Lamp you have to teach
Since 2017 I have been experimenting with machine learning as a design material. Fascinated by the possibility of redefining relationship between people and ML-powered artefacts, I started designing with focus on evoking positive emotions, usage for personal purposes and promoting understanding of basic principles behind technology, thus putting the human in the position of control. My work been part of research competitions (CHI’18 Montreal), invited talks (Hacking ML Munich) and got me to conferences (ML Prague). With my position paper on opening black box of AI through implicit learning, I am going to be a part of Emerging Perspectives in Interactive Machine Learning workshop, which is a gathering of domain experts during largest Human Computer Interaction conference (CHI’19 Glasgow). It all started with my thesis work on Silly Objects, open-ended intentionally trained ML-powered artefacts.
How to design machine learning artefacts that empower humans? The current interaction paradigms for ML-based artefacts and services don’t promote active human participation in the domain, are often obscured and indirect. Can we re-imagine the relationship between humans and technology in a home setting, so that it could be gradually built on trust?
Democratic, inclusive and engaging experiences through first stages of personal training of ML algorithms. Human-AI relationship developing over time as the agency of the artefact shifts.
A working prototype of a Silly Lamp, lamp that does nothing unless you teach it, was deployed in participants houses for 24- and 72-hours user sessions. Insights from the experiments, iterations and user studies formed a silly object aesthetics and set of recommendations for engaging ML experiences. This work was selected to be part of Student Research Competition and was presented at CHI’18 in Montreal.
- 10 weeks
- Machine learning
- Tangible Interaction
- Design based research
Silly objects are simple, open-ended technologies intentionally trained by the end users. Instead of constantly recording data, they require intentional training examples. They let the user shape their agency and gradually shift into the periphery of attention.
The concept of Silly objects was developed in a design process consisting of a survey, iterative prototyping, co-design sessions and sketching in open-sourse hardware and software.
A year later this project was followed-up by another 10 weeks exploration. This time, I focused on how qualities presented by Silly Objects could also be applied in already existing products. I experimented with focus on single interaction eg. making a choice in a conversation with a voice assistant, handing over manual control to autonomous smart home product or data access permission.
The experiments formulated a framework of qualities for designing for negotiated agency – relationship of interaction dialogue with our devices in which their agency is established. I will be presenting this project during Emerging Perspectives in Human-Centred Machine Learning Workshop as a part of CHI’19 Glasgow.