In this CoDesign study we will together use a design probe and explore the area of Machine Teaching (MT)1 in a commuting context.
By teaching during use we will explore a teaching situation where you teach when your are in the context (I have called this "teach as you go"). During the use you will, from the previous weeks use, have an understanding of how the teaching works in general.
Therefor we could get insights in how to answer research questions like these:
RQ1: How does your expectations, given this implementation, on the teaching match the outcome when the teaching takes place during use?
RQ2: How does the perceived performance of the predictions relate to measured performance of the predictions?
RQ3: How could a teaching interface be designed for the teach as you go situation?
- Clear all teaching from last week.
- If you don't get a correct prediction when you are about to take a transport, teach the app so you get the prediction you expect.
- Don't clear the teaching during the week, just teach more until you get the prediction you expect.
Traditionally a lot of machine learning takes place before the usage. This means that a ML-model is trained and then used. For example a model predicting weather can be trained using years of temperature data and then used to predict the temperature next day. The problem here is that the model uses old data, so for example if the climate changes due to global warming the model will be outdated. To handle this situation the model has to be relearned with more recent data. In our case we will teach the model new things continuously and at the end of the week it should be a bit better at predicting than in the beginning of the week.
Here is a article by Jonas Löwgern on a related experiment2 in the medical sphere.
Add feedback here both general issues, thoughts and sketches or just notes so you remember, the more the better. Especially things that relates to the research questions above is valuable:
https://github.com/k3larra/commuter/blob/master/UserStudy/week3/presentation.md
[1] P. Y. Simard et al., “Machine Teaching: A New Paradigm for Building Machine Learning Systems,” 2017.
[2] M. Lindvall, J. Molin, and J. Löwgren, “The Importance of UX for Machine Teaching,” in The 2018 AAAI Spring Symposium Series The, 2018, pp. 407–410.