Introduction

The use of machine learning within the education sector provides schools, colleges and universities with multiple opportunities to enhance and transform the heart of their services such as teaching, learning, assessment and student support. This article seeks to expand on my previous notes on machine learning by providing a number of user case scenarios for each of the machine learning agents that could be employed by institutions on their personal learning environments. [1]

Context
The use of machine learning within the education sector is based on two simple premises. The first premise states that schools, colleges and universities are essentially information-processing entities. [2] The second premise states that student outcomes can only be raised if practitioners use intervention and support strategies that are evidence based. Educational institutions who are skilled at acquiring, creating and organising information and whose employees are able to share the knowledge that transpires from it are better placed to support their students. Better still, they can also generate new knowledge which can be applied to the design of new products and services, enhance existing services and improve organisational processes to better support the student. [3]

What is the role of machine learning within the education sector?
The complexity of structured and unstructured data within schools, colleges and universities and the desire to understand, manage and use it to improve the outcomes of all students is one of the main reasons why machine learning is proving to be a fertile area for institutions and the companies that provide services to them. Machine learning agents play a critical role in this arena. Within the context of personal learning environments, agents can be described as programs that observe the behaviour of students, teachers and support teams; and they carry out activities which enable them to extract meaning and knowledge from these interactions before directing or combining their activities to satisfy the needs of students, teachers and support teams.

Machine Learning Agents
Numerous machine learning agents can be put to use to help teachers and student support teams to raise student outcomes in their institutions. In the following section I describe the role and function of the primary machine learning agents that can be used by schools, colleges and universities and I provide examples of how they can be used to enhance teaching, learning, assessment and student support.

The Instructional Design Agent
This agent reviews a student's profile and compiles instructional and assessment materials on a given topic on the course. The agent uses a wide ranging dataset on the student such the academic level of the course being studied, grades scored thus far on the course, predicted grades, targets, gender, learning support needs and more. The use of the instructional design agent enables the student to receive timely, personalised and contextualised learning and assessment materials. The design of the adaptive learning environment enables the instructional design agent to deliver learning and assessment materials to students with or without the input of the teacher. Here are examples of how the instructional design agent can be put to use by course teams:

• As students complete an initial diagnostic test as part of their studies the results are captured and used to inform the initial online tutorials that are presented to each student. This ensures that students are presented with learning and assessment materials that reflect their current academic ability at the start of the course.
• If a student's goal is to achieve a distinction grade or an A grade for a particular component of their course the instructional design agent can tailor content and activities that support the development of knowledge and skills that will enable the student to achieve the higher grade.
• All schools, colleges and universities offer an induction programme for new students. At Bolton College, an instructional design agent was used to help deliver contextualised content to students during their induction programme. When students accessed the online tutorial on health and safety construction students were presented with information, advice and guidance that was pertinent to their department. Likewise, when the College's hairdressing students accessed the same health and safety tutorial they were presented with content that reflected health and safety guidance for the College's hair and beauty salons.
• In the following example colleagues at Bolton College took advantage of the instructional design agent's ability to provide differentiated content to students. The self-employment tutorial has an activity that assesses the student's aptitude to being self-employed. Colleagues felt that the use of strong role model images within the tutorial was important. The tutorial presents female students with female role models in each of the tutorial slides. Likewise, male students are presented with male role models in their slides. The use of the gender variable by the instructional design agent demonstrated how learning and assessment materials can be tailoured to suit any number of contexts and scenarios.

In the self-employment aptitude test the instructional design agent makes use of the gender variable to present the following slide to female students.

The instructional design agent makes use of the gender variable to present the same slide but to male students.

• A large and growing library of online content allows course teams to be much more selective about the learning and assessment materials that they wish to present to each student. For example, the tutorial that is presented to a student could reflect his or her preferred learning style. Students who prefer to learn by watching videos or animations will be presented with content that is rich in video content. A student who prefers to absorb text could be presented with content that is text rich. This scenario assumes that course teams have access to a large library of content where there are multiple iterations for each tutorial or assessment activity for their course.
• A large content library can also provide other affordances to course teams. For example, the instructional design agent can be used to present learning and assessment materials that have a higher weighting; a weighting that reflects previous successes in the use of the tutorial by students. A tutorial which enabled students to achieve high scores in a subsequent test would have a high weighting. A tutorial were students existed the tutorial early would carry a low rating and so on and so forth.
• At the present moment in time teachers, subject experts and instructional designers work together to author learning and assessment materials. The composition of any given tutorial; that is - its introduction, the list of topics that are to be covered in the tutorial, subject matter, videos, images, assessment activities and re-cap slides are all decided by the teacher or the course team. What if each element of a tutorial could be compartmentalised, referenced, tagged and assigned a weighting? The instructional design agent could then be given the task of composing a tutorial from the hundreds or thousands or tens of thousands of discrete items in its content repository or database. The advantage of this mode of delivery is that it makes the authoring of learning and assessment materials completely fluid and it allows the instructional design agent to respond instantaneously to the meet the needs and demands of each student.

