=Paper=
{{Paper
|id=Vol-2192/ialatecml_paper5
|storemode=property
|title=Human-Technology Relations in a Machine Learning Based Commuter App
|pdfUrl=https://ceur-ws.org/Vol-2192/ialatecml_paper5.pdf
|volume=Vol-2192
|authors=Lars Holmberg
|dblpUrl=https://dblp.org/rec/conf/pkdd/Holmberg18
}}
==Human-Technology Relations in a Machine Learning Based Commuter App==
https://ceur-ws.org/Vol-2192/ialatecml_paper5.pdf
Human-Technology relations in a machine
learning based commuter app.?
Lars Holmberg1
Faculty of Technology and Society, Malmö University, Sweden
Internet of Things and People Research Center, Malmö University, Sweden
Lars.Holmberg@mau.se
1 Introduction
Artifacts that take advantage of Machine Learning (ML) influence our daily life
to an increasing extent. The development and use of these artifacts often leaves
out the human actors and the context and thus risks to become technocentric [6].
Many ML techniques, for example active learning, interactive learning or
machine teaching, calls for user involvement [3, 10, 12, 13, 15]. To our knowledge
there is a lack of research concerning relations between humans and ML-based
artifacts, the attention is instead often on the desired outcome [4, 11], for example
trust.
If we are to understand how ML can support humans, we need to turn at-
tention to the human-technology relations. In this work we will use a postphe-
nomenological lens and focus on how different ML techniques can create, build
and maintain relations between humans and technology [5, 14].
Our work is explorative and use material sketching to produce artifactual
knowledge [7], methodologically this fits into a Research Through Design ap-
proach [16].
In this work, our artifact and the relevance of a postphenomenological ap-
proach is the main contribution. In the work presented here the focus is on
adaptive learning in a background relation [5] and identifying situations where
the background relation can promote a transition to another type of relation
(embodied, hermeneutic, alterity). We hope to initiate a discussion on this ap-
proach and inspire further work by our contribution.
In this extended abstract, we will focus on the artifact and its relevance
for human-technology relations and only briefly expand on application context,
theory and methodology.
2 Related research and Methodology
As a blueprint for this work we have followed the directions of Ohlin et al. [9]
and we aim at promoting transitions between relations initiated either by the
?
This work partially financed by the Knowledge Foundation through the Internet of
Things and People research profile. I would also like to thank Paul Davidsson and
Carl Magnus Olsson for invaluable support during the work.
73
Human-Technology
2 L. Holmberg Relations in a Machine Learning Based Commuter App
artifact or the user. We start by creating an adaptive ML-backend for an existing
commuter app 1 2 . The ML-backend predicts and presents the users next journey
when the app is started, the prediction is based on historical individual travel
patterns and the users ML-model is trained online. To create a baseline for our
predictions, personas based on a local transportation company are used [8].
3 Result and discussion
In our process to create an artifact that meets our initial design goals regarding:
prediction accuracy, cloud service cost and performance, we have iterated and
re-framed our artifact multiple times. The users trust in the predictions is central
to be able to build relations, but trust doesn’t automatically follow prediction
accuracy [4]. This implies an adaptive backend in the background that delivers
reasonable accurate predictions for data sets that reflects use over weeks, months
and year.
The current artifact iteration is a solution that extends the Android com-
muter app and adds a backend based on Google Cloud services (Firebase, Big-
Query, compute engine), TensorFlow estimators [1, 2] and Node.js 3 . With our
personas in focus we created one travel pattern that is periodic, one that drifts
over time and one more random. Based on this data an adaptive ML-backend was
created that uses different ML-Algorithms depending on amount of labeled data
and noise in input data. The resulting system delivers journey predictions to the
app in an acceptable time-frame (less than 2 seconds) and identifies situations
were transition to another type of relation is appropriate.
The main contribution in this work is identifying situations where there is
a need for a transition to a new type of human-technology relation. Since the
work presented here builds a background relation the transitions calls for an
ML-technology that allows or invites users to get involved.
4 Conclusion
This extended abstract set out to turn attention to the human-technology rela-
tions ML-artifacts promote. We have done that by using the vocabulary of post-
phenomenology and by exploring the design space created by a commuter app.
By specifically focusing on adaptive learning in a background relation and identi-
fying situations that calls for a shift in type of relation (embodied, hermeneutic,
alterity) we have laid a base for future research.
In terms of future research we particularly suggest to include active learning,
interactive learning and machine teaching to explore the relations these tech-
niques promote.
