All over the world, institutions of higher education are making greater use of AI and
ML software for the admission of students. Schools have started using AI tools as
applications are becoming more voluminous and competitive, and they also want to
be more productive. Now, AI tools can scan applications, assess likelihood of
enrolment, evaluate essays, customize recruitment outreach and more. The global
governance of access to academia has become algorithmically scaled rather than
humanly discretized. Ever since 2019, academic literature on usage of AI in
education has been on the rise. Put differently, in one of the earliest systematic
reviews, Zawacki-Richter et al. (2019) examined AI uses throughout the student
lifecycle. They catalogue usages for profiling, predicting, tutoring, etc., adaptive
learning in that context. An increasingly popular area of research is the use of
models to support and optimize admission decisions. According to a review that
primarily utilized secondary literature from 2018 to 2023, profiling, prediction and
related applications, including admission-related predictions, is a core domain of
application of AI in higher education research, Bond et al. (2024). Li et al. (2023)
conducted a mapping of the analytical landscape of predictive learning analytics in
student admissions using 59 Scopus indexed articles. The most popular modeling
techniques include logistic regression, decision trees, random forests, support
vector machines and neural networks. The implementation of AI is raising concerns
more and more in society. Research into algorithmic bias shows that models
trained on data shaped by decades of structural discrimination can exacerbate
historical injustices. The review was published in the International Journal of
Artificial Intelligence in Education. Baker and Hawn noted that there are algorithmic
biases in education across several demographic aspects. These can be seen in race,
gender, nation and social class. According to research by Gandara and others
conducted in 2024 using data collected from across the country, models predicting a
student’s success at university level were less accurate for Black and Hispanic
students. This, they say, produces a ‘double disadvantage’ through the predictions
which under-predicts success and over-predicts failure. AI admission systems may
help in managing admissions more easily. However, nonetheless, they come with a
set of serious questions about whether we are paying a price for equity. The ethical
stakes of many deployed systems rise due to opacity. A systematic review of the
literature on AI and higher education recently published in a journal Computer and
Education: Artificial Intelligence, Memarian and Doleck (2023) found that most of the
studies improperly operationalise FATE (fairness, accountability, transparency and
ethics). Laws like the EU's General Data Protection Regulation, or GDPR as it is
popularly known, and the proposed EU Artificial Intelligence Act place transparency
obligations on high-risk algorithmic systems. However, compliance in educational
settings has not exactly been universal (Memarian & Doleck, 2023). The review will
address the following three guiding questions: (1) What AI technologies are
currently being used in the university admission process and how do they work?
What do witness statements say regarding the effects of AI-augmented admissions
on intake volumes and on institutional diversity? What structures and processes
have been proposed to ensure the equity and the accountability of AI admission
systems? The remaining sections of the paper are arranged as follows. Section 2
describes the process and procedures for conducting the review. Section 3
examines how the AI approval system has been designed. The evidence on
outcomes of enrolment is assessed in Section 4. Section 5 examines diversity and
bias in the existing literature. Ethical and governance frameworks are discussed in
Section 6. In the end, Section 7 ends with a recommendation.