Fear + Anticipation = Anxiety: What are we anxious of in the face of Gen-AI

Mar 14, 2024 2:33 PM

Ruiwen Zhou

Publication year
March 14, 2024

Photo: YASNARADA from Canva

Fear + Anticipation = Anxiety

What are we anxious of in the face of Gen-AI

by Ruiwen Zhou

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

AI Project Series 2: Tracing the anxiety in the landscape of Generative AI’

In the previous blogs of this series, "In Search of the Digital Author: Author Function in the Age of Generative AI," we built a connection between the anxious feelings that occur with the change in the author function; next, we delved into the landscape of Gen-AI to discuss the theoretical framework of the project, and explored ideas about the discourses and conversations concerning the new technology. In this blog, we will continue our journey of hunting for the anxieties around Gen-AI as well as gather insights from this searching and mapping practice.

Disassembling anxiety

According to the psycho-evolutionary theory of emotion (PTE), which was introduced and elaborated by Robert Plutchik (1962), there is a small number of prototype emotions such as joy and trust, and all other emotions like love and disappointment formed as "combinations, mixtures, or compounds" (Imbir, 2017) of the primary ones. Describing the formation and combination of emotions, PTE also highlights the role of stimuli in the environment, saying that emotions are our reactions to stimulus events (Plutchik, 1990).

To extend the propositions, psychologist Rober Plutchik (1980) proposed a structural model, often referred to as the "Plutchik Wheel" (Figure 1.) to analyze the interactions among emotions (Imbir, 2017) as well as unpack the ambiguousness of the formation of the complex feelings.

The model has two features: first, it places eight core emotions on a circle based on their semantic similarity and polarity. Between two core emotions is an emotion that combines two adjoining emotions. Second, in the vertical dimensions, the model uses petals to illustrate the variations in the intensity of emotional states; that is, the degree to which these emotions can be felt. In this vein, this model addresses not only the classification but also the relationships of emotions by linking semantic proximity/opposition to spatial proximity/opposition in the wheel.


Figure 1. the Plutchik Wheel

Anxiety as a tertiary dyad


Figure 2. Plutchik's dyads (Semeraro et al., 2021)

Adopting the metaphor of color blending (i.e., fusing two colors together to create a new color), the model also points out that our emotions are interrelated: A complex emotion, namely a dyad, can be elicited if two emotions presented in the model are triggered simultaneously. Accordingly, it is considered a primary dyad when the complex emotion is elicited by emotions which are one petal away, and a secondary dyad is a combination of two two-petal-away emotions. In their article, Smeraro et al. (2021) have comprehensively illustrated the dyads in the Plutchik's wheel (Figure 2.) and we see the emotion of "anxiety" is a tertiary dyad, compounding anticipation and fear.

Admittedly, the Plutchik's model of emotion is not perfect because our ever-changing emotional feelings are never exhaustive. However, it can offer us an efficient guideline to understand how our feelings are placed in a complex chain of cognitions, feelings, actions, and stimulating events (Plutchik, 1982; Plutchik, 1990, as cited in Lmbir, 2017). By addressing both the complicity inside the emotions and the intricate relations surrounding them, the PTE along with its model of emotion may offer us possible ways to explore how and why people experience fear and anxiety when faced with the proliferation of new technologies, such as AI.

Where does Gen-AI anxiety come from?

With anxiety as a tertiary dyad in mind, we now introduce a computational method to contextualize the Plutchik model into the landscape of Gen-AI to search situations where fear and anticipation are intertwining, and where anxieties are engendered.

Specifically, we put the text scraped from media reports (see the 2nd blog of this series) directly into GPT-4 and let it label the data. Subsequently, we input the labeled data again into “unlabeled data” after checking the consistency and the accuracy of the classification of this small labeled data. By taking the labeled data as annotation, this active processing loop can classify and analyze the text from our media report database both efficiently and reliably (Zhang et al., 2023).

The reason to adopt this AI-mediated approach is twofold: on the one hand, it has practical advantages when it comes to processing a huge amount of data.

Even though the exact number of parameters in GPT-4 has not been officially confirmed by OpenAI, a variety of sources suggest that there might be more than 1 trillion parameters (e.g., Shevchuk, May 2023; Bastian, Mar 2023). With way more parameters compared with GPT-2 and GPT-3, the large language model of GPT-4 can gain a greater degree of reliability to meet our needs. Moreover, this approach enjoys high cost-effectiveness as it enables labeled data as annotations to function as supervision in this iterative process, thus reducing the cost of time and human labour. On the other hand, as a research project that is interested in how the underlying rules of our ways of perception have been mediated by artificial intelligence, it is crucial to embrace AI technologies at each step of our research journey. By doing so, we can personally perceive, witness, and experience the permeation of Gen-AI in every aspect of our society and our lives, including conducting research.

Towards a Gen–AI anxiety taxonomy


Figure 3. Anxiety objects in Hong Kong and Germany

Following this path, we conducted our preliminary research and obtained a list of anxiety objects brought up by news reports in Hong Kong and Germany (Figure 3.). According to the list, there are more than 800 possible objects that are mentioned in the media reports database, which has not even included the concerns in India as yet. Does that mean there are hundreds of distinct Gen-AI anxieties existing in our everyday lives? Could these diverse, disordered, and incessantly emerging anxieties be variations on a certain number of the fundamental anxieties and be labeled uniquely under various contexts? If so, what are the fundamental categories of these anxieties that make AI constantly new? Before ushering in the age of the newness, it is time for us to put our emotions in check and re-examine "the breathless rhetoric of newness and unexpectedness" (Shah, 2024, p.7) that often comes along with the emergence of new technologies.

As the anxieties are constantly shaping the narratives and people's response to the technology, there is an urgent need for us to dig deeper and hit the bedrock of Gen-AI anxieties. Hence, we will be able to find out what people are concerned about, and moreover, be able to investigate if there are possible approaches through which we can classify the various anxiety objects into a limited number of fundamental categories. By building a toolkit for organizing, classifying, and analyzing Gen-AI anxieties in the domain of media reports, we aim to change the way of the narratives around these anxieties to probe into how technologies affect society, with the intervention of digital author, from media perception.


  • Bastian, M. (Mar 2023). GPT-4 has more than a trillion parameters-Report. https://the-decoder.com/gpt-4-has-a-trillion-parameters/#google_vignette
  • Imbir, K. K. (2020). Psychoevolutionary theory of emotion (Plutchik). In Encyclopedia of Personality and Individual Differences (pp. 4137–4144). Springer International Publishing.
  • Plutchik, R. (1962). The emotions: Facts, theory and a new model. Random House.
  • Plutchik, R. (1990). Emotions and psychoterapy: A psychoevolutionary perspective. In R. Plutchik & H. Kellerman (Eds.), Emotion: Theory, research and experience, Emotion, psychopathology and psychotheraphy (Vol. 5, pp. 3–42). Academic Press.
  • Semeraro, A., Vilella, S., & Ruffo, G. (2021). PyPlutchik: Visualising and comparing emotion-annotated corpora. PLOS ONE, 16(9). https://doi.org/10.1371/journal.pone.0256503