Skip to Content

Wiki source code of Kris Reflection

Version 6.1 by Kris van Melis on 2024/04/09 00:22
Show last authors
1 At the start of the project, we did a general brainstorm to see what the main problems of people with dementia are. It was quite insightful learning about the issues people with dementia have and how nurses and family current support in it. We picked specifically for the issue that people with early stage dementia are prone to forget recent events. The goal of dementia care is to keep the quality of life as high as possible. Forgetting events, especially in relationships, can cause a lot of distress and social awkwardness. That why we've chosen to support people with dementia by providing a help system for retrieving these memories.
2
3
4 At first, I was quite skeptical about the use of a robot for this. It seems like an unnecessary complexity for a solution that could just work like a smart speaker. However, over the week, I've noticed that people act differently around robots. Somehow, the life like form makes us relate more to the system. We are more likely to also ask: How are you? It shows that we feel like there is more there than just a system to serve us. It creates a personal connection which seem to hold much value.
5
6
7 In designing the system, many questions arose. How should we get the relationship information? Should the person with dementia always be able to make changes to it? Or should it be locked to stop incorrect information from entering this database? Should it lie to avoid distress? This also has giving me a lot of insights into the complexity of dementia treatment and creating a software/robotics solution for it.
8
9
10 After creating a prototype with a custom GPT from ChatGPT and a puppeteer program to move the robot, I was quite impressed with the quality of the system. The answers were quite elaborate and the text-to-speech sounded really nice, warm, and comforting. After fine-tuning the answer style, we were ready for the first tests.
11
12
13 I was quite excited to see how people would interact with the robot. The result was quite positive, the robot added a dimension that people often went quite off track with their initial quest: finding out the family and relationships. At some point, someone even asked for ideas to plan someone's birthday party, which the GPT could answer quite well. However, I've now seemed that there are quite big hurdles in the way. When the GPT picked up a question incorrectly, this can be very confusing. Also, implementing a robot for every dementia home seems like a quite logistical operation, since we already had issues getting the robot to work at the start. So imagine non-technical nurses having to maintain the robot.
14
15
16 In the end, this course has given me a better overview of Social Cognitive Engineering. Especially a look into what robots can do and how to work out such a system with use cases, specifications etc.
17
18
19 Thank you for teaching this course :)
20
21
22 For my contributions, I have mostly worked on the overall idea basis, technology, use cases and getting the demo to work for the evaluation with the customized GPT. I also create the evaluation test to test how much the participants have remembered from interacting with the Nao. For the rest, our team collaborated well on every part of the project. So I feel like we've all had a positive effect on the result of the XWiki as a whole.