21  Intelligent Machines Are Social Actors

Milena Tsvetkova, London School of Economics and Political Science,

Abstract: AI is changing not only how social scientists do research but also the social reality they study. Intelligent machines such as social media bots, LLM chatbots, AI agents, self-driving vehicles, and robots are social actors. They act in social space, interact with humans and other machines, and affect the social outcomes of teams, groups, communities, and networks. To understand our emerging hybrid society, quantitative social science should expand its purview and study intelligent machines alongside humans. This new research agenda can build on existing theory and methods but demands an overhaul of social science education and a reassessment of social scientists’ role in AI development and regulation.

AI usage statement: No GenAI was directly used to prepare this essay. However, I used Google searches and Wikipedia for brainstorming and citations, which themselves employ AI to various extents.

21.1 Introduction

AI has affected quantitative social science in multiple ways. Starting from its now older, less attention-grabbing version known as machine learning, AI has introduced new research methods for mining, analyzing, and simulating social data at scale. Deep learning and subsequent AI developments made these methods more sophisticated, allowing us to incorporate multimodal data, detect latent meanings, and generate Turing-test-passing behavior. Large Language Models (LLMs), generative AI chatbots, and agentic AI systems are further improving methods for analyzing data. But this is not why they present a breakpoint. What is novel with the current iteration of AI is the capacity it offers to do research at scale. From synthesizing literature, to scaling up qualitative interviewing (Geiecke and Jaravel 2024), writing and reviewing manuscripts, and all the way to automating the entire research process (Engzell and Wilmers 2026), LLMs and AI agents are augmenting traditional research workflows and increasing academic productivity, for better or worse (the essays in this collection discuss the various ways in which this is happening).

It was in fact LLMs that demonstrated that sometimes a revolutionary leap could occur not because of a jolting new paradigm shift but because of gradual build-up of scale (these models had to cross a critical threshold of training data and layer size to generate emergent reasoning). Hence, it would be shortsighted of me to argue that the scaling up that AI is introducing to the research methods and the research production process is merely a next-stage advancement in an ongoing scientific evolution; a true qualitative transformation of the academic profession and the academic institution might be just around the corner. However, I would leave the speculation to others. Instead, I would like to highlight a more imminent revolution specific to quantitative social science – AI has introduced a new object of study. That new object of study is AI itself: intelligent machines are now social actors, acting and interacting on par with humans, influencing and co-producing social outcomes. Consequently, to understand our current social reality, social scientists need to start studying intelligent machines alongside humans.

21.2 AI as a structural constraint

Some qualitative social scientists and social theorists are bound to cry out “old news” here. To a scholar in Science and Technology Studies, AI is clearly a social construction whose development, deployment, adoption, use, and regulation can only be understood in terms of social processes, cultural biases, institutional frameworks, organizational constraints, etc. From the opposite direction, social anthropologists and psychologists approach technology such as AI as a medium or infrastructure that constrains and molds human perceptions, decisions, and social interactions (Suchman 2007; Turkle 2013). The perspective of AI as a structural constraint on human action has also been popular among computational social scientists in the last two decades, who have investigated the effect of recommendation systems, personalized search engines, and prediction algorithms on informational isolation, political polarization, and bias in decision making, among others.

Thus, social scientists have and will continue to study AI as a macro-level social phenomenon that both affects and is affected by social action. The radically new angle I am suggesting here is approaching AI as a social actor. Thinking about contemporary LLM chat bots makes this idea intuitive: the news keeps bombarding us with anecdotes about regular folks who anthropomorphize their AI assistants and build (often unhealthy) relationships with them. But I have a broader definition of intelligent machines in mind: online political bots, self-driving vehicles, robot vacuums, trading algorithms, military drones, etc. These artificial agents may be encased in shiny futuristic bodies or exist virtually in digital space, powered by sophisticated neural network algorithms or simple if-then scripts, and offering general human-like conversational capabilities or repetitive high-precision skills. They may be human-guided or human-supervised and not entirely autonomous, and we should never forget that there is always a human or corporate entity behind them, whether in the form of an engineer, programmer, client, or owner. These artificial agents are, nevertheless, actors, in the sense that they perceive information (input), make decisions (processing), and execute those decisions (output). Because these actions take place in social space (online communities, city traffic, financial markets, family homes, industrial warehouses, etc.), they are necessarily social. And social actions and interactions combine to influence and produce social outcomes. Understanding how this happens is social scientists’ domain.

21.3 A research agenda

To steer their thinking and research into this new direction, social scientists need to go back several steps to escape the trap of the current hype fueled by financial speculation, managerial hysteria, political techno-alarmism, and workers’ backlash. First, forget about the term “AI” as it is in the singular and also ill-defined both among the public and amid scholars; use the term “intelligent machines” because it clearly conveys the idea that artificial agents are plural, diverse, interactive, and potentially embodied. Second, take in a healthy dose of anthropomorphization because humans do: to understand interactions with intelligent machines, one can draw a parallel to humanity’s relationship with animals [DeMello (2012); Darling (2021)). Third, dedust the social psychology and behavioral economics tomes because much of what behavioral and social scientists know about cognitive biases and intergroup relations can be revised and applied to human-machine interactions. Similarly, many of the social research methods scholars have used in the past to study humans and human society can be repurposed to investigate machine behavior, human-machine interactions, and hybrid groups and communities.

