Greetings from the ANU Generative AI in Health Education Symposium in the ANU Moot Court in Canberra. This reports on work to produce and trail guidelines for teaching medical students. This felt like an extension of ASCILITE 2024 in Melbourne last week, as it covered some of the same topics and some of the same people were running the event.
Professor Martyn Kirk, Associate Dean Education, College of Health and Medicine, suggested using generative AI for formative assessment, bit not summative. The idea being the AI could be used for helping students learn, but then they be tested at the end to check they really know what they need to know. I suggest blending the assessments, not having this split between formative and summative. In equity terms not having a large test at the end will allow inclusion of students (such as myself) who can't cope with large exams. It will remove the unnecessary anxiety it causes for many other students. It will also result in more authentic testing, under conditions like a workplace.
The guidelines were trailed in second semester 2024. Students were comfortable using the tool (Microsoft Copilot) and found it helpful for learning. Most students did not receive any training on the tool, but did receive guidance on its use. The approach taken was to issue guidance and leave it to course conveners to tell students about it. Staff and students were provided with assistance in the use of such tools. This is similar to the way in the ANU Techlauncher program we ran a workshop to run students through what Copilot could do, or not do, to help them with an assignment.
One question which came up was the energy used by Generative AI. Students worry about the effect on the environment. In 2008 I was commissioned to design a course on Green Computing. In this students looked at energy consumption by data centers generally. AI is a more energy intensive form of data center. There are ways to reduce the energy use and carbon emissions resulting. This is something which perhaps should be raised with students generally, not just technical specialists.
One of the courses using generative AI was on climate change and health. In this course tutors demonstrated the use of AI to students in their work. This included using AI to simulate a person for the student to interact with. This was done with face to face and online tutorials. In this case students were given prompts to use with the AI. This extended the student's understanding, where previously they just asked one question and pasted the answer. The approach of simulations could be applied, I suggest, in other disciplines.
In a course on immunology students could use AI for preparing presentations, but not for writing up their laboratory notebook. How you stop students using AI for the notebook is a challenge. One way would be to require the students to compose directly into an online tool. As with the climate change course, students were given sample prompts to help them.
The symposium then switched from AI for teaching medical students to teaching students to use AI in medicine. Dr Andrew Tagg, Senior Clinical Lecturer at the University of Melbourne and Emergency Consultant at Western Health, Melbourne, pointed out that AI is already used routinely in diagnosis of cancer from x-rays. He argued that AI should be used more widely to deliver medical care. As someone who recently had to wait 11 hours for medical treatment, this could be a good idea. One area where AI might be useful would be in helping patients filling in the forms they are given.
The issue of the environmental effects of AI came up again in discussion. It was suggested generating one image took as much energy as charging a mobile phone. This sounded an over estimate to me by three orders of magnitude. The Jevons Paradox came up in discussion: rather than saving effort will AI just result in more resources being used, rather than oit being used sparingly.