The Assessment Agent
The assessment agent monitors coursework grades, initial diagnostic test results, interim test results, end of unit test results, qualifications at the start of the course, exam dates and more to compile assessment materials before presenting them to each student on the personal learning environment. The agent can utilise information such as actual grades, target grades and predicted grades to help it compile appropriate tests before presenting them to the student. The agent is designed to support and prepare students for formal exams and tests and to maximise the grades achieved by students. Like all other agents that are utilised by the personal learning environment, the assessment agent will not act in isolation. For instance, it will exchange information to and from other agents to inform its behaviour and to inform the behaviour of other agents. Here are examples of how the assessment agent can be put to use by course teams:

• In the following example a student needs to achieve a distinction grade on her current course in order to meet the entry requirements for her chosen university. If the student is performing at a level below a distinction grade average the instructional design and assessment agents will work in unison to deliver learning and assessment materials that will support the student to gain the desired distinction grade.
• You will be familiar with the most common forms of online assessment activities such as the use of multiple choice questions, drag and drop exercises, matching exercises and short form free text questions. In recent years learning management systems have become better at differentiating and adapting the types of questions and activities that are presented to each student by using real-time student performance data.
  • For example, if a student has answered a question incorrectly the student could then be presented with a similar question or with information, advice and guidance on how to answer these types of questions.
  • The questions or assessment activities that are presented to the student can also be determined by how the student performed on a previous assessment, assignment, or diagnostic test.
• Assessment agents can take advantage of other information or knowledge about a student. If a student is performing below expected levels it would be unwise to present assessment activities that are at a higher academic level to the student. The assessment agent will take current academic performance into account before presenting activities to the student. Similarly, the assessment agent will use information regarding expected, targeted or predicted grades to deliver assessment activities that progressively enable the student to achieve the higher grade over a period of time.
• Some subjects such as English or maths are delivered to large numbers of students. For these subjects it makes sense to offer contextualised assessment activities to the student. For example construction, civil engineering, hairdressing or art students would be presented with assessment activities that reflect their vocational settings.
• The following video from Turnitin demonstrates the use of natural language processing and assessment agents to provide formative feedback through medal and mission. In this instance the assessment agent (e-rater) acts on behalf of the teacher to provide immediate feedback to the student which has the potential to raise the quality of student work that is submitted to the teacher.

Academic and Pastoral Support Agents
Academic support agents and pastoral support agents are designed to support and enhance the role of traditional student support teams. These agents take advantage of their ability to operate at scale. They are particularly useful when a large number of students require support with their academic studies or with pastoral matters. This group of agents review a wide variety of datasets and knowledge sources on any given student and act in a manner that reflects the overall profile of the student. Here are examples of how the academic and pastoral support agents can be put to use:

• The academic support agent can work closely with the assessment agent to deliver additional support to students. If a student has paused or is seen to be hesitating during an assessment activity the academic support agent may ask the student if he or she requires help. If the student replies yes, the student will be presented with information, advice or guidance on the topic.
• As mentioned earlier, if a student is falling short of achieving the grades required to get into university the academic support agent can adapt tutorials and assessment materials on the course that will support the student to achieve a higher a grade profile on his or her course.
• The agent can also prompt the student to schedule an appointment to see the course tutor or the agent may schedule the appointment on behalf of the student and tutor.
• The agent will also monitor if students are failing to communicate on a regular basis with their teachers. The reason for monitoring the communication channels between students and teachers is to identify those students who are at risk of not succeeding on their courses. Algorithms make use of the hypothesis that those students who do not communicate regularly with teachers are more likely to withdraw from their studies or they may not succeed in getting their desired grades on the course. The agent can also monitor what is being communicated between student and teacher; and act accordingly.
• Academic support agents are particularly useful for identifying students who are likely to perform poorly with their studies. The agent can identify this group of students in two ways. Firstly, the academic support agent can review real-time data to determine the progress made by any given student on a course. Secondly, the academic support agent can examine historical data or use profiling information to ascertain the probability that a student has a high chance of not succeeding on a course. The academic support agent can then alert members of the course team or academic support staff about the status of students; particularly those that have been identified at risk of not succeeding with their studies.
• In the following video Ashok Goel shares how he believes he can use artificial intelligence to scale personalised learning. He used AI to build Jill Watson, an AI teaching assistant. Over time the use of these digital assistants will become broader to encompass any student enquiry about the school, college or university.

