Embrace the formation of social relations in old age.

Explore how to increase tenfold the opportunity for older adults to stabilise their social convoy.

We aim to disrupt the steadily decreasing social convoy in old age by exploring an intelligent matchmaking algorithm that promotes long-lasting friendships by matching users from a broader social network.

Social convoy theory holds that people maintain a network of social relationships that escorts them over the life course like a convoy, like a group of fellow travellers on the road of life. Relationships in this convoy differ in levels of closeness and dependency on social circumstances and are therefore differently affected by changes in a person’s circumstances. Relationships with people in the innermost circle of the convoy, such as spouses and the core family, should be highly stable throughout the lifespan. Relationships in the periphery of the convoy, such as acquaintances, co-workers, and neighbours, are assumed to be less stable. These relationships may end with external circumstances, such as changes in social roles or location. Various studies provide first support for convoy theory for late adulthood: Close, core relationships remained stable, whereas peripheral relationships decreased in number and contact frequency.

Explore how to increase tenfold the opportunity for older adults to interact directly within a diverse social network.

We aim to explore an intelligent matchmaking algorithm that initiates and facilitates social interactions, directly engaging users within a diverse network to promote well-being and quality of life.

The link between health and social networks has gained attention in the scientific literature over the last two to three decades. Social networks are moderators of disease and recovery histories of ageing-typical diseases such as heart disease, dementia, and cancer. Also, empirical results across populations and different network types show that a complex social network structure enriches well-being; well-being is lowest in older people with small, socially isolated, restricted networks. Although specific structures appear associated with certain functions, we found notable heterogeneity, particularly in friend-focused and limited network types. Diversity in networks of family and friends, on the other hand, goes hand in hand with increased well-being, both mentally and physically.

Explore how to increase tenfold the resilience in social network ecosystems to ensure the frequency of social interaction.

We aim to explore a scalable matchmaking system with existing social network coordinators to ensure the frequency of social interactions in social network ecosystems and support equality and the freedom to choose with whom, for how long and how often individuals want to engage in a social network.

Meta-analysis on age-related differences and changes in social networks shows that the global network increases in adolescence and young adulthood, reaching a plateau in the mid-20s to early 30s and shrinking after that together with the number of named friends. The personal and friendship networks decreased throughout adulthood by almost one person per age decade (average size of friends’ network in the 55+ age group: 4.5). The family network was stable in size from adolescence to old age. Meaning that people manage to keep their family networks relatively stable, especially to the highest age, while friends, for example, are more likely to be lost.

Also, looking beyond the Coronavirus pandemic, we face several forecasted global risk factors like climate change with extreme weather and human-made disasters that can impact our society with natural disasters and infectious diseases. This risk factor is accompanied by research showing that older age in association with chronic illness or multimorbidity and/or unfavourable social, socioeconomic, and environmental conditions are significant risk factors for social isolation and loneliness.

A social agent adapts to users changing moods and well-being over time and captures matchmaking dimensions relevant to human behaviour and satisfaction.

The use of natural language processing techniques and the output from our co-creation and user involvement processes will serve in a cluster analysis that will train a machine learning (ML) model. The purpose of this model is to identify questions to be used by the matchmaking algorithm. Initially, we will use supervised learning – but aim to build a fully unsupervised matchmaking algorithm capable of producing decision tree questions by itself. Our further aim is to develop a robust training set that improves the matchmaking algorithm from the very first beginning.

A series of breakthroughs in machine learning (ML) has led to a surge in artificial intelligence research progress in the last few years. Despite this progress, ML still has a critical shortcoming: a lack of social intelligence. Personal assistants cannot understand the meaning behind a user’s tone of voice, and recommender systems cannot improve by adapting to the user’s changing mood and wellbeing over time.

The HannaH project is pursuing social and emotional AI to provide the immediate benefit of a system that can collaborate more effectively with users in the matchmaking process. We consider an AI that is intrinsically motivated to produce positive social responses in the humans it interacts with. When a user responds with an angry or frustrated tone, this could act as a negative incentive, training the model not to repeat the action that led to the user’s frustration. Rather than requiring the user to train the device manually, such passive sensing of the user’s emotional state could allow the model to learn quickly and at scale, enabling human-in-the-loop training without extra human effort.