The Hidden Power of AI in Building Human Connections

New Research Reveals Promising Pathways

In an age where many of us feel increasingly disconnected from one another, a latest research review has uncovered how artificial intelligence might help bridge the gap in human social interactions. The comprehensive analysis, conducted by researchers at Eindhoven University of Technology, explores the emerging role of generative AI in fostering meaningful social connections—a finding that could transform how we approach the growing epidemic of social isolation.

The Loneliness Paradox in a Connected World

Despite our seemingly endless digital connections, approximately one in three adults in Europe and the United States experiences social disconnectedness—a condition linked to poorer mental health, reduced life expectancy, and diminished overall wellbeing. This contradiction sits at the heart of modern society: we have more ways to connect than ever before, yet meaningful human connection remains elusive for many.

"Social disconnectedness has been recognized as a 'global epidemic' in recent years," note the researchers, highlighting the urgency of finding new approaches to this widespread problem. Their work examines how generative AI (GAI) technologies—systems that create content like text, images, and videos based on user requests—might offer unexpected solutions.

The timing couldn't be more relevant. Since ChatGPT burst onto the scene in 2022, bringing generative AI into mainstream awareness, questions have swirled about how these technologies might impact our social lives. While much attention has focused on potential harms, this research illuminates the positive potential of AI to strengthen human bonds.

"GAI can be used to facilitate social interactions by generating personalized content and activities that continuously adapt to users' changing needs, behaviors, and capabilities," the researchers explain. Unlike earlier technologies, today's AI systems can understand nuanced communication, adapt to individual differences, and engage across multiple modes of interaction—from text to images to speech.

This versatility makes GAI uniquely positioned to support meaningful social connections in ways previous technologies couldn't achieve.

How AI Takes Shape in Social Spaces

The research team analyzed 30 studies published since 2020, mapping out how designers are creating GAI applications specifically to enhance social interactions. They discovered these applications fall into three distinct categories based on how the technology functions in social settings:

AI as a Social Participant

In most cases (19 of 30 studies), AI took on the role of an active social agent—essentially becoming a conversation partner. These applications, typically chatbots or embodied conversational agents (like social robots), directly engaged with users as social actors.

For example, researchers developed MindTalker, an audio-based chatbot created using GPT-4 to support reminiscence therapy through meaningful conversations about personal photos with people experiencing early-stage dementia. Another system used LLM-driven chatbots in virtual reality environments to help autistic individuals practice job communication skills.

These AI agents were designed to create experiences of intimacy and immediacy—key components of social presence that can foster feelings of connection.

AI as a Communication Channel

In some applications, AI served as a medium through which human-to-human interaction occurred. SocializeChat, for instance, helps individuals with speech and motor impairments communicate by offering sentence suggestions through eye-gaze technology and adapting communication styles to social context.

Rather than participating as a social actor itself, the AI in these cases functions as a bridge between people who might otherwise struggle to connect.

AI as a Social Facilitator

The third category includes AI systems that function as tools to enhance interaction between humans without becoming social actors themselves. These applications typically support activities like storytelling, reminiscence, or collaborative creation.

Treasurefinder, for example, is a small device that uses an LLM to create open-ended questions when users scan NFC-tagged items in their home. The device plays previously recorded stories about the objects, giving users a foundation for conversation. Another example, Closer Worlds, is a two-player game where participants take turns describing an imaginary world through text, which AI then transforms into images, interspersed with reflective questions to encourage personal sharing.

Creating Connections Across Different Contexts

The research revealed six primary social activities where GAI applications are making an impact:

1. Storytelling and storymaking: Seven applications helped users create and share stories, often as a way to facilitate intimate conversations and self-reflection. Many targeted children, using storytelling as a method for developing social-emotional skills.

2. Social-emotional skills training: Eight applications focused on helping users develop skills necessary for social interaction, such as emotion recognition and expression. For example, ChaCha, a chatbot, guides children in sharing personal stories and emotions to promote emotional understanding.

3. Reminiscence: Four applications supported reminiscence activities, particularly for older adults or people with dementia or mild cognitive impairment. These systems often generated new text and images based on memories to foster recollection and conversation.

4. Collaborative learning: Three studies used chatbots to make collaboration between people more efficient, such as facilitating smooth transitions through stages of group discussion.

5. Music making: One study focused on collaborative music composition, examining how AI-generated musical phrases affected social dynamics between users.

6. General conversation: Ten applications aimed to establish natural human-machine interaction without targeting a specific activity context.

Who Benefits Most? Focusing on Vulnerable Groups

A striking finding from the research is that more than half of the studies (17 of 30) designed GAI technologies specifically for vulnerable user groups—people who face challenges with self-reliance and social participation.

