The field grew, the programme did not: differential growth and substrate portability in artificial intelligence and education research, 2016–2026
This study challenges the misconception that declining publication shares indicate field decline by demonstrating that while the overall AI in education literature grew fivefold from 2016 to 2026, specific subfields experienced differential growth rates driven by "substrate portability"—the adaptability of research constructs to new data environments—rather than functional obsolescence.
Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer
For decades, researchers have tried to use computers to understand how people learn. They built systems to track student behavior, predict who might struggle, and adjust lessons in real time. This work relied heavily on digital platforms that recorded every click and keystroke, creating vast trails of data to analyze. Around 2022, a new kind of technology arrived: large language models capable of generating text, answering questions, and creating content instantly. This shift captured the imagination of educators and researchers alike, causing a massive surge in new studies. But as the volume of research exploded, a confusing pattern emerged. The share of papers dedicated to the older, data-heavy methods seemed to vanish, leading many to believe those fields had died out. The question remained: had the old ways of studying learning been abandoned, or had they simply been drowned out by the sheer size of the new wave?
A team of researchers set out to answer this by looking at the actual number of studies published, not just their proportion. They gathered nearly 1,200 academic papers from 2016 to mid-2026, splitting the timeline into two eras: before the rise of generative artificial intelligence and after. Their goal was to separate the composition of the literature from the actual activity within it. Imagine a city that grows from 100 people to 1,000. If a specific neighborhood, which used to house 30 people, stays at 30 people, its share of the city drops from 30% to just 3%. To an observer looking only at percentages, that neighborhood seems to have shrunk or disappeared. But in reality, the number of people living there has not changed at all; the city simply got much bigger. The researchers applied this same logic to the field of artificial intelligence in education. They found that while the share of papers on older topics like educational data mining and learning analytics did indeed fall dramatically, the actual number of papers on these subjects did not always disappear. In fact, for some topics, the number of studies remained steady or even grew, just not as fast as the new generative topics.
The study revealed a story of differential growth rather than total replacement. The field as a whole expanded more than five times over the decade. The older research programs, which depended on collecting long-term behavioral data from students, did not grow at that same furious pace. Two specific areas, educational data mining and learning analytics, saw a genuine drop in the number of new papers published each year. Another area, intelligent tutoring systems, held its ground, publishing roughly the same number of papers as before, even though its share of the total field shrank significantly. Meanwhile, new topics like academic integrity, ethics, and the use of generative tools for creating course materials exploded in popularity, growing far faster than the field's average. The researchers discovered that the survival of a research topic depended less on what it was called and more on what it required to exist. Topics that needed complex, instrumented platforms to collect years of student data struggled to keep up. Topics that could be explored with simple text or immediate interaction, such as discussing ethics or generating lesson plans, thrived.
This distinction matters deeply for anyone trying to understand the current state of the field. A new researcher entering the field today might look at the list of popular topics and conclude that the ability to analyze student data has been abandoned. However, the data shows that this capability is still being developed, just at a slower, more steady rate than the hype surrounding new tools. The older work has not been erased; it has been outgrown. The study suggests that the shift is not a failure of the old methods but a natural consequence of a new technology that is instantly available to everyone, creating a flood of new questions that do not require the same heavy infrastructure to answer. For institutions or educators looking to adopt these technologies, the lesson is clear: the most visible topics are not the only ones that exist. The foundational work of understanding learning through data continues, even if it no longer dominates the headlines. The field has not lost its past; it has simply expanded around it, carrying both the old and the new forward, but at very different speeds.
Drowning in papers in your field?
Get daily digests of the most novel papers matching your research keywords — with technical summaries, in your language.