The growth of dating apps has changed the way relationships are formed. These applications have enabled users to widen their dating pool of potential partners through profile recommendations based on proximity and shared interests. However, dating app usage was also found to impact well-being negatively, with the AI systems of these platforms, contributing to it. In this project it was therefore explored how well-being can be enhanced, by mitigating the detrimental impacts of dating apps.

Contextualising well-being
Literature research was conducted to explore the features within a dating app along with its effect on well-being. By linking the various experiences with dating apps to well-being theories, a theoretical model could be developed. This contextual model of well-being was validated in a participatory workshop with dating app users. Findings from this workshop indicated that autonomy and relatedness were mostly affected, especially during the profile setup, which also influenced the overall app experience such as during swiping.

Operationalising well-being
As relevant constructs were identified for further study, it was necessary to determine how these facets of well-being could be measured, in particular for the context of dating apps. In-depth research on the relationship between autonomy and relatedness indicated that in order for people to become more autonomous (showing one’s true self to others), the need for relatedness (being understood or feeling a sense of belonging) had to be fulfilled first. For dating apps, a sense of relatedness could be achieved through sharing similar interests (or by experiencing a sense of belonging) with others, while autonomy relied on choosing qualities that are not necessarily desirable, but valuable to the user themselves (when it concerned self-expression).

Optimising well-being
Based on these insights, a design direction was formulated. This concerned, enhancing social connection by highlighting individuality (uniqueness of a person) within similarity (through a shared connection). This direction was used as a base for ideation. AI Ideation cards, which showcased the capabilities of the technology, were utilised. The possible interactions enabled by the technology were linked to desirable well-being outcomes defined previously.

Implementing well-being
This resulted in the design of a new dating app, MiHue. The app highlights a person’s unique attribute and tries to find a common ground with others either through an uncommon trait or through mutual passions. The experience of autonomy and relatedness upon this design was measured and the outcomes were used for recommending future cycles.

Initiating a new cycle
For the next cycle, it was recommended that other well-being related aspects which were not considered in the design, but present in the theoretical model (such as self-acceptance, positive emotions and physical health) to be considered for future research. Furthermore, it was advised that the involvement of minorities should also be included and that gender differences should be taken into account for the next cycle.

Is healthy eating simply the intake of correct quantities of certain nutrients, regardless of how and where? This project addresses improving wellbeing within the broad context of eating.

Eating-related guidance products have a hyper-focus on nutrition and/or weight loss, to the detriment of a wider definition of “health”. Wellbeing is not only physical health metrics, but has a deeper psychological foundation. Eating experiences should focus on this in order to achieve a truly healthy relationship with food, and an improved level of overall wellbeing. Positive AI builds on the framework of positive design in order to do this, and this project produces a tangible example of how this can be done.

This project focuses on the issue within the specific context of eating at home. The simplified context retains vital nuance, but simplifies the focus. By gaining an understanding of this and how design and AI can be applied in determined ways, actual results can be obtained. The problem is then to promote mindful home-dining experiences by providing a service in which users can feel both in control of what they eat and closer to their loved ones.

The problem is addressed by designing an app-based intervention system - FoodVibe - which harnesses the power of AI to create a highly personalised recipe and experience recommendation system. It facilitates these with the goal of improving wellbeing based on a psychological wellbeing model tailored to the specific context, and learns based on user trends and past experiences. The wellbeing model, based on design analyses of various apps in the broader eating context in combination with wellbeing theories, gives initial direction. This is then thoroughly scrutinised in an evaluation section, leading to future recommendations becoming clear.

Recommender systems have become a core strategy of countless online sites and platforms including music streaming services like Spotify. These recommender systems are associated with limiting our exposure to a narrower range of content, this is even the case for features designed to promote discovery like Spotify’s Discover Weekly. Discover Weekly is curated with the aim to match users’ current preferences, it analyses its users’ listening habits and searches for similar songs that they haven’t heard before. While presenting users with new songs, Discover Weekly keeps users from discovering new and diverse music. This project aimed to design an intervention that increases user’s exposure to diverse music and fosters wellbeing.

For this, literature research was conducted. First an understanding of the current landscape of features of music streaming service was created. This led to the conclusion that personalised recommendations were the most relevant for this project, as they are driven by recommender systems. Next, different personalised recommendation features were linked to wellbeing. It was established that mood-based recommendations have the potential to contribute to hedonism, that activity-based recommendations have the potential to contribute to personal functioning and that discovery recommendations have the potential to contribute to personal development.

