Presenting the winners
The Hackathon AI in Higher Education 2023 at the University of Twente concluded with a unique twist: two winners. Due to the unexpectedly high number of 12 participating teams, the event was judged by split juries, leading to two outstanding presentations that brilliantly utilized Large Language Models (LLMs) in the realm of higher education:
- Team UT/Bremen (from the University of Twente and Universität Bremen) developed a proof-of-concept LLM-based interactive tool that could analyze short presentation videos in real-time, offering immediate feedback and suggestions for improvement.
- Team sAxI (from Saxion University of Applied Sciences) proposed an LLM-based system to retrieve useful information of the university documents and support personalized coaching sessions, potentially preventing student dropout.
Let's find out more about these winning innovations. I was pleased to interview both teams and get more details.
Could you tell us a bit about how you got to your solution?
Team UT/Bremen
“(Student) teachers are more and more making use of prerecorded micro lecture videos to instruct their students on a specific topic. However, checking and giving feedback on these explanatory videos take a lot of time from teacher trainers.
What we suggested is that a LLM-based system could facilitate this process by for example using Mayer’s Multimedia Principles and recent scientific evidence on high-quality instruction. In our tool, learners can upload a video into the AI system and based on a number of criteria that we use as inputs, the AI analyzes both the graphics of the slides and the spoken text in order to give feedback on the use of multimedia principles and on the quality of instruction. This approach not only streamlines the assessment process but also provides a more tailored educational experience for each student.
It was really a brainstorming process starting from the dinner. We decided to build on top of what was already done in Bremen with video analysis and discuss how Mayer’s multimedia principles and evidence on what constitutes high-quality instruction could be helpful in providing feedback on the explanatory videos. I think what was really nice is that the different expertise from Bremen and the UT really complemented each other so well.”
Team sAxI
“Currently at Saxion (and also other universities of applied sciences), there are many students that drop out and there are a lot of reasons why students drop out. For example, every student has a study coach, but some students don't know that they have a student coach or that the student coach is not available 24/7. There is the Study Success Centre to help students who are struggling, for example, with education in general. Students don’t always know about it or there is a ‘distance’ to it. Lastly, Saxion offers a lot of data that can inform students but all these documents are placed over the intranet and not many students know where to find them, and when we were students, we also missed that.
We then thought that an emphatic interface for searching would be helpful for students in their situations, and we had several ideas. But how do we support students in their planning? We immediately came up with a chatbot, named sAxI, since for the hackathon it could be a doable idea. We then discussed how could we use existing platforms ChatGPT or similar technology as a solution to our problem.
The dinner was a very good pressure cooker to come up with ideas with the team.”
What are some technical challenges your team faced?
Team UT/Bremen
“Ethics, of course, is a big concern. You do not really know what happens to your data if you use a system such as OpenAI. You need an AI on your own server, so the data are protected.
Secondly, AI is complex and you need a lot of different competencies to use it. Just having a tech guy is not enough, just having an educational scientist is not enough. Just having a teacher trainer is not enough, so you need to have all these different expertise to work with AI effectively.
Also, the time constraints of the hackathon were quite tough. It was enough time to get something up and running, but more time would have helped.”
Team sAxI
“Answering how to support students in their planning meant that privacy and security were definitely key challenges. How do we get information from previous cases to support their planning? Then we faced the fact that this is not possible as these previous students have their privacy and we don’t have access to this information. A chatbot having access to student information, how should it work, what kind of documents should we use, and how should the chatbot act on these kinds of problems? Organizing the student programs and regulations into a knowledge base is a challenging complexity.”
What future participants should know?
Team UT/Bremen
“Try to find an inter or multidisciplinary team so that you can collaborate and use your own expertise. It's helpful if some of the team members already have some experience with hackathons.
Also, what is important in AI solutions for explanatory videos is having criteria and this is not just some idea out of thin air, we used scientific literature, frameworks on multimodal learning and effective instruction. Not just coming from the technology and doing something, but it's the other way around, starting with a clear goal and asking ourselves: Do we have some sound theoretical understanding and bases to see how we can use technology to support reaching that goal?
Looking back, we also think it’s important to find a balance between brainstorming and just trying things out because at one point we were brainstorming a lot and we had a lot of ideas. You have to decide at a certain point, that this is the idea that we are going to try to develop a prototype for. The danger is getting stuck in the brainstorming phase. It's good to focus on some idea to mock it up or to make a prototype and do a live demonstration.”
Team sAxI
“Having already an idea in place can be quite helpful, so you can spend the heavy discussions about the solution on the main day.
Take time to think of the ethical aspects, not focusing on making a fancy tool.
Also, a few times we asked the Jedis, trying to ask questions what are the right questions. Just the process of talking to them helped a lot. Also, it was interesting to see that working from different disciplines we all knew something that was helpful at different times.
The collaboration was the fun part. It is a process of trial and error, and it’s quite a process to reach a good presentation. The time pressure was also something that you had to go through.
For the presentation, keep it short and simple. Early before the presentation, we separated the tasks, we knew who was doing what, so hours we were better prepared.”
Conclusion
Looking back, UT's Hackathon AI in Higher Education 2023 was inspiring in many ways. Twelve teams registered, great inspiration sessions, intense collaboration, expertise from various sectors, helpful jedis, coffee, delicious food, brilliant organization, and in the end, two winners.
Two innovative solutions, UT/Bremen’s tool for explanatory videos and sAxI for academic coaching, pave the way for a more adaptive and personalized learning experience. The success of the winning teams reflects their visionary approach to integrating AI into education, setting a milestone for future innovations in the field.
The winning teams advise future participants to come in with some ideas in advance, preferably with a multidisciplinary team, and to enjoy the process. In the next editions of UT’s Hackathon AI in Education, we do hope to see the same inspiring enthusiasm as we had this year!
