Experimental
Authorship
Narrative
Change
Reality
Check

Tracing the anxiety in the landscape of Generative AI

Created
Jan 29, 2024 7:08 AM
Author

Ruiwen Zhou

Publication year
January 29, 2024

image

Photo from Kirill Savenko on Getty Images

Tracing the anxiety in the landscape of Generative AI’

by Ruiwen Zhou

AI Project Series 1: An Anxious Encounter: Redefining author function in the age of Generative AI

In the previous blog, we delved into the nature of the anxieties existing around Generative AI (Gen-AI) and connected those anxious feelings with the change in the author function. Building upon that discussion, this blog will offer an explanation of the theoretical framework of this project, as well as a record of our current progress.

Anxiety as the starting point

Before starting our journey in search of the digital author, it is crucial to address why we have used people’s anxieties as the entry point to this landscape. As Lazarus (1991) points out, emotion, pertaining to “organized cognitive–motivational–relational configurations”, changes its status with “changes in the person–environment relationship as this is perceived and evaluated” (p.38). In this vein, the variety of anxieties that people have experienced when confronting Gen–AI can be considered prompts for different situations — these new technologies impact people in an emotional sense, and the relationship between humans and technologies is then filled with uncertainty. This is where the role of the author is changing.

Additionally, the anxieties surrounding Gen–AI play the role of the “tricks that have to be invented to make them [the objects] talk” (Latour, 2007, p.79). In other words, the feeling of uneasiness, tension, stress, and being out of control (Li et al., 2017) often accompanies the occasion where accidents and breakdowns happen, and it is these events that actively present the ways in which routines and norms are modified, making the relationship, especially the changes of the relationship, between the new technologies and cultural practices visible and thus, traceable.

The “author function” reveals the manner in which discourse is articulated on the basis of social relationships (Foucault, 1979, p.28). With this in mind, the anxiety around Gen–AI becomes a vehicle for us to enter the terrain of the author function, driving us to where the discourse exists operates, and circulates (Foucault, 1979).

Mapping with two dimensions

By using anxiety as a vehicle, we now need to find the ways in which Gen–AI is producing what we think of as reality. Therefore, we have adopted a mixed–methods approach, building upon the crisis mapping approach from Actor–Network Theory as the navigation, guiding us to the fluid connections between the new technologies and people’s daily lives.

Currently, we are in the first phase of the journey — mapping the discourse around Gen–AI to create a taxonomy of Gen–AI anxieties. By doing so, we hope to shed light on the consistency of the anxieties across cultures, and building upon that, we will create a symptomatic archive to understand the changes in author function and a new framework to frame the newly emergent digital author. Basically, there are two dimensions for mapping, namely locations and media materials.

Locating the anxiety

We chose Hong Kong, India, and Germany to localize our analysis. The significance of locating the anxieties surrounding Gen–AI in these three areas is two-fold: first, the scenarios from both the Global North and Global South allow us to investigate, distill, and integrate the discourse through the lens of various economic and political environments, along with different attitudes towards the application of Gen–AI. Second, the three sites also differ from each other in terms of cultural norms, as they are rooted in unique yet interconnected historical backgrounds, where the social role of the author functions and becomes regulated (Foucault, 1979, p.19). Accordingly, the huge and ever–changing divergence within the sites can not only enable us to capture the diversity around the discourse of Gen–AI anxieties to some degree, but more importantly, it will also contribute to the credibility and the reliability of the change in the author function, since there is a certain level of consistency across cultures in time.

The domains of anxiety

We will check the symptoms of Gen–AI anxieties in three domains: social media websites, media reports, and legal cases. While the discourse about anxieties circulates within a much wider range, our focus here is not to cover the full complicity of the variations. Think of it like a symphony orchestra concert: before it begins, we can see the musicians tune their instruments and quickly practice their sections. Although they are going to perform the same pieces of music, this preparation right before is rather chaotic, and contained to their individual contributions to the performance. Similarly, the discourse from different domains is something like a collective tuning exercise. Our mission is to be the conductor of the orchestra and make those intertwined voices come together coherently so as to gain some insights from them.