1
https://skanependlaren.firebaseapp.com
2
https://play.google.com/store/apps/details?id=se.k3larra.alvebuss
3
https://github.com/k3larra/commuter/
74
Human-Technology Relations
Human-Technology relationsininaaMachine Learningbased
machine learning Based Commuter
commuter app.App 3
References
1. Cheng, H.T., Haque, Z., Hong, L., Ispir, M., Mewald, C., Roumpos, G.,
Sculley, D., Smith, J., Soergel, D., Tang, Y., Tucker, P., Wicke, M.,
Xia, C., Xie, J.: TensorFlow Estimators: Managing Simplicity vs. Flexi-
bility in High-Level Machine Learning Frameworks. In: Proceedings of the
23rd ACM SIGKDD International Conference on Knowledge Discovery and
Data Mining. pp. 1763–1771 (2017). https://doi.org/10.1145/3097983.3098171,
https://terrytangyuan.github.io/data/papers/tf-estimators-kdd-paper.pdf
2. Cheng, H.T., Koc, L., Harmsen, J., Shaked, T., Chandra, T., Aradhye, H.,
Anderson, G., Corrado, G., Chai, W., Ispir, M., Anil, R., Haque, Z., Hong,
L., Jain, V., Liu, X., Shah, H.: Wide & Deep Learning for Recommender
Systems. In: Proceedings of the 1st Workshop on Deep Learning for Rec-
ommender Systems. pp. 7–10 (2016). https://doi.org/10.1145/2988450.2988454,
http://arxiv.org/abs/1606.07792
3. Fails, J.A., Olsen, D.R.: Interactive machine learning. Proceed-
ings of the 8th international conference on Intelligent user inter-
faces - IUI ’03 p. 39 (2003). https://doi.org/10.1145/604045.604056,
http://dl.acm.org/citation.cfm?doid=604045.604056
4. Glass, A., McGuinness, D.L., Wolverton, M.: Toward establishing trust in adaptive
agents. Proceedings of the international conference on Intelligent user interfaces
(IUI ’08) pp. 227–236 (2008). https://doi.org/10.1145/1378773.1378804
5. Ihde, D.: Technology and the Lifeworld, vol. 1 (1990).
https://doi.org/10.1049/et:20060114
6. Johnson, D.G., Verdicchio, M.: Reframing AI Discourse. Minds and Machines 27,
575–590 (2017). https://doi.org/10.1007/s11023-017-9417-6
7. Löwgren, J.: On the significance of making in interaction design research. Interac-
tions 23(3), 26–33 (2016). https://doi.org/10.1145/2904376
8. Nielsen, L.: Personas - User Focused Design.
Springer (2013). https://doi.org/10.1007/978-1-4471-4084-9,
http://link.springer.com/10.1007/978-1-4471-4084-9
9. Ohlin, F., Olsson, C.M.: Beyond a utility view of personal informatics.
Proceedings of the 2015 ACM International Joint Conference on Pervasive
and Ubiquitous Computing and Proceedings of the 2015 ACM International
Symposium on Wearable Computers - UbiComp ’15 pp. 1087–1092 (2015).
https://doi.org/10.1145/2800835.2800965
10. Porter, R., Theiler, J., Hush, D.: Interactive machine learning in data
exploitation. Computing in Science and Engineering 15(5), 12–20 (2013).
https://doi.org/10.1109/MCSE.2013.74
11. Ribeiro, M.T., Singh, S., Guestrin, C.: ”Why Should I Trust You?”: Explaining the
Predictions of Any Classifier (feb 2016). https://doi.org/10.1145/2939672.2939778,
http://arxiv.org/abs/1602.04938
12. Simard, P.Y., Amershi, S., Chickering, D.M., Pelton, A.E., Ghorashi, S., Meek, C.,
Ramos, G., Suh, J., Verwey, J., Wang, M., Wernsing, J.: Machine Teaching: A New
Paradigm for Building Machine Learning Systems. Tech. rep., Microsoft Research
(2017), http://arxiv.org/abs/1707.06742
13. Smith, A., Kumar, V., Boyd-Graber, J., Seppi, K., Findlater, L.: Closing
the loop: User-centered design and evaluation of a human-in-the-loop topic
modeling system. In: International Conference on Intelligent User Interfaces,
Proceedings IUI. pp. 293–304 (2018). https://doi.org/10.1145/3172944.3172965,
http://cs.colorado.edu/
75
Human-Technology
4 L. Holmberg Relations in a Machine Learning Based Commuter App
14. Verbeek, P.P.: What things do. No. 9 (2013).
https://doi.org/10.1017/CBO9781107415324.004
15. Zhu, X., Singla, A., Zilles, S., Rafferty, A.N.: An Overview of Machine Teaching
(2018), http://arxiv.org/abs/1801.05927
16. Zimmerman, J., Forlizzi, J., Evenson, S.: Research Through Design as a
Method for Interaction Design Research in HCI design research in HCI. Pro-
ceedings of the SIGCHI conference on Human factors in computing systems
pp. 493–502 (2007). https://doi.org/http://doi.acm.org/10.1145/1240624.1240704,
http://repository.cmu.edu/hcii
76