Equipped with current social science theory and methods, social scientists can tackle a band of new research questions:

How do machines behave? Cognitive and personality psychologists are starting to investigate emergent LLM behavior such as cognitive biases, higher-order reasoning, and personality features (Serapio-Garcı́a et al. 2023; Webb et al. 2023). Simultaneously, social psychologists and behavioral economists are studying LLM’s tendency to cooperate, reciprocate, regard others’ welfare, and identify with groups (Leng and Yuan 2023; Fontana et al. 2025). In the tradition of Turing’s and Minsky’s pursuit of human-like artificial intelligence, much of the current focus is on the extent to which AI can substitute humans. However, a more productive path forward is to think of AI as complementing humans. Just as humanity came to benefit from the power of the ox, the vigilance of the wolf, and the speed of the horse, it has much to gain from intelligent machines that are different from us in terms of persistence, predictability, information processing capacity, and error-free execution, as well as self-awareness, emotional attachment, susceptibility to influence, and adaptivity. Machines’ behavior is often most consequential when it supplements or counteracts human behavior (Tsvetkova et al. 2024).

How do humans treat intelligent machines compared to how they treat other humans? Already in the 1990s, sociologist and human-computer interaction pioneer Clifford Nass emphasized the similarity between human-computer and human-human interactions (Nass et al. 1994), and recent research keeps uncovering parallels (Bazazi et al. 2025; Mutzner et al. 2026): for instance, people say please, reciprocate, and apply gender stereotypes to artificial agents. But there are also important differences that we need to study and understand further. One well established by now finding is that people show lower emotional commitment to and cooperate less with artificial agents (Chugunova and Sele 2022; Makovi et al. 2025). As artificial agents proliferate in online forums, office communication, student work, job interviews, and dating markets, this behavior is bound to spill over to human-human interactions, decreasing generalized trust, informal social participation, public engagement, and achievement motivation. It will be impossible to comprehend human society in the 21st century without studying and understanding human-machine interactions.

How do machines interact with each other? Not all intelligent machines are explicitly designed for social interaction and even fewer have been specifically instructed how to behave towards other machines versus humans. Nevertheless, any action in a shared social space is social because it can indirectly affect other actors and their actions. Thus, even the simplest bots and scripts that operate in a common environment may end up facing off in unintended ways (Tsvetkova et al. 2017). The problem of how machines interact with other machines is especially salient when it comes to multi-agent AI systems: orchestrating efficient network patterns and workflows for AI agents benefits from an informed consideration and targeted application of concepts such as flat organization, chain of command, team composition, redundancy, and social brokerage from management and organization studies. Approaching intelligent machines as social actors opens up a whole new class of social interactions to identify, describe, design, and regulate. Social scientists have a role to play in this endeavor.

How do intelligent machines affect the teams, groups, and communities they participate in? Machines’ direct actions and indirect influence can alter social dynamics in unintended and unexpected ways, with short-term effects that contradict the long-term consequences. Trading algorithms make financial markets more efficient but also more volatile, self-driving vehicles make driving safer but also prone to traffic jams, while AI assistants improve office productivity in some ways but also introduce new workloads (Tsvetkova et al. 2024). Understanding and evaluating the effects of intelligent machines in hybrid communities requires a systematic and rigorous comparison to human-only communities, as well as machine-only communities. Perhaps there are realms of societal activity where human presence is no longer needed and multi-agent systems can take over to the effect of smoother operations and higher efficiency. At the same time, there are aspects of social life where meaning-making and personal connection remain the goal and hence, machine interference is undesirable and unproductive. Social scientists have a duty to uncover problems and offer solutions as society adjusts to the surge of artificial social actors.

21.4 Concluding remarks

Quantitative social scientists can lead the way in this new research direction as they already have experience working with artificial agents. From imputing missing survey data to simulations with agent-based-models and controlled experiments with bot confederates, quantitative social researchers have been developing and using synthetic representations of humans to study human behavior, interactions, and networks. Bots and algorithms have left the model world, however, and now roam freely among us, making decisions and taking actions that are changing social reality. Intelligent machines have become social actors in their own right and as such, they merit social scientists’ curiosity and attention. The future of quantitative social science involves the study of intelligent machines as social units alongside human individuals, households, organizations, and nation states.

This carries consequences for the practice and profession of quantitative social science. Computer programming and algorithm design should become essential training for social science students. Cross-disciplinary collaborations between social scientists and AI engineers and roboticists should be encouraged and funded. And social science professionals should be required on research and development teams working on large-language models, agentic AI systems, autonomous vehicles, and robots. It is time for social scientists to take on a proactive role, informing design decisions and preventing unintended effects, rather than the usual reactive stance, discovering problems and demanding solutions. New players have entered the game, making the game more complex. Intelligent machines are introducing new types of behaviors and interactions, making social outcomes more unintuitive and unpredictable. Social scientists have more to observe and study but also more opportunities to participate and make a difference. The future of quantitative social science will be busy.