SOHAM | Centre for Sociology of Humans and Machines

What is SOHAM?

We are living through one of the most profound shifts in human history. Devices using artificial intelligence are in our pockets and on our desks, they are managing our workplaces and moderating our conversations. This rapid integration of intelligent machines needs a new sociology that goes beyond the study of human beings and their relationships to research hybrid societies where humans and machines co-exist, collaborate, and transform society.  

SOHAM, the TCD-TU Dublin Joint Centre for the Sociology of Humans and Machines, was founded in April 2025 to address exactly this need. Our mission is to build a new sociology to study the emerging world in which people and AI systems think, decide, and act side by side. 

The questions SOHAM asks do not belong to any single field:

• How do gender biases travel from human culture into AI systems?
• When does artificial intelligence make groups smarter and when does it make them worse?
• How should platforms be designed to support democratic discourse rather than undermine it?

These are simultaneously technical, social, ethical, and political questions and they need a multidisciplinary response. SOHAM gathers researchers from the humanities, sociology, psychology, data science, physics, mathematics, philosophy and engineering to explore how humans and intelligent systems can be understood, designed, governed and harnessed for a better future. 

What are some of SOHAM's research highlights?

Gender and AI

SOHAM has only just completed its first full year and already it is clear that human social patterns, our biases, our norms and our political identities, do not disappear when we interact with machines. This is particularly clear in our work on gender and AI.

• In a series of studies, SOHAM researchers have shown that people bring gender stereotypes into their interactions with artificial intelligence.
• In one study of over 400 participants, people were more likely to exploit AI systems labelled as female and more likely to distrust those labelled as male. This mirrored and even amplified the biases we see in human workplaces.
• A related study found that people reacted more negatively when a female AI manager decided not to give them a reward, even when the decision was identical to one made by a male AI manager.  

We have found that the way an AI is presented matters. When an AI system was described in text, people evaluated it on competence and gender doesn’t really matter. But when the same AI system is shown with a human-like face, gendered perceptions emerge and competence information became less influential. 

These findings carry real implications for how AI products are designed and regulated. If adding a face to a chatbot introduces gender bias that wasn't there before, that is a design choice with social consequences. Policymakers need to take this research into account. 

In one study, we analysed a year's worth of data from a social media platform populated entirely by autonomous language model agents. Despite the absence of any human users, stable patterns of gendered behaviour emerged through interaction. The work highlights how human cultural patterns embedded in language can organize collective behaviour in AI systems. Even in an all-machine environment, our human social world reasserted itself. 

AI Agents Working With Human Peers

A second strand of our research examines what happens when AI agents act alongside humans as peers. A study using experimental public-goods games found that when a single AI agent is in the minority, group behaviour is driven primarily by human normative dynamics. Whether or not people know their teammate is an AI matters less than the social norms already in play. This has important implications for how we introduce AI into workplaces and collaborative settings. 

Online Information

A third area of focus is online information and the challenge of keeping public discourse healthy. SOHAM has produced a comprehensive resource on Community Notes, the community-based content moderation system originally developed by Twitter and now used on X, Facebook and Instagram. We have shown that collaborative note writing produces better-quality contextual notes than individual efforts. However, this benefit disappears when contributors know each other's political affiliations. Political identity, in other words, can undermine the very collective intelligence that makes community moderation valuable. Understanding this dynamic is essential for platforms seeking to design moderation systems that are both effective and politically resilient. 

You can read more about our research and find links to the papers and preprints in our newsletter. You can sign up for our bi-monthly newsletter here. 

How might SOHAM improve our future?

The questions SOHAM asks are not only academic, they are consequential and practical questions and the answers have the potential to shape policy, platform design and public life in tangible ways. 

Tackling Misinformation

Take for example the challenge of misinformation. Social media platforms are struggling to keep up with the volume of false or misleading content. SOHAM's research on community-based moderation offers a way forward, not by replacing human judgement with algorithms, but by understanding how to harness human judgement better. 

AI has the potential to enhance humanity’s capacity for collective intelligence which could help communities make fairer and/or faster decisions. But AI can also undermine collective intelligence. SOHAM's research is mapping these dynamics so that the conditions under which AI genuinely helps can be identified and built upon. 

Addressing Climate Change

Many of our most pressing environmental problems are, at their core, problems of collective action, of getting large numbers of people and institutions to cooperate in pursuit of a shared goal. SOHAM is exploring how mechanisms can promote the kind of large-scale cooperation that could address climate change. 

Integrating With the Labour Market

AI is increasingly being used in decision-making in a wide variety of applications and sectors. SOHAM's research shows that AI systems reflect the biases embedded in their design, their training data, and their interfaces. Understanding this is the first step towards designing AI systems that are genuinely fair. 

Across all of these areas, SOHAM’s aim is to ensure that as machines become more powerful, our societies become more fair and more resilient.

Visit SOHAM's website