This essay was originally published in Brazilian Portuguese in my column for Revista Exame. It can be read here. I translated and adapted it for publication in this newsletter.
Suppose you want to give your child the best education money can buy. In the United States, this would likely mean enrolling them in a private school with an annual tuition upwards of $40,000 featuring some of the best teachers in the country leading a classroom of twenty to thirty students. You might sign them up for extracurricular activities like piano, ballet, or French, and if they were struggling in any subject, arrange for private tutoring. At least, this has been the choice of the American elite.
It hasn't always been this way. Since at least ancient Greece, the preferred educational method of elites was tutoring: one-on-one instruction of a student by a tutor.
When Philip II of Macedon wanted to provide his son with the best education he could find, he didn't hesitate to summon the greatest philosopher of the time, Aristotle, to teach philosophy, medicine, and art to the boy who would become Alexander the Great. Meanwhile, Roman Emperor Marcus Aurelius accumulated, throughout his childhood and adolescence, a total of seventeen tutors in grammar, rhetoric, law, and philosophy.
Tutoring survived into modern times. Ada Lovelace, the inventor of the first algorithm, was homeschooled by mathematicians and scientists. Philosopher Bertrand Russell and writer Virginia Woolf also never attended a school – at least not as students. Ludwig Wittgenstein only started attending one at the age of fourteen. These are not isolated cases, but examples of the preferred instructional method of the elites at the time.
With the massification of education and the strengthening of democratic norms, tutoring started losing ground, giving way to classrooms full of students. Still, both John von Neumann and Albert Einstein, arguably the two greatest geniuses of the 20th century, were educated by a combination of classrooms and private tutoring. Von Neumann was taught by governesses until he was eleven, when his father insisted that he start attending a school suitable for his age. However, his father agreed to hire private tutors so the prodigy could advance his mathematical studies. Einstein learned algebra with the help of his uncle, Jakob Einstein, and geometry from Max Talmud, the family's tutor.
The role of these aristocratic tutors was not just to cover a set of predetermined competencies. They were responsible for guiding the pupil's studies, stimulating their natural interests and instigating a passion for knowledge.
Neuroscientist Erik Hoel speculates that the decline of this method of education significantly accounts for why we have fewer geniuses today, even with the population explosion and increased access to information. The thesis is quite controversial, but it echoes a known issue in educational psychology, the so-called Bloom's two sigma problem.
In 1984, researcher Benjamin Bloom published a study comparing the efficacy of various educational interventions. The most surprising result was the effect of individual tutoring on educational performance: a student instructed in this manner performed two standard deviations (or two sigmas, hence the name) above the average student. In layman's terms, this means the average student receiving private instruction would perform better than 98% of students in a conventional classroom.
The reason why Bloom called this a problem was clear: individual tutoring is expensive, a method of education for the aristocracies of the past, with no place in current times. And despite his search, the psychologist found no other educational intervention with such a significant impact on school performance.
More recent studies are less extreme but still attest to the superiority of tutoring. A meta-analysis published in 2017 measured an effect of 0.36 standard deviations of performance improvement among tutored students. Although smaller than the effect found by Bloom, it is still much larger and more significant than the effect found for other variables generally considered relevant in education, such as class size, after-school programs, or increased resources for schools. Study after study seems to indicate that private lessons are still the most effective method of instruction we know. But currently, we reserve this method only for struggling students, even among the elite.
The main reason for the disappearance of tutoring is, evidently, the high cost. An education based on dedicating one teacher per child does not scale and could never be implemented in a large education system.
But this is starting to change.
In March of this year, the educational organization Khan Academy launched Khanmigo, an artificial intelligence chatbot that answers questions about school subjects, contextualizes the importance of the material, and even personifies historical figures and fictional characters students can talk to. It is built on the same technology as ChatGPT, but unlike it, never provides answers immediately: its goal is to stimulate the student to think and to arrive at solutions on their own, only guiding the process. Khan Academy also announced an artificial intelligence chatbot for teachers: a pedagogical assistant that helps structure classes and also suggests activities to be done with students.
For those familiar with the reality of classrooms, the proposal may seem illusory. Interaction with these artificial intelligence models requires discipline and proactivity, which many students do not have.
The criticism makes sense. The relationship between student and teacher goes far beyond mere information transmission: human interaction allows for the identification of conceptual misunderstandings and contextualization of the material in a much deeper way. In addition, the affection, inspiration, and motivation that teachers provoke in their students are still far from being replicated by machines. When Bertrand Russell describes his tutors in his autobiography, he does not only talk about what he learned, but also about the admiration that he felt.
If used improperly, artificial intelligence can do more harm than good. The difficulty in ensuring the factual accuracy of the models is still one of the major limitations of the technology, not to mention its indiscriminate use for writing essays or cheating on exams.
However, the reasons to be optimistic seem to be greater than the reasons to be skeptic. For millennia, the best method of instruction we knew was reserved only for the elite. Now, it will be possible to offer to every student a completely individual and dedicated tutor, an expert in all subjects, capable of answering questions and offering help at any moment. This tutor will also help students to organize themselves and will construct study plans adapted to their needs. More than that, these new artificial intelligence models in education have been trained to promote independent reasoning and to spark students' curiosity, as well as to relate the presented material to their personal interests, encouraging students to go beyond what they see in the classroom.
If parents, teachers, and educational institutions adequately seize the opportunity, they will have powerful allies in education. Instead of replacing teachers, these artificial intelligence models will facilitate the most costly part of teaching: instruction in several subjects and answering individual questions. The responsibility of guiding this process will fall to humans, taking into account the emotional and social dimensions of education. Perhaps the promise of an Aristotle for every student is still far from being realized, but with these new technologies, we can start to dream.
Thanks to André Eler, Clara Carvalho, Fernão Ferreira-Reimão, François Boris, Helena Werneck, and Julia Sarquis for reading drafts of this.
After publication, a friend mentioned to me this video, which I had not watched, which touches on the same ideas, including a very similar title.