Loading…
CUHK Expo 2019/20 has ended
Back To Schedule
Wednesday, July 29 • 12:00pm - 12:30pm
P34: DeepMind and Beyond: Using Machine Learning to Teach an Artificial Intelligence Anatomy for Medical Education

Log in to save this to your schedule, view media, leave feedback and see who's attending!

In the COVID-19 era of distance learning, Artificial Intelligence (AI) agents conversing through dialogue systems offer a way to capture an important pedagogy; the individual student-teacher discussion. In order to achieve this, the first step is training the AI on the subject matter to be discussed.

Customising open-source AI tools from Google’s DeepMind, we attempted to answer the research question; ‘Can we train an AI to discuss human anatomy via Machine Learning?’

Methods

The research team constructed a customised training database of anatomical information linked to the UK Anatomy Syllabus for Medical Graduates and trained an AI agent using Machine Learning.

A subset of research team members independently formulated questioned to be posed to the AI and typed them via a dialogue interface. The AI gave an answer with an associated confidence value, and these were reviewed by a separate panel of experienced anatomy teachers.

Results:

The confidence values ranged from 0.56 to 0.70, indicating the degree to which the algorithm thinks its answer is correct. All of the questions were assessed to have been answered correctly by the anatomy teaching panel (N=15/15, 100%), despite no explicit programming of questions or answers.

Discussion

This pilot study has demonstrated the ability to train a truly intelligent AI agent which was able to handle a variety of question formulations, the like of which may be encountered in day to day teaching. Unexpected questions, simple spelling and grammatical errors posed no issue for the AI, as they might do in other non-intelligent systems.

View Poster
View Video



Wednesday July 29, 2020 12:00pm - 12:30pm HKT
Room B
  Online Teaching I, Poster
  • Session Type Poster
  • Presentation Type Poster
  • Poster No. P34
  • Order in Theme Session 3
  • Submission No. 89