Yesterday I attended the Jisc/UCISA Artificial intelligence chatbot symposium, held in Birmingham. As part of this, Bolton College and Lancaster University shared their chatbot journey – demos and presentations that, I think it would be fair to say, greatly impressed us all.
I thought it worth sharing a little on these and some thoughts on how Jisc could support chatbot/personal assistant developments going forwards.
Before I go on, I’ll just get one thing out of the way! A core theme during the day was that these chatbots were not aiming to reduce or replace human interaction. The aim is to allow staff to spend more time on the stuff humans are great at – explaining, understanding, empathising etc. In some ways they were trying to provide an easier way to access the kinds of information and transactions that you might search for and look up on websites or apps.
You can find our more about Bolton College’s bot, Ada, here :
http://www.aftabhussain.com/responsive-boltoncollege.html
and Lancaster University’s bot, L.U., here:
What they do
Both bots aim to provide students with an easy way to get answers to common information, both by text chat and voice services. For Ada and L.U. this includes both general information and personal information. In addition, Bolton’s Ada could also provide answers to some subject/domain knowledge questions and also provides nudges to students, and Lancaster’s L.U. could handle some transactions, such as allowing a student to book a study room in the library.
Examples of general information:
- “What time does the students union stop open”
- “What week is it?”
- “Where is the library?”
Examples of personal information:
- “What lectures do I have today?”
- “What is my attendance like?”
- “What are my assignment marks?”
- “Who is my tutor?”
- “What is my email address?”
How they work
The general architecture of both systems is similar. They use a cloud based chat/voice service to handle the input, and then use APIs to interact with various information systems in the University, either directly, or via an intermediate API. Ada uses IBM’s Watson for the chat service, whereas L.U. uses Amazon’s Lex. We’ve experimented with both in Jisc – they are dead easy to get started with and experiment with, and you can start for free:
Watson: https://www.ibm.com/watson/how-to-build-a-chatbot
Lex: https://aws.amazon.com/lex/
What could Jisc do?
In Jisc we are looking at how we could best support chatbots. We’ve got a few ideas, which I’ll share.
Provide a chatbot service?
The obvious thing would be to provide a chatbot service for those that don’t wish to develop their own. If we based this on our learning data hub (the core infrastructure for learning analytics – https://www.jisc.ac.uk/learning-analytics) we could provide answers to quite a few questions out of the box, for example:
- “Who is my tutor?”
- “What is my attendance like?”
- “What are my marks?”
- “When is my next lecture?”
We’d then need to be able to provide a mechanism to provide more general campus information – perhaps by scraping existing data sources.
One variation that we’re particularly interested in, which I’ll share more on later, is the idea of a ‘sidebot’ (working name!) – one that participates in a forum or discussion as member, chipping in when it has something useful to add, rather than being something students or staff talk to directly.
Provide chatbot friendly APIs?
Bolton College shared a diagram showing the data sources that they were currently using, and explained they’ve love to see more services being integrated, including from beyond the college. This is something that Jisc could, so we could provide domain knowledge from, say, our hairdressing training resources (https://www.jisc.ac.uk/hairdressing-training) or information on open access research from our CORE service (https://www.jisc.ac.uk/core).
Shared conversation design?
One thing that stuck me that if we don’t share conversation design, there is going to be a huge amount of repeated work in the sector. There are only so many ways a student can ask about their timetable, or to book a room. In chatbot terms, these will usually map to ‘intents’ – have a read here to understand a little more:
https://chatbotsmagazine.com/chatbot-vocabulary-10-chatbot-terms-you-need-to-know-3911b1ef31b4
We could provide some kind of repository of proven, well designed conversations for all the common activities we see in universities and colleges, ready to integrate in Lex, Watson and so on.
Data readiness assessment?
In Jisc we’ve been talking about the idea of a ‘data x-ray’, and what this might look like. One idea would be to frame this as a series of questions that students might ask of your data, which you can then assess against various factors to figure out whether it might be a viable data source for a chatbot (or any other service!). Factors might include:
- Does it exist at all? – for example Lancaster University explained they wanted to answer questions on opening times, but this was not held anywhere centrally, so they needed to create a data source.
- Accuracy – are you confident the data is correct?
- Timeliness – how often is it updated? For example if marks are only entered into a central system just before exam boards then it will be hard to answer ‘what are my marks’ in a timely fashion
- Accessible – is there any API to access the data? If not, is there any other automatic way of getting to it?
If you are interested in working on us with any of these ideas then get in touch!
michael.webb@jisc.ac.uk
@michaeldwebb