Tuesday, December 4, 2018

Chatbot Tutors for Blended Learning

Greetings from the University of Wollongong where I am taking part in a workshop on "Chatbot Tutors for Blended Learning" by Chi-Un Lei, Yuqian Chai, Xiangyu Hou, and Vincent Tam from TELI at University of Hong Kong. This is part of the IEEE 7th International Conference on Teaching, Assessment, and Learning for Engineering (TALE).
"Creating a pedagogical agent requires complex computer programming skills and it is usually built from scratch to fit the intended educational purpose. This makes it difficult for teachers to adapt existing systems or to attempt in creating a similar version. Recently, we have leveraged the IBM Watson Assistant chatbot engine to develop some chatbots. These chatbots have been adopted in a common core (general education) course for inquiry-based learning, with a promising outcome. In this workshop, we would like to share our experience gained from developing a set of chatbots for an online blended learning environment. Through the workshop, attendees will identify what and how chatbots can be designed for their own blended or fully online courses. They will also develop a simple chatbot system that can responses to some assessment inquiries."
I had in mind using this for routine questions from students. I provide tutors with a list of standard answers, but perhaps these could be provided by a chatbot.
The example given in the workshop is for  training forensics  scientists in crime scene analysis. The chatbot answers the easy questions and passes on the hard ones to the tutor.

The workshop is using the free version of the IBM Watson tool.  A JSON file was provided for the workshop to upload to Watson.We were cautioned that there were "25 steps" to do this from scratch.We were simply asked to enter the "intent", which we have to be specific about (such as the second assignment deadline). Then the "Entities", such as a specific assignment the student is asking about. The "Dialog" would be the answer.

While the example in workshop was to answer simple logistics questions (is the exam open book?), the idea is to use this for teaching. A simple example might be, rather than say "No, the exam ins not open book", but to guide the student through where they find the answer. This might address teascher's frustrations where students don't read the material given.

The prepared chatbot is not very bright and very single-minded. It reminds me of the talky toaster in Red Dwarf, which was obsessed with toast. One trick shown to make the dialogue more interesting was simply random versions of the same responses.A dialogue is added with answers for the questions. One of the features of using Watson is that it has a dictionary and lists of synonyms. As an example with the Q&A I prepared was about "books", Watson correctly interpreted "document" as being a sort of book (it even understood "Bring Dongle?").

Next we added an "entity".  These can be physical objects (the textbook for the course) or something more abstract.

This chatbot process reminds me of my first encounter with preparing a multiple choice online quiz. The process of setting up the questions and answers was very tedious. The reward only comes when you have many students using the system. The process with the chatbot doesn't look any more time consuming that a quiz, with the AI system providing flexibility.

What seems to be missing from this process is the intelligence to create the answers. For example, I would like to just give the system the rules for the course assessment and have it work out the possible questions and answers. As it is this form of AI seems to making the type of promises fourth generation programming language, did last century. Claims were made about programmers not being used, as it was easy for anyone to do. The reality was that it was easy for anyone to write a simple program, but it required a highly skilled expert for complex tasks. The result was a lot of time wasted by unskilled people writing bad code, and experts trying to untangle the resulting mess.

ps: The workshop was held in a UoW  lab. This had about 32 desktop PCs in four long rows of eight. There is a large display screen near the instructor's console at one end and on the opposite end. However, it would be useful f there were screens on the other two walls. The room is carpeted, but was still a bit noisy, with multiple tutors helping students. Perhaps the ceilings should be sound absorbing in these type of rooms(at the cost f requiring a microphone for the presenter).

No comments:

Post a Comment