From the 27th to the 30th of July I attended the 19th AIED conference
taking place during the Festival of Learning
at UCL in London, the U.K. The event featured three co-located conferences: the International Conference of Learning Science (ICLS), Learning at Scale (L@S) and Artificial Intelligence in Education (AIED). The conference was highly participated (around 1000 attendees) and allowed for cross-pollination between different communities. I attended only AIED and in this post I try to summarise my experience.
Several interesting talks on the relationships between computers and the human brain and how machine learning can complement human learning. Interesting discussions happened around new techniques to make student cognitive modelling, including scrutable models to allow easier diagnosis.
A great talk I attended was "What Vygotsky thinks about deep neural networks?" This research paper was presented by Iikka Tuomi and analysed how Google's inception (deep learning algorithm) performs the classification of geometric objects. Tuomi derived that there is a substantial difference between human and deep neural networks. The latter show no syncretism property, no complex chaining of concepts, and no relational knowledge. Deep learning only leverages associative rules, which in many cases is more than enough.
This parallel between human and machine brain was present in the keynote by Michael Thomas who gave a more neural psychology perspective and stated something I had the feeling during my studies in AI. The study of machine learning can reveal constraints on learning that stem from how the brain works.
The keynote Paulo Blikstein focused instead on the current biggest challenge, which I summarise in short as “saving Education principles from the AI/data hype”. Blikstein called for re-evaluating the format we are using in edTech and he called the AIED community to become the champion for an *ethical* use of AI in education.
Best AIED paper
Great attention deserves the research by Holstein et al. concerning the development of an AI tutor for teachers using Microsoft Hololens. The project, called Lumilo, uses Hololens for real-time teacher awareness of the classroom. The research shows that students learn more if the teachers are informed by the analytics. For more information see http://kenholstein.com/TeacherAI
My presentation at the Young Research Track
My six minutes of fame took place during the Young Research Track (doctoral consortium). Roughly 10 doctoral students presented their projects and I was one of them presenting my “Multimodal Tutor for CPR”. The room was packed form the presentations and questions could be asked both after the presentation or using an online form. I received a lot of interesting comments to improve my experiment. It was probably me, more open for feedback, but I felt this year the comments were more constructive compared to last year's doctoral consortium. My presentation will be uploaded on Slideshare soon.
Multimodality still a niche 🙁
It was unfortunate to notice that, despite the great announcements, very little studies have used multimodal approaches. Dominant approach is to collect learner-computer interaction data, especially during multiple choice question in mathematics or physics. The last panel sounded promising on paper for its title “Learning in 2050 beyond keyboard”, in practice it was disappointing. Speakers had troubles thinking beyond tablets and MOOCs. Very little examples of ubiquitous and pervasive technologies such as wearable devices or augmented reality. As a researcher involved in the CrossMMLA community, I believe there needs to be more attention to the evolution of human-computer interaction paradigms which are progressively multimodal and their possible applications in the domain of learning.
- The AIED Proceedings are available for FREE until 21st July
- Next two AIED conferences will be organised in Chicago, the USA in 2019 and in Ifrane, Marocco in 2020.
- Pictures of the Festival of Learning can be found here