Researching society and culture with generative AI
Guest post by Carol Rivas
Artificial Intelligence (AI) technologies are evolving at a remarkable pace; what was cutting-edge yesterday may be obsolete today. In 2014 I led a study that took months to develop an AI-type method for analysing text that could only be applied to healthcare surveys. That approach, once innovative, is now significantly surpassed by freely available AI tools that can respond instantly to a typed question in normal language on any topic or data.
The fast-changing nature of AI shaped our editorial decisions for the 5th edition of Researching Society and Culture; we chose not to explore AI in depth, focusing on pragmatic advice for navigating real-world research complexities. AI was mentioned briefly in chapters on ethics, literature reviews, writing, and analysis, acknowledging its relevance, without attempting to capture a moving target. But here we give a short overview of current AI capabilities for researching society and culture.
Origins of GenAI and social science research
Back in 2014, the term AI was reserved for Artificial General Intelligence (AGI), a still-unrealised form of adaptable, human-like intelligence. What is nowadays colloquially called AI is more accurately termed ‘generative AI’ or GenAI. This refers to computer models trained to find patterns in data presented to them (predominantly using machine learning) and to then generate new content from new data or from simple written requests by mimicking those patterns using statistical techniques. Understanding this distinction is crucial; although powerful, GenAI systems are narrowly task-specific and rely heavily on human input and statistical approaches, lacking the adaptable intelligence of AGI.
There are two common forms of GenAI. Large language models (LLMs) such as ChatGPT are most frequently used to generate text or analyse data including images. Diffusion models are typically used to modify and generate images, video and audio.
GenAI was first adopted within social sciences research for its ability to process vast amounts of data, such as text, images, and social media; this is known as Big Data analytics. GenAI has a synergistic relationship with Big Data, which is used to train it to analyse more Big Data. Ironically, technologies like GenAI, wearable sensors and the Internet of Things (IoT) have accelerated Big Data growth, creating a feedback loop that increases reliance on GenAI.
GenAI capabilities
The processing power of GenAI can give us rapid insights into cultural shifts, public opinion, and emerging trends, and can be time-saving, for example when summarising or coding large datasets. The auto-coding within Computer Assisted Qualitative Data Analysis (CAQDAS) software, an early form of this, looked for keywords; nowadays ‘natural language processing’ (NLP) systems give the illusion of actually interpreting complex text. For example, they can distinguish between a trial run, a medication trial, and a court trial in complex sentences through linguistic context (word patterns and parts-of-speech, such as verbs), using statistical techniques to group the data. We humans then name these groups using our own conceptualisations; if we encode those names within the GenAI NLP programming and they don’t include the original words, it can seem as if GenAI exhibits real understanding.
Beyond analysis, GenAI can automate mundane tasks, such as producing contents lists for reports, or correcting grammar, freeing researchers for more conceptual or creative thinking. GenAI also creates new possibilities for digitising and analysing cultural artifacts, graphics (including medical scans) and paper-based materials (including handwritten documents).
Its capabilities have been described as transformative by researchers with disabilities or neurodivergence, supporting them with tasks they find challenging; for example, AI-generated summaries of reading lists can help prioritise sources. The Model Context Protocol (MCP), using GenAI like Claude can even engage people in seemingly natural conversations, tailored to their needs, including accent and voice.
Challenges of using GenAI in research
Despite its impressive outputs, GenAI is still learning. It can produce convincing results, but these often contain errors or ‘give-aways’ that reveal their non-human origin. More concerning are the biases embedded in GenAI systems, through every decision made in their development and use. If data used to train GenAI reflect existing societal biases, GenAI algorithms can perpetuate and amplify those. This can result in "vanilla-ization", where nuanced insights are flattened and generic or centrist conclusions normalised. The risk is heightened by GenAI’s algorithmic monoculture, that is, a reliance on just a few GenAI systems with similar or identical underlying algorithms or training data. At worst, this can lead to serious discrimination.
Ethical concerns also arise around intellectual property. Some GenAI models are trained on research outputs and artwork without permission, raising issues of misrepresentation, fraud, and loss of earnings or recognition. These unapproved uses can also disseminate sensitive information without the ethical safeguards researchers typically apply. While you might consider some of these misuses amusing or easy to spot, they can enter the IoT ecosystem, where future GenAI use incorporates them as legitimate data, a phenomenon known as ‘AI hallucination’. In other words, AI-generated outputs feed back into the system, reinforcing errors at scale.
To address these challenges, SAGE, along with universities and research institutions, have developed guidelines for responsible AI use. This requires collaboration across disciplines (e.g. AI, social sciences, humanities, and ethics). AI can and does make mistakes, though they tend to be systematic and so detectable, and human oversight remains essential. Ultimately you should aim to use AI as an assistant, to augment, not replace, human judgment and critical thinking. AI is only as good as the data and humans behind it.
Further resources on the use of AI in social science
de Manuel, A., Delgado, J., Parra Jounou, I., Ausín, T., Casacuberta, D., Cruz, M., Guersenzvaig, A., Moyano, C., Rodríguez-Arias, D., Rueda, J., & Puyol, A. (2023). Ethical assessments and mitigation strategies for biases in AI-systems used during the COVID-19 pandemic. Big Data & Society, 10(1). https://doi.org/10.1177/20539517231179199 (Original work published 2023)
Gillespie, T. (2024). Generative AI and the politics of visibility. Big Data & Society, 11(2). https://doi.org/10.1177/20539517241252131 (Original work published 2024)
Milana, F., Costanza, E., Musolesi, M., & Ayobi, A. (2025). Understanding Interaction with Machine Learning through a Thematic Analysis Coding Assistant: A User Study. Proceedings of the ACM on Human-Computer Interaction, 9(2), Article CSCW197. https://doi.org/10.1145/3711095
Riedl, C., & De Cremer, D. (2025). AI for collective intelligence. Collective Intelligence, 4(2). https://doi.org/10.1177/26339137251328909 (Original work published 2025)
Order your copy of Researching Society and Culture (2025) by Clive Seale and Carol Rivas here
