Every day, our brains take even the smallest behavioural clues from those around us to build a picture of what they might be thinking or feeling. We generalise this information to help make sense of others and even ourselves, a process called ‘self-other generalisation’.
In this interview, Joe Barnby of King’s College London, UK, discusses his recent eLife article on self-other generalisation, describes what happens when the process goes wrong, and explains what modelling experiments can tell us about human relationships more broadly.
Joe Barnby. Image from Social Computation and Representation (SoCR) lab.
What was your recent eLife article about?
Self-other generalisation is a process that allows us to build a stable picture of how people behave – and how we should behave – and this helps us to navigate social exchanges with efficiency and adaptability. For some people, however, this process can go awry.
Borderline personality disorder (BPD) is a diagnosis provided to help support those who often experience distressing, sudden and intense shifts in how they feel about others in response to tiny events that wouldn’t normally have such a big impact. Why does this happen? Can theories about how we learn and integrate social information help us explain this?
How did you go about investigating this?
Myself and Jen Nguyen, a clinical psychologist, worked with colleagues at various institutes in the UK, Germany and the US*. Data collected by Julia Griem, Tobias Nolte, and Peter Fonagy were particularly important for the project. We designed an interactive social task called the Intentions Game and used it to study how people with and without a diagnosis of BPD use what they know about themselves and others to make decisions. We wanted to ask if the actions of others caused people to adapt their beliefs and behaviours during social interactions. What we found was striking: people with and without BPD use very different strategies to understand others.
In people without a diagnosis of BPD, beliefs about the self and others are blended, each helping shape the other. But in people with BPD, these beliefs remain separate, and someone with BPD is less likely to update their beliefs based on someone else’s actions. We also found that early life trauma can dictate how much others influence us: more trauma equates to less change.
Can you tell us more about the Intentions Game?
The game has its origins in a paper on social values which I published in 2022 with Nichola Raihani and Peter Dayan, and it draws on previous research on theories of social insertion and social contagion. Henry Burgess, a software engineer at Washington University School of Medicine in St. Louis, US, working with Linda Richards, helped turn the original version into a gamified, aesthetically pleasing and highly controlled experimental setup.
The game has three phases. Participants play with anonymous partners to gain points, with a total score of over 1,000 entering them into a lottery for a cash prize. The ‘partners’ are simulated based on the choices of the participant, allowing us to tightly control the experiment.
Participants start the game as a decider, choosing between two options for distributing points between themselves and their ‘partner’ over 36 rounds. The options are a mixture of prosocial (where both players get the same number of points), individualistic (both players score high, but the decider gets the most points), or competitive (the decider favours the option with lower points but that maximises their points relative to their partner).
In the second phase, participants play again with a new partner, this time as the recipient. This ‘partner’ is programmed to behave about 50% differently based on the participant’s choices in phase one. The participant is asked to predict their new partner’s choices before each round. Then, after the game, they rate how selfish they believed their partner had been. In the final phase, the game is repeated as in phase one with a new partner, and the participant as the decider. The game allows us to test whether people updated their predictions and strategies based on the actions of others.
What inspired and motivated you?
Understanding other people is hard. But somehow, from just a few small clues – a look, a hesitation, a word – we manage to build rich ideas about what others are thinking or feeling. This ability helps us make sense of a complicated social world. The great unknown of how and why humans do this so efficiently is a huge motivation for me, my group, and my collaborators.
Our goal was to develop a model that could capture how people transfer social information in both directions: from self to other, and from other to self. Did people transfer back and forth as expected, or just in one direction, or not at all? We also wanted to explore whether conditions characterised by unstable personal relationships, such as BPD, can be better understood through this framework. We were inspired to bridge the gaps in a range of previous research, from early psychodynamic studies, cognitive science, and very recent computational work.
What might happen next?
The approach as a whole encapsulates the use of precision modelling to explain complex social dynamics that frequently underpin social life. The findings also raise important new questions: i) How does the way we generalise between ourselves and others change as we grow up? ii) Could early life trauma affect how we build a stable sense of self and others? iii) What social or environmental factors help or hinder these abilities?
Importantly, these answers can help scientists and clinicians understand and support those who seek clinical help. However, I believe our research is relevant to human social relationships in general, opening the door to a deeper understanding of how we build relationships and turn sparse information into complex intentional maps of others.
Interview by Daisy Veysey.
About Joe Barnby
Joe is a senior lecturer (associate professor) and director of the Social Computation and Representation Lab at King’s College London, UK. He is also a senior research fellow at the Centre for Artificial Intelligence and Machine Learning, Edith Cowan University, Australia.
*The other authors of ‘Self-other generalisation shapes social interaction and is disrupted in borderline personality disorder’ were: Jen Nguyen, Julia Greim, Magdalena Wloszek, Henry Burgess, Linda J Richards, Jessica Kingston, Gavin Cooper, P Read Montague, Peter Dayan, Tobias Nolte, Peter Fonagy, and the London Personality and Mood Disorders Consortium.
