The experience of apprentices and trainees in the real world is not that different from Luke’s. Take, for example, the junior lawyer who chokes under a pile of papers, or the corporate intern who takes coffee orders, or a series of aspiring chefs who shuck oysters all day. Exhausting as it may seem, this drudgery is often seen as a necessary evil because it is the only way many apprentices can repay their teachers for their training.
“The master can say, ‘Okay, I’ll train you. But in return, I’ll make you work for me for a while and I’ll pay you less than what you produce,” he says. Luis Rayoprofessor of strategy at the Kellogg School and expert in contract theory and organizational economics.
But the recent rise of artificial intelligence has complicated this long-standing dynamic. Large language models such as Claude and Gemini perform many common learner tasks increasingly well, often with much greater efficiency than the average practitioner. Actually, recent research already shows a decline in hiring for entry-level, but not experienced, workers in occupations most exposed to genetic AI.
“Apprenticeship-like systems are a very common way that economies solve the difficult problem of transferring human capital from one generation to the next,” says Rayo. “Artificial intelligence risks destroying this system, and the problem is, who will train the next generation of experts?”
To better understand the possibility of this happening, Rayo partnered with Luis Garicano of the London School of Economics to mathematically model what happens when artificial intelligence enters the apprentice-master relationship.
They found that the more tasks the AI takes over from the apprentice, the less viable the apprentices become. However, they also found that AI could help raise the peak ability and productivity of advanced apprentices, which could increase the value of apprentices and give teachers more incentive to hire one.
“Well, it’s a fight between the two [effects of AI]says Rayo. “One shrinks profits, the other increases profits, and whichever wins determines whether an apprenticeship is profitable.”
This is where AI comes in
Rayo’s interest in this topic stems from his previous research, in which he and his colleagues modeled the traditional apprentice-master relationship. The model has two key assumptions: the apprentice does not have enough money to pay the master for training and cannot simply promise to pay later. Thus, the master wants to extend the training period to get the most value from the apprentice while the training takes place.
Rayo and Garicano, both partners in Center for Economic Policy Researchidentified two key outcomes after introducing artificial intelligence into this model.
First, AI raises the floor for what learners need to do to deliver value. Since AI tools can complete many of the apprentice’s core tasks almost entirely by themselves, the apprentice is no longer needed for menial work. As a result, the apprentice is less valuable to the master.
“Now that AI is doing that [work] essentially free, the currency that learners use to buy knowledge disappears,” says Rayo.
Second, AI raises the ceiling for what an advanced learner can deliver. AI tools can supplement some of the higher-level work that learners do, which increases the value of their work while giving them much more to learn.
“So for high enough levels of knowledge for the learner, AI is good for you. It complements your skills, makes you more productive,” says Rayo. “And if the cap gets extremely high, then there’s still a lot of potential for productivity growth.”
A fight for the ages
Economists use the setting of a law firm to show how this can happen in real life. With artificial intelligence, a junior associate—the apprentice—no longer has such menial work to do, so he is trained to do more advanced work and charge more for it. In other words, the floor is higher. But because junior associates get to do more advanced work, they can complete their training much faster than before. Thus, despite the increased value they initially bring, shorter apprenticeships ultimately reduce the long-term value they bring to business partners. So the more AI can do, the less incentive the company has to bear the cost of hiring and training junior associates.
At the same time, AI is also helping not only senior associates, but also highly trained junior associates to operate more efficiently and charge more than before. There is also much more that junior associates can absorb and do. If this increases their profitability high enough to cover the cost of hiring and training them, then they become even more desirable to the business.
“The question is which grows faster—the floor or the ceiling?” Rayo says.
To answer this question for a particular apprenticeship, a company would need to track how much value a highly trained person creates with the help of AI versus the value that AI creates on its own. According to the model, an apprenticeship is guaranteed to be worthwhile if this ratio is greater than Euler’s ‘e’ number (2.71828), the famous mathematical constant widely used in the world of finance.
A bright future?
The wider the gap between the floor and the ceiling of an apprenticeship, the more valuable the apprenticeship. The smaller the gap, the less value it has.
But if the gap between the ceiling and the floor narrows enough, not just for a single apprenticeship but for the entire workforce, society could face serious consequences.
“Human knowledge would start to disappear, and then robots would do what they are capable of doing on their own, without the added benefit of advanced human knowledge,” says Rayo. “It will be a case where, instead of accumulating knowledge as a society, we start to lose knowledge.”
To deal with this worst-case scenario and keep apprenticeships alive, businesses could require apprentices or trainees to repay the business after completing their training by agreeing to work for lower pay for an extended period of time. Or the government could subsidize apprenticeships through grants or loans. But both of these options come at a significant cost, either to the learner or to society.
A more logical approach could be to adapt and improve the education system. Universities, for example, could help prepare future apprentices and interns to enter the workforce at a high enough level to immediately provide more value to the company than AI alone.
For this to happen, Rayo believes that the education system should teach its students extremely valuable knowledge, especially general principles that are widely applicable to many types of work.
This is the kind of information that “enables humans to perform tasks that AI cannot do on its own,” he says. “The more fundamental principles you learn from different disciplines, the better positioned you are to direct AI and judge the quality of AI output.”
These strategies may be less necessary, however, if AI widens the gap between the floor and the ceiling for apprenticeships as a whole. In this scenario, apprentices would naturally start doing higher skilled work. And they will still have a lot to learn and offer because of how much AI has increased their productivity and expanded the scope of their work. With proper training, for example, a young padawan like Luke could have started his apprenticeship with much more control over his emotions and the Force and ended up developing a greater level of power.
“We’d spend more time doing advanced stuff, and our final level of productivity—the ceiling we reach once we’re fully trained—would continue to rise,” says Rayo. “This is a bright future with extreme levels of productivity and with the transmission of knowledge from one generation to the next.”
