MBZUAI team awarded Google Academic Research Award to study loneliness in the age of AI

A team from MBZUAI has received a Google Academic Research Award (GARA) for a groundbreaking project that will explore how artificial intelligence can better understand and respond to human loneliness in digital spaces. 

The project, A Psychosocial Loneliness Framework for Safer AI Companionship, is led by Thamar Solorio,  Vice Provost of Faculty Excellence and Advancement and Professor of Natural Language Processing (NLP); Monojit Choudhury, Professor of NLP; and postdoctoral researcher Aseem Srivastava. This work is in collaboration with a team from the Georgia Institute of Technology led by Professor Munmun De Choudhury. Together, they will examine how loneliness is expressed online, how conversational agents can detect it, and what healthier AI companionship could look like. 

The GARA program provides funding to leading researchers around the world to advance cutting-edge work in computer science and related fields. The awards support projects that align with Google’s areas of interest – from NLP and health to sustainability and education – fostering collaboration between academia and industry.  

Each winning proposal is selected through a competitive process and receives direct financial support to pursue high-impact, open research that can benefit both the scientific community and society at large. 

“Loneliness is turning into an epidemic,” says Choudhury. “One out of six people on the planet suffer from loneliness at some point in time. As people are increasingly using chatbots and AI companions to express their feelings, especially when they are lonely, we wanted to ask: is it healthy in the long run? And can chatbots even understand whether the user is expressing something to do with loneliness?” 

Defining loneliness 

Solorio explains that while there has been progress in applying NLP to mental health, there remains a gap in addressing loneliness. “For mental health, such as depression, there’s a significant amount of research but not on understanding loneliness,” she says. “Digital loneliness is different. We interact more with our devices, but that doesn’t alleviate the problem of not having deeper human connections. In fact, it may amplify it.” 

The team is seeking to define what loneliness looks like in the digital age, where interactions with chatbots are very different from earlier online “third places” such as chatrooms or forums. 

“In those cases, there is a person who is interacting on the other side,” says Choudhury. “The equation completely changes when it is AI. And the bigger problem is that AI has this sycophantic behavior. If you say you are lonely, the chatbot says, ‘Oh, I’m so sorry to hear you are lonely. You should talk to me more.’ They are designed to hook the user.” 

Srivastava highlights the difficulty of formally capturing loneliness in online conversations. “We see a lot of people expressing loneliness on Reddit, for example,” he says. “But there is a blurred line where exactly we need to define whether a post is loneliness-related or not. That is where the challenges lie, and we are trying to solve that by creating a taxonomic understanding of the complete loneliness structure – how this social disconnection functions in an online setting.” 

The researchers also plan to explore how loneliness is expressed differently across cultures, languages, and social contexts. 

“Cultures are structured very differently,” says Choudhury. “Some are collectivist, family-oriented societies, others are more individual-oriented. Depending on that, even the expressions or the symptoms of loneliness vary a lot. This is little understood in physical spaces, but much less in digital spaces. And when it comes to chatbots – not understood at all.” 

Solorio adds that these differences can be especially pronounced for non-native English speakers. “The way I feel things in Spanish is not the way I can express them in English,” she says. “So even distinguishing what people really mean in their cultural and linguistic context is important.” 

Measuring success 

Looking ahead, the team envisions AI companions that not only engage users but actively encourage them to seek meaningful human interaction. 

“As a result of efforts like this, I would love that when a user is engaging with a chatbot, the bot is aware of the context and history of that interaction,” says Solorio. “And then it can prod the user to say: have you talked to someone? How about you set aside time to have in-person interactions with your friends or family? Then we know it’s not replaceable by the chatbot.” 

Choudhury adds: “I would consider it a great success if we could have benchmarks and a set of policies that any company rolling out a conversation agent can use to evaluate whether their system is loneliness-compliant.” 

Srivastava also emphasizes the human stakes: “People are forming relationships with AI companions and that is a booming space. There is a debate about whether it is right or not – but people are doing it. We just want that connection to be more meaningful. That’s my vision.” 

This latest recognition highlights the University’s growing global profile and commitment to advancing AI for social good. 

“By developing this technology, we can support human experts to be more efficient or maybe scale what they do, because we know there’s a shortage of psychologists and therapies that can support humans,” says Solorio. 

“If therapy can be made more accessible, cheaper, but reliable – in a sound and ethical way – that would be great.”