How digital assistants are promoting enhanced accessibility at the Open University

The Open University uses Taylor, an AI-based digital assistant, to improve the student experience for disabled learners. In this instalment of the Here and Now series, the Jisc AI Team speaks with Dr Tim Coughlan, Senior Lecturer in Education Technology at the Open University, to explore the benefits and practicalities of using digital assistants in this context.

Can you tell me a bit about Taylor and explain how is it being used to improve the student experience at the Open University?

Taylor is a digital assistant that is designed to have a dialogue with students who disclose a disability. A conversation with Taylor has two purposes. The first is for the student to provide information, covering things such as the nature of their disabilities, any assistive technologies they use, and areas where they could require support or adjustments in study. Taylor is an alternative to the usual process of filling in forms to provide this information.

The second purpose is to help each student to better understand what OU study entails and the support that they could benefit from. Students are given some introductory information on key topics, and they can also ask Taylor questions. This is designed to be part of the conversation throughout, with the idea that students can learn from the conversation and give better answers to the questions they are asked.

For example, the university needs to capture which disability categories a student will be identified in. But we know that students struggle with these categories. So an exchange with Taylor may go like this:

Taylor “…can you tell me a bit about any disabilities or study needs that you have?”

Student “I struggle to type with my hands”

Taylor “From what I’ve understood, I think these categories apply: Manual Skills”

Student “can I ask a question?”

Taylor “what would you like to ask?”

Student “what is manual skills?”

Taylor “this is a category of disability that the OU recognises related to dexterity and motor skills. It includes for example: difficulties to use the mouse, keyboard and struggling to write.”

Student “OK, continue”

In this case, the student has learnt how this disability category is defined and can confirm that this category does indeed apply to them.

What motivated you to utilise digital assistants in this way, and why was this type of technology an appropriate solution to your problems?

Listening to our disabled students, we recognised that the administrative burden placed on them is a substantial barrier to success and experience (Coughlan & Lister, 2018). With rising numbers of students disclosing disabilities and needing timely support, there are also organisational challenges to tackle here, in order to create effective processes. Our aim was to create an alternative to completing forms and to explore with students and staff how this could be designed to be accessible and helpful as a part of these processes.

We also know that students benefit from conversations with our disability advisors, but this expert human support is a limited resource that needs to be used wisely. In this space, a digital assistant offers a lot of potential. Because it supports a dialogue, there is the potential for both the assistant and the student to ask questions and clarify anything that was said, before finalising the information that is shared. It is also our goal that the student learns something useful from the conversation and that it feels more beneficial than filling in a form.

Can you give us an overview of how Taylor works?

Taylor was created by our Learning and Teaching Technologies team using a combination of Microsoft Azure services. When students access it, the system authenticates them using their university login, so it knows who they are. It then offers them the option of enabling speech, so if they want to they can use voice in the conversation, and listen to what Taylor has to say. Or they can just use text. Supporting the use of voice or text for input and output is one example where assistants such as these could be more accessible and flexible than forms. The student can also ask Taylor to speed up or slow down, and we worked on making the speech input patient with students who may take more time to respond or need to pause if they are explaining something in detail.

Taylor uses a conversation flow with a script that was iterated several times with input from students and advisors to provide essential information and ask questions in a succinct way. This takes the student through topics such as their disabilities, study materials, and access to tutorials. Taylor can use natural language processing to ‘understand’ what the student has said, for example when identifying which recognised categories their disabilities fit with. This can then lead to appropriate responses, allow the student to use their own terms and result in more useful data being captured from the conversation.

At any time, the student can interrupt the assistant, and it then expects questions which it answers using a knowledge base. Working with students and staff we developed answers for over 100 questions related to disability, technology, support processes and study, and this can be continually expanded and improved on.

Variations on the same question are added to the system to improve accurate interpretation of questions, so if a student asked “how do I contact my tutor?”, “when do I get to talk to my tutor?”, or “how do I call my tutor?” the similar intent is recognised and the student will receive the same response from our knowledge base. If the student asks a variation using similar word combinations this will be assessed by the language understanding model to give a level of confidence in an answer. If this is above a threshold the answer will be given to the student, otherwise they will be asked if they can rephrase the question in order that it can be understood.

What reception has Taylor had from students?

Early interest and feedback around this was really positive, but we wanted to get a realistic sense of whether this was working and what students thought of it. So we ran a trial in which new students disclosed disabilities using both Taylor and the existing form-based process for comparison. 65% of the 134 participants preferred using Taylor, and the feedback suggests that we’d be able to convince most of the others with some improvements to the design.

How else have staff and students at the Open University benefited from Taylor?

Staff are really positive about Taylor and interested to get engaged in taking it further. There’s been a lot of engagement with different areas of the university as we developed the design, got it in front of students and staff, and shared the results. As well as addressing a particular issue, we’re building up the general knowledge that the university has about the potential and challenges for using chatbots, virtual assistants and other artificial intelligence technologies. There are other pilots and ideas in development and we’re contributing to the technical understanding of what is possible and the strategic decisions that need to be made to make the best of these technologies.

This project has also given us the opportunity to respond to what disabled students at the OU have told us, and to work with them to address the barriers they face to successful study. We have worked with Microsoft on this and their support for the development of Taylor has helped us to push forward understanding of the potential for AI to enhance accessibility. We’ve published papers about the design and evaluation research undertaken for the project at key accessibility conferences and won an award for Best Communication Paper at the Web for All 2021 conference.

What else do you think can be achieved through using digital assistants, and do you have any further plans in this area?

Taylor is really a foundation for achieving a lot more in the future. We’re planning to enhance it in several ways, for example by integrating more information about the student and their study, which will mean it can give more targeted advice, and also looking at how staff can monitor Taylor and contribute to its ongoing improvement. Administrative burden for disabled people is a wider issue both across the education sector and in wider society. We’re in a good place to work on it at The Open University because of the huge population of disabled students who study with us. From the start of the project we’ve been discussing with other institutions to understand the similarities and areas of difference in their disability support processes. We’re keen to talk to anyone interested in collaborating to develop assistants that could work for disabled students in any institution, and in applying our understanding to improving the wider administrative processes that disabled people need to complete.

What advice would you have for a university or college who was considering using a chat bot or digital assistant in their setting?

It was important for us to identify a real problem which had the potential to be addressed through these technologies. There’s a combination of features here: barriers to student success, limited advisor capacity, and potential to make a better, more efficient experience. That is a good fit to the potential of digital assistants. We’ve been collaborating with Jisc’s subject specialists in accessibility and assistive technology to share knowledge about this project and discuss the wider potential with other universities and colleges. Taylor is great as an example of what the technology can do, and people have recognised the same issues in their own institutions, so they are interested in whether it can be adapted to their needs. At the same time, they often come up with different problems where an assistant could reduce the burden on staff and students and provide timely personalised information.

Another issue that has been central in these discussions is the readiness to create or adopt these systems. The assistant won’t stand alone. It will need to integrate with current processes and interact with data. It’s important to understand the current systems and also to work with stakeholders across departments and roles to make the best of the technology.

We were very clear that our aim wasn’t to replace expert advisors with an AI-based system, but instead to work with our staff and students to find and address challenges they faced. For example, they told us that they were having to repeat the same information to new students in lots of their phone calls, which took time away from applying their expertise and responding to complex issues. Taylor could take some of this burden away, and get that information to the student earlier.