Machine learning offers schools, colleges, universities and the companies who provide digital services to the education sector with an opportunity to improve personalised and contextualised learning to students. In this article I will explore how machine learning can enhance the management of differentiated and adaptive learning; and the management of the student life cycle. I will also examine some of the challenges that arise from the use of machine learning.
Differentiated and adaptive learning
The ability to differentiate and adapt learning and assessment materials is founded on the multiple connections that a personal learning environment has with the datasets and services beyond itself; and its ability to make sense of the millions of pieces of information that are held by the educational establishment before making timely, accurate and contextualised decisions that support each student. Imagine logging into a personal learning environment and being presented with learning and assessment materials that appear to be curated specifically for you and no one else. How does it adapt its content to suit your learning support needs? When you sit beside your friend in the library and open the same online tutorial why does it offer you different content and questions from your friend? Observations such as these and many more are becoming common place as students login into their personal learning environments in schools, colleges and universities around the world. The one aspect that always stands out for me is the effortless manner by which the personal learning environment processes millions of pieces of information in a fraction of a second before presenting contextualised learning and assessment materials to each student. What makes it contentious is that it may do it better than the teacher. I say contentious because professionals across the education sector have yet to debate if machine learning is proven to enhance the quality of teaching, learning and assessment and if it can raise student outcomes.
With supervised machine learning the decisions that are made by a personal learning environment are hard wired into it because a team of educational professionals have told the algorithms to follow certain actions when particular conditions are met. However, what happens when the personal learning environment begins to take advantage of unsupervised machine learning techniques to deliver personalised and contextualised learning and assessment materials to the hundreds or thousands or to the tens of thousands of students that you have on your campus? When personal learning environments use unsupervised machine learning techniques to distribute learning and assessment materials teachers and academic support teams are unlikely to know how or why the personal learning environment has made its decisions. Please note that a personal learning environment will use unsupervised machine learning techniques to identify clusters or patterns in the dataset which are hard to detect or identify by teachers and academic support teams.
Ethics - When the wider education sector starts to take advantage of machine learning techniques to enhance teaching, learning and assessment teachers and academic support teams will have to hard code their ethics into their personal learning environments. The coding of ethics into everyday products and services will be challenging. Here is a set of hypothetical examples that could play out in schools, colleges and universities:
- If a student is under-performing present the student with a learning path that will take her to a grade C. If a student is doing well, present him with a learning path that will take him to a grade A.
- Male and female students learn in different ways. If this is the case then present each group with learning and assessment materials that reflect their learning traits.
- A group of students has been been identified as historically under-performing. If a new student applicant has the characteristics of this group don't enroll that applicant onto courses x, y and z because there is a high probability that the individual will not complete their studies.
Educational institutions must be prepared to be open and transparent about the algorithms that they use. Students and parents are entitled to ask schools, colleges and universities about the algorithms that are used to make decisions regarding their application to courses, access to academic support, their choice of progression routes and the learning pathways that are assigned to students.
The following Educause video describes the main elements of adaptive learning and how it can be used by students, teachers and support teams.
In the following Educause video academics and education professionals define their perception of personalised learning. It is important to recognise that machine learning and adaptive learning should never be designed to displace the teacher. Rather, the technology should be used to empower the teacher to deliver personalised instruction, learning and assessment.
The management of the student life cycle
Schools, colleges and universities have teams of people who are dedicated to support every aspect of the student life cycle. The student life cycle begins with an initial inquiry to the institution about the courses on offer, the application for a place, enrollment, induction, study and support during the course, assessment and ultimately graduation from the course. The knowledge and expertise needed to support each element of the student life cycle rests with individuals and teams within the institution. Teams will take advantage of information systems that allow them to query courses, to help them manage the enrollment process, to distribute learning and assessment resources via a learning management system or to manage the assessment process at the end of a course. Now imagine a world where virtual machines have more knowledge about each aspect of the student life cycle than individuals and teams within the institution. This sounds far fetched but note that the private sector has been using virtual machines for years to determine whether or not to grant you a personal loan, mortgage, an insurance policy or the choice of what to purchase online.
The elements of the student life cycle that could benefit immediately from the use of machine learning are recruitment, selection, induction, academic support and pastoral support during the course.
Promotion and Recruitment - The advent of short courses, top-up courses, just in time learning and the opportunity for students to study at a time, pace and place that suits them means that the traditional college and university website is not suited to the task. The use of machine learning will enable institutions to offer information, advice and guidance to potential students that is informed and relevant. For example, an accountant may visit and login to a college or university website wishing to find a short course to top-up her knowledge on tax law. The virtual machines behind the website recognise the visitor and recommend appropriate courses to her; courses that are at the right level and at times and dates that suit the accountant. The website may present courses that the accountant had not considered before. This information is not presented randomly to the accountant. In this case, the virtual machines behind the website has learnt that accountants with similar profiles have tended to do certain courses and it is these courses that are presented this new visitor to the institution's website.
