Embedding prior knowledge into your campus bots

  

In this short article I would like to explore how schools, colleges and universities can start to embed prior knowledge into their campus bots. Prior knowledge is important for the development of cognitive services on the campus because without that knowledge these services will be unable to act on behalf of the institution to address the day-to-day needs of students and colleagues. At a simple level prior knowledge enables services to present answers to all the frequently asked questions that are posed by everyone on the campus. At an intermediate level prior knowledge enables services to read, analyse, assess and act on the information that is stored across all the datasets within the institution and beyond. At an advanced level prior knowledge enables services to assimilate and learn from the environment around them. Let’s explore each of these statements in more detail.

At a simple level prior knowledge enables services to present answers to all the frequently asked questions that are posed by students, teachers and support teams on the campus.

Chatbots are representative of a new breed of services within the education sector that are capable of answering thousands of day-to-day questions that are posed by students, teachers and support teams. The performance of a campus chatbot is governed by the prior knowledge that is imbued into the service. Its ability to articulate responses to a myriad of questions reflects on the design, implementation and on-going support that a school, college or university invests in the service. The prior knowledge in these services give institutions the ability to offer one-to-one support to every one of their students and at all times in the day. As chatbots develop within the education sector they will begin to support students and colleagues with simple tasks, activities and transactions; such as applying for courses, making payments, handing in assignments, renewing library books, booking rooms, making appointments and more besides.

Ada, the chatbot at Bolton College enables students to find answers to a wide range of questions which will eventually relate to every point on the student life-cycle. When the College embeds prior knowledge into the service colleagues do more than just instill basic facts and information into the service. Since Ada acts on behalf of the College her responses are designed to reflect the sentiment and ethos of the College. In many ways, Ada can be seen as an additional member to multiple teams across the campus; hence the use of the word 'she'.

Chatbots will sit on top of every online service that students, teachers and support teams use. A student will have access to the institution's chatbot(s) on different platforms such as the campus learning management system, the library website, the student home page and more. As the student uses these platforms he or she will occasionally use the chatbot to seek out further information or to carry out specialist tasks to support their studies.

prior knowledge image

At an intermediate level prior knowledge enables services to read, analyse, assess and act on the information that is stored across all the datasets within the institution and beyond.

When the Ada service at Bolton College is asked a question it checks to see if the answer can be sourced locally. If not, Ada will attempt to source the answer from beyond the campus by using service APIs such as Wolfram Alpha. The College also plans to utilise Microsoft Academic in the near future. Were no answer can be sourced Ada will present the contact details for the Student Services team. Going forward, the ability to extend the capabilities of a campus chatbot will be crucial to their success. Within the education sector we could see the advent of specialist service platforms whose APIs augment the capabilities and knowledge domains of an institution’s primary chatbot.

A cognitive service such as Ada is the sum of multiple agents. The cooperative nature of these agents determines what Ada can do and how she will help students, teachers and support teams around the campus. For example, if a student asks about the hand-in-date for a forthcoming assignment, the service would be expected to respond by stating the date and time for handing in the assignment. Ada will also state the desired grade for this particular piece of work. The advice around the desired grade is predicated on the student’s current academic performance on the course and the goal that the student is aspiring to; such as seeking employment, further training or moving onto further or higher programmes of study. Ada may also prompt the student to see her tutor; even scheduling an appointment on behalf of the student and subject tutor. In this example, the agents must work in unison to support the student. The conduct of these agents is determined by prior knowledge and the parameters which shape and guide their behaviour. These agents represent the institution and should act on behalf of the institution to serve the needs of its students.

At an advanced level prior knowledge enables services to assimilate and learn from the environment around them.

Prior knowledge also has a bearing on the performance of machine learning models. For example, the behaviour of the machine learning model that is being developed to support automatic marking at Bolton College is governed by desired expectations and outcomes. At a simple level, the model needs to learn which parts of the long form answers that are submitted by students are correct and acceptable and vice versa. The machine learning model improves and fine tunes its behaviour as more training data is supplied to it. Once the automatic marking service goes live subsequent student work informs and improves the performance of the model. If the automatic marking model proves successful colleagues at Bolton College will apply the lessons learnt to many other use cases in other curriculum areas.

When traditional organisational workflows are combined with cognitive services educational institutions have the opportunity to take advantage of intelligent process automation to drive improvements in the services that are presented to students. When cognitive services are used in this manner they are transformative. At times it becomes difficult to imagine designing a service that does not take advantage of a cognitive service.

One of the interesting aspects of using machine learning models on the campus is their ability to produce new insights or knowledge. They are already deployed on adaptive learning environments; enabling institutions to improve the delivery of personalised and contextualised teaching, learning and assessment materials; and to do so at scale. Machine learning models augment the capabilities of every teacher and support team on the campus. They enable institutions to deliver one-to-one services at scale and they add value to the student experience.