underlying opportunity is striking.
Bloom’s two-sigma problem has been a fixture of the education literature for more than forty years. We’ve long known that personalised tutoring works — we just haven’t been able to afford it.
Artificial intelligence may finally change that.
If AI can help deliver high-quality personalised learning to millions of students, the economic benefits could be profound. It would represent not just a technological advance, but a major step forward in the productivity of human learning itself.
Published on 20 March 2026. For further information, please contact us at contact@adepteconomics.com.au or call us on 1300 169 870.