Selection - The decision to accept or reject applicants to a course can be difficult at times. A course team will ask themselves if the student has the ability or propensity to benefit and succeed on their course. This will be repeated for every applicant to the course. The use of machine learning could support course teams in a number of ways. For instance, the virtual machines employed by the institution have access to a multitude of records that pertain to every applicant to the course. Anonymised records could go back many years. In this instance, a virtual machine has used machine learning to analyse the anonymised records of previous students who closely match the characteristics of the present applicant before presenting the course team with a probability that pertains to the applicant's likely success or failure on the course. The likelihood of success or failure is backed up with a dashboard of data. It this instance, the course team makes the final decision of either accepting or rejecting the application. However, what if you belonged to a course team that received hundreds or thousands of applicants to a course that only had 50 places? Would you welcome the opportunity for machine learning to shortlist potential applicants to your course? The parameters that are used to either select or reject applications need to be carefully considered and institution's need to be wary of breaking legislation pertaining to age, ethnicity, gender and more.
Induction - The logistics of managing an induction programme for new students are very similar wherever you are. Information pertaining to the wider campus needs to be cascaded to new students as well as specific information pertaining to the course that they are embarking upon. The picture is complicated by the individual needs and requirements of each student. For example, students will have specific learning support needs, requirements for specific medical support, if they will reside on or off the campus, fee schedules, information pertaining to part-time and full-time students and many more. Institutions will typically invite students to talks on a whole manner of subjects that pertain to specific services or they will publish printed or online material that is presented to students regardless of context. The volume of information relating to induction is also a factor to consider. The complexity of induction lends itself perfectly to machine learning and it offers an opportunity for the whole institution, its departments, course and academic support teams to improve the induction programme for its new students. Machine learning will enable institutions to present timely, relevant and contextualised information to each student. Virtual machines will query a given student's dataset and gather information that correlates with the needs and requirements of the student.
Academic and Pastoral Support - Students are supported through their studies by a large and varied team of people. Individuals will typically meet up on a regular basis to discuss the needs of individual students on a course or as and when required. Teams will also use information systems and work-flows to manage the needs and requirements of individual students. The use of machine learning to offer differentiated and adaptive learning and assessment materials has already been discussed. However, how can machine learning be used to improve the success of a student during their studies? An increasing number of institutions are using predictive analytics to ascertain the students who are likely to fall short in their studies. Variables such as entrance qualifications, attendance data; actual, targeted and predicted grades; the number of visits to the personal learning environment; communication between students and staff; and even historical data are all factored into a department's student at risk register. Machine learning is well suited to the task of identifying those students who could be at risk because the virtual machines that are deployed by the institution continuously monitor the progress of each student.
In the following video Richard Culatta describes how digital technologies can be used to enhance personalised learning.
The use of machine learning marks a significant milestone within the education technology paradigm because it introduces a new agent into the classroom. There are numerous reasons why this aspect of machine learning is likely to remain contentious for years and decades to come. The primary reason centres on the personal learning environment's ability to make decisions autonomously from the teacher; such as determining the curation and distribution of learning and assessment materials to students across the campus. The algorithms that are used by the personal learning environment could become so complex that the teacher is unable to ascertain why his or her students were being presented with specific learning and assessment materials. The caution, hesitation or reservation that is shown by teachers in these circumstances is only natural and it can only be eased by providing evidence that machine learning does indeed raise student attainment.
The development of machine learning within the education sector will be broad ranging. Here are a few user case scenarios:
- As natural language processing improves the personal learning environment will not only collect work from students but it will also mark, grade and offer feedback to each student. Companies such as Turnitin are leading the way in this field.
- The use of natural language will have many uses beyond marking and grading. For instance students will be able to query the semantic capabilities of the personal learning environment for any matter relating to their studies. For example: 'when is my next assignment due in?', 'are there any resources in the library that can support my current project?', 'who else in my class is working on the same task as I am?' and so on.
- The development of natural language processing will offer personal learning environments with the tools to curate content that is bespoke and contextualised to each student. It will be the first time that learning and assessment materials will be curated and distributed by an agent other than the teacher, teaching assistant or authors employed by educational publishers.
The successful use of personal learning environments and machine learning tools will depend on teachers acknowledging the contribution that such technology can make to enhance teaching, learning and assessment. If it enables teachers to better support their students, if it enables students to raise their grades and if it enables students to progress onto further education, training or employment then machine learning will have proven itself.