Groundbreaking ideas and research for engaged leaders
Rotman Insights Hub | University of Toronto - Rotman School of Management Groundbreaking ideas and research for engaged leaders
Rotman Insights Hub | University of Toronto - Rotman School of Management

AI could fundamentally change the way systems are built...if we let it

Read time:

Ajay Agrawal,  Avi Goldfarb,  Joshua Gans

When rules have existed for a long time, it can be hard to see the system that they are embedded in. Because rules are reliable, a myriad of rules and procedures can stick together. If something moves, they have to all move at once.

On the free version of music streaming service Pandora, every user receives the same amount of advertising. This is generally true of advertising-supported media. Network television stations used eight minutes every half hour for advertising. This was the rule that drove network revenue. Around this rule a variety of other processes developed. Programs were designed to be twenty-two minutes or forty-four minutes, which meant the writers needed to write every episode of a show at the same length, with natural pauses at the intervals when commercials take place. This rule is affixed to this system.

YouTube provides an example of an alternative system design for content. Unlike network television, YouTube content creators can create content of any length. The system’s AI can predict which viewers will be most attracted to which content. The AI driving the search engine and the recommendation engine enables viewers to find suitable content despite a seemingly infinite catalog of options. Furthermore, the AI can predict which users will be most attracted to which advertisements. Importantly, this prediction capability is much more valuable in a system that allows different users to view different content. Even if network television had an AI that was able to generate similar predictions, the value would be much less because its system forces every viewer to view the same content. So, the best it could do is to predict which advertisement would be most appealing to the most viewers.

In other words, the same AI that predicts viewer attraction to content and advertising is much more valuable in the YouTube system than in the network television system. And while the AI directly enables discovery in a vast catalog of content and enables advertisement matching, it indirectly enables flexible content lengths because the solutions for discovery and advertising solve the problem of an infinite number of combinations of content, advertising, and timings that make flexible content lengths intractable for network television.

In school systems, students in the same grade learn the same things. There is a fixed curriculum. “Students are educated in batches, according to age, as if the most important thing they have in common is their date of manufacture.” For example, in Ontario, where we live, almost all students born in 2009 enter first grade in 2015 and high school in 2023. These rules are in place to manage the uncertainty around which level a given student should be at, for both academic and social reasons. These rules, in turn, glue together in a system: teacher training to manage a limited diversity in learning needs; modest extra help and resources to students who fall behind. And, at the high school level, there are nominal programs for students who do not conform to the standard process for their cohort, including alternative schools, work-study programs, and processes to get high school equivalency certificates.

An AI that predicts the next best learning content for each student would personalize education, allowing students who master a topic quickly to move to something new before they get bored and at the same time allowing students who need more practice on a topic extra time, examples, and exercises to develop competency in that area before moving on. As a point solution, this AI could enhance learning in the existing school system to some extent, although the impact would be limited because once a student completes their grade’s age-based curriculum, they would be done for the year or need to continue any further learning with limited teacher support because teachers are often trained for a specific grade (e.g., grade six math). In the existing system, this problem would become increasingly severe in later grades as the spread between faster and slower learners in a subject area grows over time. To support their students, teachers would need teaching mastery over an increasingly large range of topics.

Imagine, instead, a system where students progress through school as a class (their physical and social development is paced by biology), but many different tutors and teachers come and go to support different students, depending on their individual learning needs. The tutors and teachers that students work with are independent of the students’ ages but are instead determined by the nature of their questions and ability in a subject area. The impact of the AI would be much greater in this new system compared to the impact of the same AI in the existing system because each student could receive an education personalized to their learning needs and style. Students who learn fast in one subject and slow in others could be accommodated. Students who need to focus on particular skills would have teachers who specialize in those areas. The teacher would not need to select the style that helped the most students. Teachers that are great at helping struggling students to read, and those that excel at helping students shine in mathematics competitions, would spend all their time on what they do best.

Rules such as 22-minute programming and the age-based curriculum were put in place to deal with uncertainty. Then, various forms of scaffolding were developed to optimize the performance of the system. While invisible to the casual observer, the rules became the glue that holds the system together. So, introducing an AI that enables a rule to be transformed into a decision might seem attractive at first glance, but its impact may be limited because the rule it’s replacing is tightly coupled with other elements of the system.

Dropping an AI that predicts the next best content into the existing school system would have limited impact because the age-based curriculum rule with a single teacher per class is a cornerstone of the current educational system, especially in elementary school. In contrast, using exactly the same AI, but embedding it in a new system designed to leverage the AI’s personalized content and pacing by coupling it with personalized discussion, group projects, and teacher support, which would require much more flexible tutor and teacher allocations and modified educator training, would likely result in a much bigger impact on education and personal growth and development.

In other words, the age-based curriculum rule is the glue that holds together much of the modern education system, and so an AI that personalizes learning content can only provide limited benefit in that system. The primary challenge for unleashing the potential of a personalized education AI is not building the prediction model, but unsticking education from the age-based curriculum rule that currently glues the system together.

That’s why it’s hard to replace a single rule with an AI-enabled decision. Thus, it’s often the case that a very powerful AI only adds marginal value because it is introduced into a system where many parts were designed to accommodate the rule and resist change. They are interdependent — glued together.

This excerpt has been condensed from the new book Power and Prediction: The Disruptive Economics and Artificial Intelligence by Ajay Agrawal, Avi Goldfarb and Joshua Gans. 


Ajay AgrawalAjay Agrawal is the Geoffrey Taber chair in entrepreneurship and innovation and a professor of entrepreneurship at the University of Toronto's Rotman School of Management.
Joshua GansJoshua Gans holds the Jeffrey S. Skoll chair in technical innovation and entrepreneurship and is a professor at Rotman.
Avi Goldfarb is the Rotman chair in artificial intelligence and healthcare and a professor of marketing at the Rotman School of Management. He is also the chief data scientist at the Creative Destruction Lab.