Given the alignment with the broader concerns of recommender systems, the project continued to focus on discovery features and personal development.
By analysing how current recommender systems operate in these feature, it was determined how that the recommender system of the new intervention should implement an ‘exploration mode’ approach and gradually introduce users to new items, allowing them to slowly acclimate to new genres over time.
Next, the design focus was narrowed down to: promoting personal development through expanding user’s musical horizons by introducing them to diverse music that differs from their current listening behaviour, fostering greater understanding of others and oneself. This design focus let to two different concepts. The first concept is ‘Discover More’, which adopts a similar approach as Discover Weekly, an automatically created playlist that gradually introduces users with new genres. The other concept is ‘Explore More’, a concept in which users are encouraged to actively explore new genres themselves. After evaluation it was decided to continue with Explore More. This concept was further developed and after evaluation of this concept, an improved version of the concept is proposed for the next cycle.

The concept highlights the gap between the vast musical landscape and the music the user listens to, creating awareness of their musical bubble and inspiring them to start exploring new music genres. Additionally, Explore More helps users gain deeper awareness and understanding of their own preferences. This is achieved with the help a guided self-reflection. Furthermore, the feedback on the reflection is utilized by the AI in selecting next genres and aims to align with the mood and taste of the user. After several exploration journeys, Explore More provides a recap—an opportunity to revisit the musical discoveries and providing insights into remarkable trends, such as specific elements that repeatedly spoke to the user, contributing to self-awareness.

Lastly, the report ends with additional suggestions that could be taken into consideration during the next cycle.

The Netflix documentary, ‘The Social Dilemma’, has shown that ubiquitous digital platforms such as Facebook, YouTube tend to use artificial intelligence to optimize the system for user engagement. The documentary provides an example of why metrics that optimize engagement, such as time-on-site, can be detrimental to our society. Overemphasizing current metrics resulting in manipulation, short-term concerns, and other inadvertent negative consequences.

OPTIMIZE FOR WELLBEING
Why do companies not optimize for wellbeing since it is in their best interest in the long-term to keep their users subscribed to their service? The answer is that it is easier and, in the short term, profitable to measure time on-site than wellbeing. In a 2016 TED talk, Tristan Harris provided an example: “Tinder, where instead of measuring the number of swipes left and right people did, measuring the deep, romantic, fulfilling connections people created.” Here, the existing metric, the number of swipes left and right, is easy to measure; conversely, the alternative, fulfilling romantic relationship, is not. However, the goal of a dating app should be to connect two individuals meaningfully; the number of swipes not equal to a positive relationship. This problem highlights the difficulty of translating human values into feasible metrics.

DEEPER QUESTION BEHIND ETHICAL AI
What if we want ethical AI, systems that can do good to people? What if we want AI value alignment, for instance, ensuring AI systems obey human value. According to the paper “Artificial Intelligence, Values, and Alignment.”, behind each vision for ethically-aligned AI sits a deeper question. How are we to decide which principles or objectives to encode in AI and who has the right to make these decisions?

INCLUDE STAKEHOLDERS TO REACH AI VALUE ALIGNMENT
One possible way of answering this is by including stakeholders and those most impacted in the design process and combine quantitative measures with qualitative information. Columbia professor and New York Times Chief Data Scientist Chris Wiggins stated,“Since we cannot know in advance every phenomenon users will experience, we cannot know in advance what metrics will quantify these phenomena”. In other words, we first need to understand the user’s perspective of AI experience to be able to develop suitable metrics. Thus, participatory design can be a way to translate human values into wellbeing metrics that fit the context. In this project, we focus on wellbeing as our value.

PROJECT FOCUS
To operationalize wellbeing, we need to know which aspects of wellbeing require focus by giving users a voice to inform us of their perspectives. This project will focus on developing a method that could operationalize wellbeing concept for participatory design. This methodology will be applied to sensitize end-users and allow them to share their wellbeing concern freely and meaningfully. Although the motivation of this project comes from the needs of AI for wellbeing, the project focuses on a problem that exists outside of AI alone, namely, sensitizing wellbeing. Because to be able to design an AI for wellbeing system, we first need to enable people to talk about wellbeing.