'Urgent question': University of Limerick researchers question future of homework in AI age

Short, regular homework can improve maths and science performance, but the growing use of artificial intelligence by students could undermine its benefits, new University of Limerick research suggests.

The research – conducted by Dr Darragh Flannery and Professor Dónal Palcic from UL’s Kemmy Business School, in collaboration with Dr José G. Clavel from the University of Murcia, Spain – found a “small but significant relationship” between the allocation of homework and test scores, with one additional hour of homework assigned to students per week found to be roughly half as effective as an additional hour of in-class instruction time found in other prominent studies.

The study, published in the Oxford Bulletin of Economics and Statistics, analysed data from almost 100,000 eighth grade (13- to 14-year-old) students across 23 countries – including Ireland – drawn from the Trends in International Mathematics and Science Study (TIMSS) 2019 dataset.

The research team compared each student against themselves, analysing their performance in the subject in which they were assigned more homework against the subject in which they were assigned less. They then looked at homework through two lenses: how often it was assigned, and how long each assignment took to complete. Noting the increase in test scores relative to the assignment of additional homework, the authors highlighted the significance of their findings.

“They show that in some countries, on average, homework can be used as a low-cost tool (from a financial perspective) to increase the test scores of secondary school students; a finding that has not been generally true for children in lower levels of education.”

The study showed that the benefit of the additional homework allocation for students was largely driven by more frequent allocations rather than increased volume of work, with those students given short, frequent homework (three to five times a week, taking up to half-an-hour each time) scoring higher than those given no homework at all. The researchers did note, however, that this positive association didn’t follow through for the weaker performing students.

“A teacher who sets a quick homework task after most lessons appears to be doing more for their students than one who occasionally assigns a long, involved piece of work – even if the total number of hours is similar. This lines up with decades of cognitive science on ‘spaced repetition’; recalling and applying material in short bursts, spread over time, builds a stronger memory trace than cramming the same amount of practice into one sitting,” the researchers explained.

The team also found the positive link between additional homework and improved test scores to be stronger where students are being educated by more experienced teachers, and weaker where teachers have adopted a more modern teaching style.

“Our results suggest that more modern styles of teaching do not lend themselves towards complementing the assignment of homework in improving student test scores. This may help inform discussions around teacher and school practices while contributing to the wider debate around modern versus traditional teaching styles,” they wrote.

While acknowledging the need for a more complete understanding of the relationship between homework and student achievement, the researchers also highlighted the “urgent question” of how increased reliance on AI tools by students may impact the value of homework in education going forward.

“The mechanism we identify – short, frequent tasks that force students to retrieve and apply material themselves – depends on homework actually being done independently,” they explained.

“If pupils increasingly turn to AI to generate answers rather than work through problems on their own, the retrieval-practice benefit we document could weaken or disappear entirely, regardless of how homework is structured. Whether that is happening, and how schools might design homework and assessment to preserve its value in an AI-saturated environment, is now an urgent question for researchers to answer with more recent data – ours included.”