Aroma is a physical-digital fictional prototype that I built for my Master Thesis in Interaction Design at Cyprus University of Technology.

As a digital designer who mostly focuses on mobile and web applications, I have experienced a growing usage of Artificial Intelligence and Machine Learning. But I also realized that AI and ML are difficult design materials, even if they are established technologies. Some of the difficulties that I constantly face involve an accurate understanding of the nature and use of ML, how to prototype it, and how to purposefully design with it. It has been argued that Human-Centered Design, when applied within the IoT domain, causes opacity and unintentionally reduces the acceptability of IoT devices.

This thesis aims at providing support for practitioners in the task of designing transparent smart objects for our future homes. Therefore, it uses mixed methods under the umbrella of Research-through-Design approach.
Background
Recent works in Explainable AI have found that a more transparent and explainable communication of autonomous decisions can increment acceptance and trust of IoT technology. However, in the literature, we can not find guidance for practitioners to bring XAI into design practice. Moreover, current design approaches tend to hide the complexity, resulting in opaque interfaces. The need for an interdisciplinary approach to put in communication the XAI theory with the design practice emerges, as designers lack guidance on how to employ AI and Machine Learning as design material to prototype and design for IoT devices.
Problem and Goal
Despite a general agreement that a lack of transparency on IoT devices can lead to a loss of user trust, satisfaction, and acceptance of this technology, Explainable AI doesn’t provide practical guidance for practitioners to bring XAI into design practice. Furthermore, it emphasizes the challenges that designers face in dealing with connected devices, claiming that they don’t find support in established design methods as Human-Centered Design.

The goal of this study was to present design recommendations to support practitioners in the task of designing for IoT. With this research I wanted to find out: 

How can we design connected objects that communicate their autonomous decisions transparently with users in the context of a smart home?​​​​​​​
Thing-Centered Design
The design process started with an observation from the point of view of a coffee machine, using a Things-Centered Design approach. The tool used for the observational study was Thing-ethnography, which involves the use of cameras and sensors attached to objects to collect data from everyday home practices from the object’s perspective. This session consisted of collecting video material through a single point of view: an action camera attached to a standard coffee machine. Later, a participatory design method was used with five designers to create Thing-personas, which were then used to define the product’s features.
Designing explanations
In order to explore how the design of the future connected product may communicate more transparently with humans, the study combines lessons from Explainable Artificial Intelligence. According to the product’s features, different explanations of AI decisions were generated after selecting significant scenarios: Waking up, Study break, Proactivity, and Data usage. 

The second phase consisted of a survey created to understand what information users may find useful and the preferred level of detail for each scenario.
Design Fiction
For its ability to take care of its owner, the coffee machine interface was designed to be classy, exclusive, inspired by the actual high-end brands in the market. Besides being smart, the most differentiating feature was reutilizing the coffee grounds to diffuse coffee aroma during the preparation, so it was called Aroma. Aroma is a product of a near but plausible future, when smart home objects will no longer be seen as a trend, but they will be established in every home.

A logo and a fictional Amazon product page were created to instill in the potential user the idea of a plausible fictional context built around the product.
Interface Design
The first step in building the prototype was designing the Graphical User Interface. The tool used for this purpose was Adobe XD, which allows to draw the artboard and connect them into an interactive prototype that can be shared with a link or via its native application. Four user flows were selected, corresponding to the four scenarios previously described: Waking up, Study break, Proactivity, and Data usage.
Evaluation and results
This concept has been qualitatively evaluated through a structured interview process. I asked the participant to describe their understanding of AI through the interaction with the prototype. Results showed that some participants did not always improve their understanding, but they were able to interpret how the machine arrived at the solution. Less skilled participants also mentioned that they learned new notions about AI by interacting with the artifact, demonstrating that the level of interpretability was acceptable.

During the interview process, the participants had also to complete a trust questionnaire. A comparison was made between the trust questionnaire taken before interacting with the prototype and after the interaction, demonstrating an average increase in trust by a substantial 18,4%.
Conclusion
The results of the study have exceeded the expectations, indicating that the design process has reached the final goal of designing smart home devices with transparency, avoiding the black box problem.

This thesis project allowed me to gain experience with AI and Machine Learning. I realized that in my work I was trapped in the standardization of methods that designers take for granted. But IoT is a fast-evolving field and conducting academic research gave me the opportunity to discover and put into practice what may become a new standard.  
Back to Top