The Library Agent
The library agent reviews the schedule of classes that a student is timetabled for. The agent references the topic of each class or tutorial, the reading list associated with the class or module and the topics associated with the coursework or assignment currently being undertaken by the student. The library agent will then correlate this information with the catalogue of books, journals and online resources that are indexed in the library management system that is licensed to the school, college or university. As students login into their personal learning environment they will be presented with timely and contextualised resources, suggested articles to read and even the availability of books or journals that can be loaned out to them.

The role of the teacher in a machine learning environment
Machine learning agents will invariably alter some of the roles and functions of the teacher. I prefer to see machine learning agents as an extension of the teacher who act on behalf of the teacher to aid and support students. This is especially the case if course teams are given the opportunity to inform and guide instructional designers, data scientists and systems developers who develop and deploy the algorithms to the institution's personal learning environment. If course teams are actively involved in the design of machine learning agents these agents can be set to work with a degree of autonomy. However, safeguards need to be put into place in order to ensure that agents in a multi-agent environment behave in the manner that they were designed to. For example, if the primary goals of an assessment agent are too narrow the assessment agent may deliver easy questions to students in order to achieve a 100% pass rate on a given course. Here are some examples of how machine learning could impact on teachers and course teams:

• If the instructional design and assessment agents deliver timely, differentiated, adaptive and contextualised learning and assessment materials to the student on scale (by that I mean the lion share of a course), it will enable course teams to focus on developing the higher order skillset in students; namely a student's ability to analyse, evaluate, critique and to problem solve; the skillset required to work in the modern knowledge economy.
• For some course teams the use of machine learning agents will enable them to accelerate instruction, learning and assessment. Courses that previously took a whole academic year to complete could be done in half the time. Funding agencies and institutions will need to work together to accommodate for this change.
• The volume of data that teachers and course teams manage is increasing; with colleagues expressing concerns about their capacity to effectively manage this rising tide of data. The advent of machine learning will mitigate these concerns and it will enhance the ability of course teams to make informed decisions about their students.
• The introduction of a new, active and semi-autonomous agent in the classroom will be something that teachers and course teams will need to get used to. In a classroom with 20 or more students, were students have mixed abilities, needs and requirements; and where students are able to work at their own pace the machine learning agents will be play an important role in the management of the course - particularly in the delivery of course materials, assessment, tracking and monitoring of student progress. As instructional design agents and assessment agents develop further and as they make use of much larger repositories of material, teachers will be unable to ascertain what content and assessment activities have been presented to each student in the classroom. This will be the case if machine learning agents are given the opportunity to act in a more autonomous fashion. For this reason, institutions who wish to deliver courses with much larger numbers of students through MOOCs are likely to welcome the introduction of machine learning agents.
• Finally, it is likely that the use of machine learning agents alongside good teaching, assessment and student support will invariably improve student outcomes. Course teams or institutions who state that they already have good attainment rates will find that the use of machine learning will add substantial value to the products and services that they provide to their learners and to the communities that they serve.

Summary
The growing use of learning analytics within the education sector will generate new products or services such as machine learning agents. Companies that offer services around information systems, data management, learning management systems or library management systems will begin to use machine learning to enhance or transform their services. The use of machine learning also provides opportunities for companies to create radically different products and services for the education sector.

The use of agents blurs the rigid software boundaries between the traditional virtual learning environment, the learner management system, an institution's pool of information systems, library management systems, communication and news channels, the student home page and the personal learning environment. The boundaries will also be blurred with regard to the agents themselves; particularly when students and teachers will not be able to distinguish the work of individual agents. The interdependency of agents will also mean that agents will be increasingly reliant on each other to fulfill their tasks.

The schools, colleges and universities that will do best will be the ones who are the most effective at manipulating data, information, ideas and knowledge to deliver additional value to themselves and most importantly to the students and communities that they serve. However, as you have seen, in order to take advantage of machine learning they must learn new ways of doing things; especially with regard to the work that generates value to the institution.

Footnotes:

1. Learning Support Agents (Aftab Hussain, October 2016)
2. The Wealth of Humans. Chapter 2.5 - The Firm as an Information-Processing Organism (Ryan Avent, 2016)
3. Choo, Chun Wei (1996) "The knowing organisation: how organisations use information to construct meaning, create knowledge and make decision" International Journal of Information Management, 16(5), 23-40.