Research on artificial intelligence in education is large but uneven. A 2019 systematic review of 146 articles on AI in higher education found that only nine first authors worked in education departments and that two of the 146 articles critically discussed ethics or risks. A 2024 review of 66 earlier reviews found that 40.9 percent named a need for more ethical attention and that 45.5 percent had done no quality assessment of the studies they summarized.
Key takeaways
- Zawacki-Richter and colleagues (2019) screened 2,656 publications from 2007 to 2018 and synthesized 146; computer science supplied 61 first authors and education departments nine.
- Bond and colleagues (2024) analyzed 66 reviews published between 2018 and July 2023, and 40.9 percent of them called for more attention to ethics.
- Most primary studies in the 2019 review were quantitative (73.3 percent), while one study was qualitative.
- Both reviews cover higher education only, so they say little directly about K-12 schools or about recent generative AI tools.
- A reader can check any AI-in-education paper against a short list: who wrote it, what was compared, how long it ran, and what the authors admit.
What the 2019 review mapped
The first large map of the field came from Olaf Zawacki-Richter, Victoria Marín, Melissa Bond, and Franziska Gouverneur, in the International Journal of Educational Technology in Higher Education, published October 28, 2019. They searched for work from 2007 through 2018, identified 2,656 publications, and kept 146 under stated inclusion rules. The articles fell into four application areas: profiling and prediction, assessment and evaluation, adaptive systems and personalization, and intelligent tutoring systems.
The authorship pattern gave the review its subtitle, "where are the educators?" Computer science departments supplied 61 first authors and STEM departments 29, while nine first authors came from education departments. Counting dual affiliations, 13 papers (8.9 percent) had a first author with an education background. Half of all articles came from four countries: the United States, China, Taiwan, and Turkey.
What the 2019 review found missing
Of the 146 articles, 73.3 percent used quantitative methods, 5.5 percent mixed methods, and 0.7 percent (one study) qualitative methods, and 20.5 percent were theoretical or descriptive. Only five articles, 3.4 percent, gave an explicit definition of artificial intelligence. Only two, 1.4 percent, reflected critically on ethical implications, challenges, and risks; one of them raised student privacy in agent-based personalized learning, and another raised the cost and time of building such systems. The authors concluded that there was almost no critical reflection on risks and a weak link to pedagogical theory.
What the 2024 review of reviews added
By 2023 the literature had grown enough that researchers began reviewing the reviews. Melissa Bond and eight co-authors published a meta systematic review in the same journal on January 19, 2024. It covered 66 reviews published in English between 2018 and July 2023, drawn from seven databases plus snowballing. Of these, 66.7 percent were systematic reviews, 89.4 percent were written by teams, and 71.2 percent of the collaborations were domestic only. Authors from North America wrote 27.3 percent.
The reviews mostly addressed AI in higher education in general (47.0 percent) or profiling and prediction (28.8 percent). The authors judged that adaptive systems and personalization dominated the findings. The most common research gap, named in 40.9 percent of reviews, was a need for more ethical attention, followed by more methodological rigor, named in 36.4 percent.
The quality of the reviews themselves was uneven. Among the 66, 31.8 percent searched only one or two databases, 51.5 percent reported nothing about how reviewers settled coding disagreements, 45.5 percent did no quality assessment of the included studies, and 34.8 percent did not reflect on their own limitations. Only 24.2 percent shared their exact coding scheme.
The two reviews side by side
| Feature | Zawacki-Richter et al., 2019 | Bond et al., 2024 |
|---|---|---|
| What was reviewed | Primary studies, 2007 to 2018 | Reviews, 2018 to July 2023 |
| Items synthesized | 146 of 2,656 screened | 66 reviews |
| Setting covered | Higher education | Higher and continuing education |
| Ethics | Two of 146 articles reflected critically (1.4 percent) | 40.9 percent of reviews called for more ethical attention |
| Main caution | Few educators as authors, little theory | Uneven review quality, mostly domestic teams |
Neither review measures how much students learn. They describe who studies AI in education, with what methods, and with what gaps. Studies that do measure learning, such as randomized trials, are covered in other articles on this site.
A checklist for reading an AI-in-education paper
- Who wrote it, and does the team include teachers or education researchers?
- Is the setting a school, a university course, or a lab exercise, and in which country?
- Was there a comparison group, and were students assigned at random?
- How long did the study run, and was learning measured after the tool was removed?
- Does the paper define the AI tool, and does it discuss privacy, bias, or cost?
- Do the authors state limits, and does the review they cite report its own quality checks?
Both reviews predate most classroom use of generative chatbots, and Bond and colleagues stop in July 2023. Newer studies may differ in methods and quality, and neither review reports study results separately for Canada or the United States, so country-specific findings cannot be drawn from them.
This article summarizes published research for general information. It is not professional advice on adopting any tool.
FAQ
Does the research show AI improves learning?
These two reviews do not answer that. They describe the literature, not learning outcomes. Controlled trials answer it for specific tools and settings.
Why do computer scientists write most of the papers?
The 2019 review counted first-author departments, not reasons. Its authors read the pattern as a sign that educational perspectives are needed in the field.
Are the reviews still current?
They cover work through 2018 and through July 2023. Later research, including studies of generative AI in classrooms, falls outside both.
Do the reviews cover K-12 schools?
No. Both focus on higher education, so the findings are only an indirect guide for elementary and secondary schools.
Related reading
- problems in AI research
- current trends in AI for education
- dropout prediction as a case study
- AI in scientific research
More in AI in education.
Last verified: October 2026
Sources
- Zawacki-Richter O, Marín VI, Bond M, Gouverneur F, Systematic review of research on artificial intelligence applications in higher education: where are the educators?, International Journal of Educational Technology in Higher Education 16, article 39, 2019
- Bond M, Khosravi H, De Laat M, Bergdahl N, Negrea V, Oxley E, Pham P, Chong SW, Siemens G, A meta systematic review of artificial intelligence in higher education: a call for increased ethics, collaboration, and rigour, International Journal of Educational Technology in Higher Education 21, article 4, 2024