These included:

• Children aged 4 to 15 years

• People with dementia or mild cognitive impairment

• Older adults

• Autistic individuals

• People with speech and motor impairments

This focus highlights the potential of GAI to support those who might benefit most from enhanced social connection opportunities. For example, RetroMind leverages large language models and a social robot to help people with dementia engage in personalized reminiscence sessions, where AI-generated images based on their recollections stimulate memory recall and reinforce their life narrative.

The Technology Behind the Connection

While the applications varied widely in their approach, the researchers found some common technological patterns:

• Most applications (26 of 30) used GAI for text generation, primarily through GPT models

• Seven used image generation (primarily Midjourney or DALL-E)

• One focused on music generation

• Only four applications were truly multimodal, combining text and image generation

This predominance of text-based applications suggests designers are still exploring the most effective ways to harness GAI's capabilities. The researchers note that opportunities for "designing richer, embodied, and emotionally resonant interactions remain largely untapped," particularly in combining multiple types of generated output to create more immersive experiences.

Design Challenges: Who Gets a Voice?

Despite the promising applications, the research uncovered important gaps in how these technologies are being designed. Only about one-third of the reviewed studies adopted a participatory design approach, where end-users help shape the technology from the beginning.

While studies often included end-users in evaluation (20 of 30), they rarely involved them in the initial conceptualization phase (9 of 30). This was especially true for vulnerable populations, who were often excluded from early design processes due to concerns about cognitive strain or ethical considerations.

This pattern raises questions about power and voice in technology design. In many therapeutic or support contexts, GAI functions as a tool deployed by care professionals rather than one co-controlled by end-users themselves. As the researchers note, this model "raises ethical questions: who gets to shape the interaction, and whose voice is amplified?"

A few studies demonstrated that meaningful co-design is possible even with vulnerable populations when approached thoughtfully. Zhou and colleagues, for instance, allowed children to craft physical objects and use image generation to create virtual agents for storytelling, showing how inclusive design practices can lead to more effective and personally meaningful technologies.

Measuring What Matters

Another challenge identified in the research concerns how these applications are evaluated. While many aimed to foster social connectedness, relatively few measured this outcome directly. Most studies focused on system performance, usability, or feasibility rather than the deeper relational impacts of the technology.

When social interaction was assessed, it was often through limited metrics like conversation turns or thematic engagement rather than deeper measures of connection. This gap likely reflects the preliminary nature of many designs but points to the need for more sophisticated evaluation approaches that can capture the nuanced ways technology might enhance human connection.

Some promising approaches did emerge. Chen and colleagues quantitatively measured how users of their collaborative world-building game experienced intimacy and closeness with study partners. Others used qualitative methods like interviews and observations to assess social connectedness, such as Suh and colleagues, who found that some music collaboration participants "felt close to their partner in their shared humanness due to the unpredictable presence of the GAI."

Ethical Considerations: Bias, Representation, and Inclusion

The research also highlights important ethical considerations in using GAI for social connection. The widespread use of GPT-based models brings benefits but also risks amplifying biases in training data, potentially leading to stereotyping or exclusion of certain groups.

While many studies acknowledged these risks, few provided concrete strategies to address them. The researchers call for more than abstract ethical principles—designers need "practical, hands-on guidance" including "actionable design prompts, prototyping heuristics, and content guidelines that address issues such as cultural sensitivity, output curation, and bias mitigation."

This is especially critical when designing for vulnerable populations, who were the primary focus in many studies. Excluding these groups from design processes risks reinforcing harmful assumptions and undermines the relational goals these technologies aim to support.

The Future of AI-Enhanced Social Connection

Despite these challenges, the research paints an optimistic picture of GAI's potential to enhance social connection. The technology's ability to generate personalized content that adapts to changing user needs offers unique opportunities to support meaningful interaction, particularly for those who might otherwise struggle to connect.

As the researchers conclude: "By addressing current gaps and encouraging collaboration between researchers, designers, representative end-users, and other stakeholders, the future of GAI-based technology holds promise for effectively fostering social interactions for its users, thereby proactively supporting social connectedness."

The findings suggest several promising directions for future development:

1. Exploring multimodal approaches that combine text, image, sound, and other outputs to create richer, more immersive social experiences

2. Developing more inclusive design practices that meaningfully involve end-users—including vulnerable populations—from the earliest stages of technology development

3. Creating more sophisticated evaluation methods that can capture the nuanced relational impacts of technology use over time

4. Establishing practical ethical guidelines that help designers address issues of bias, representation, and inclusion in concrete ways

As social disconnection continues to affect millions worldwide, these AI-enhanced approaches to fostering human connection could become increasingly valuable tools in our collective toolkit—not replacing human connection, but helping to create new pathways toward it.