Discourse circulating within social media platforms reflects public perceptions from a relatively bottom-up perspective whilst the legal precedents, with the ability to connect a case and a general rule, reveal an up–down approach in which certain cultural practices are inscribed into the technologies to make an association last longer and extend wider (Latour, 2007; Philip, 2007). In this sense, media coverage seems to be situated at the junction as it is embedded with wider social and cultural factors (e.g., Duan & Miller, 2021; Ahmad & Sahu, 2019), and meanwhile, can mediate public’s opinions on social issues (e.g., Schlueter & Davidov, 2011). Admittedly, these domains more or less overlap and are intricately related to each other. Our purpose here is not to draw clear distinctions, but utilize the three layers as guidelines, helping us target narratives from different stakeholders. In addition, the three domains, in combination with the three countries/regions, offer a degree of triangulation and a way to double–check the validity of the taxonomy.

Tracing news stories around Gen–AI

As mentioned before, media coverage, with its intersectional location, has a rich diversity of situations where anxious encounters with Gen–AI occur. Therefore, we decided to start creating the framework with narratives in the domain of media reports, and subsequently generate the preliminary framework. We will then apply the preliminary framework to the other domains, namely social media websites and legal cases, for validation and modification.

As for the tool to compile the relevant media reports within the two dimensions, we used Media Cloud Search, a content retrieval and analysis tool powered by the open–source platform, Media Cloud. The parameters for the search and the corresponding results are as follows:

Dates: June 1, 2022 — November 20, 2023

We decided to set our time frame according to the date on which ChatGPT was launched (November 30, 2022). By extending the specific time point into a relatively longer period, we are aiming to catch the Gen–AI anxieties in various contexts, enhancing the representativity of the archive.

Keywords: “Artificial Intelligence” OR “ChatGPT” OR “Google BARD” OR “Dalle-E” OR “Midjourney” OR “Stable Diffusion” OR “Machine Learning” OR “neural network” OR “Generative AI” OR “Gen* AI”

Media collections:

For media reports in Germany: Germany–National newspaper collection

For media reports in Hong Kong: Hong Kong–National newspaper collection

For media reports in India: The Times of India; The Indian Express

We selected news reports written in English and published at the national level. Therefore, the taxonomy of anxiety, which will be generated based on these texts, will retain some level of comparability with the other two domains.

The search results:

After collecting 7431 URLs within the aforementioned dimensions on the Media Cloud, we finally downloaded 6842 full–text news stories around Gen–AI, ready for further examination. The details of the results are presented in

image

Table 1. The results of the Medial Cloud Search

Although there is still a lot of work left pending in the first phase of mapping, such as checking the relevance of the text content and mitigating the discrepancy in size among these three countries/regions, an impressive amount of data has been obtained via Media Cloud Search. To some extent, digital tools empower us to grasp the fluid person–artifact relationships happening in different landscapes at the same time, and offer us a holistic view of these locations, not only spatially but temporally as well.

However, in order to experience the multi–space travel, it is inevitable that it comes at a price — the loss of context. What we mean by that is that when we extract texts from the websites, we also decontextualize them from the original time sequence and the surrounding cultural, social, and geographical relations, making them separate performances. Hence, the comparative case–studies are indispensable at this point for a “thick description” (Geertz, 1973, p.6) of the anxiety around the Gen–AI as well as for the consistency of the author function changes to speak out.

References

  • Ahmad, A., & Sahu, G. K. (2019). “Newspaper coverage on human rights issues: A comparative study of The Times of India and The Indian Express.” Indian Journal of Communication Review, 7 (1), 19–26.
  • Duan, R., & Miller, S. (2021). “Climate change in China: A study of news diversity in party-sponsored and market-oriented newspapers.” Journalism, 22 (10), 2493–2510. https://doi.org/10.1177/1464884919873173
  • Foucault, M. (1979). “Authorship: What is an author?” Screen, 20 (1), 13–34, https://doi.org/10.1093/screen/20.1.13
  • Geertz, C. (1973). The Interpretation of Culture. Basic Books.
  • Latour, B. (2007). Reassembling the social: An introduction to actor-network-theory. Oxford University Press.
  • Lazarus, R. S. (1991). Emotion and Adaptation. Oxford University Press.
  • Li, X., Wang, Z., Gao, C., & Shi, L. (2017). “Reasoning human emotional responses from large-scale social and public media.” Applied Mathematics and Computation, 310, 182–193. https://doi.org/10.1016/j.amc.2017.03.031
  • Philip, K. (2007). “What is a technological author? The pirate function and Intellectual Property.” Postcolonial Studies, 8 (2), https://doi.org/10.1080/13688790500153596
  • Schlueter, E., & Davidov, E. (2013). “Contextual sources of perceived group threat: Negative immigration-related news reports, immigrant group size and their interaction, Spain 1996–2007.” European Sociological Review, 29 (2), 179–191. https://doi.org/10.1093/esr/jcr054