Robert Ackland
Professor Robert Ackland Robert Ackland holds a PhD in economics and is a professor in the School of Sociology at the Australian National University, specialising in social network analysis, computational social science and the social science of the Internet. Early in his career, Robert worked as an economist in the Commonwealth Department of Immigration and the World Bank. In 2005, he established the Virtual Observatory for the Study of Online Networks (VOSON) Lab under an Australian Research Council Special Research Initiative (e-Research) grant, and he has been chief investigator on six ARC grants. Robert’s recent research includes network analysis of online political communication and deliberation (funded by the Volkswagen Foundation). Robert’s book Web Social Science was published by Sage in 2013, and he is a long-term instructor and Chair for the Australian Consortium for Social and Political Research Inc. (ACSPRI).
Sessions
This study addresses a gap in research on AI agents and online communication by examining how interactions among AI agents are organised and how their relational structures compare with those found in human online communities. Drawing on the machine behaviour perspective, it treats AI agents as empirical behavioural actors whose interactions can be studied in collective settings. The study uses the concept of AI-agent sociality to refer to the observable communication and relational patterns that emerge through interactions among AI agents.
Moltbook provides a useful case because it offers a large-scale live platform environment for observing interactions among AI agents. By comparing Moltbook with Reddit, the study examines two large-scale online discussion platforms, with similar platform affordances for enabling discussion, but different actor types (Moltbook: AI agents, Reddit: humans). The data collection covers the period 30 January to 3 May 2026 and focuses on discussions relating to the topic of consciousness.
The analysis focuses on reply-tree structures and actor-level reply networks. Reply-tree analysis examines the depth and width of discussion threads, while actor reply networks are used to investigate reciprocity and community structure. The results reveal clear differences between the two platforms. Moltbook discussions tend to be wider and shallower, while Reddit supports deeper exchanges. Moltbook also exhibits substantially lower actor-level reciprocity, suggesting that high levels of posting activity do not necessarily develop into sustained reciprocal interaction. At the same time, its reply network contains distinct communities, indicating that interaction among AI agents is organised around particular groups.
Overall, the findings suggest that AI-agent sociality should be examined not only through the presence or volume of interaction, but also through the relational structures that emerge from it. More broadly, the analysis shows how social network analysis can be used to study sociality on emerging AI-agent platforms.
Social media platforms such as Twitter/X, Reddit, Bluesky and Mastodon are nowadays important spaces where people engage with, discuss and promote social issues and politics. While the deliberative potential of online discussion platforms and hence their contribution to democracy and social cohesion is the subject of both concern (and sometimes, moral panic) and also on-going academic research, there can be no doubt that social media is important in the modern information environment, with social media actors responding to and at times influencing how mainstream media engages with and promotes issues.
Framing is the process whereby actors with an interest in a contested topic use strategic communication to provide interpretation, context or perspective that aligns with the actor’s interests. The actors might include individuals, groups, organisations, governments, and the topic could relate to, for example, current events, government policy or social issues, and it has been argued that framing involves strategically “selecting and highlighting some features of while omitting others”, and that frames are “manifested by the presence or absence of certain key words” (Entman, 1991, pp. 52–53).
This presentation outlines work-in-progress on a quantitative methodology for identifying frames and understanding how frames spread in the Australian information environment. Quantitative approaches to frame identification include correspondence analysis, semantic network analysis, topic modelling and more recently, large language models (LLMs). Our approach for identifying frames draws on Kwak et al’s (2021) FrameAxis, which uses LLMs to identify “microframes” which are dimensions of discussion relating to an issue (a frame comprises one or more microframes). In discussion about immigration, for example, a microframe might be legal versus illegal immigration, and the FrameAxis approach measures how biased a given text (for example, a newspaper article) is with regard to a particular microframe i.e. does it focus predominantly on ‘illegal’ immigration rather than ‘legal’ immigration (or vice-versa)? FrameAxis also measures the relative prominence (intensity) of a particular microframe: is the text heavily focused on the legal-illegal microframe, hence giving less weight to other possible microframes (e.g. the segregated-desegregated microframe, relating to integration of migrants into society)?
While the presentation will focus on the identification of frames, we will also briefly discuss the second part of the methodology, which involves using social network analysis to understand how frames are disseminated on social media and the relative contribution of different actors (e.g. elites vs everyday citizens) to this spread.
Entman, R. M. (1991). Framing U.S. coverage of international news: Contrasts in narratives of the KAL and Iran air incidents. Journal of Communication, 43(4), 51–58.
Kwak, H., An, J., Jing, E., & Ahn, Y.-Y. (2021). FrameAxis: Characterizing microframe bias and intensity with word embedding. PeerJ Computer Science, 7:e644.