The emotions of Lil Miquela
How AI-generated influencers can actually influence our interactions on social media.
The Intelligent Friend - The newsletter about the AI-humans relationships, based only on scientific papers.
Hello, dear readers! This post is exceptionally on Monday because I didn't have my computer with me for a few days. The topic is very interesting and I really think we'll talk about it again. As always, have an enjoyable read and comment with your ideas and opinions on the topic!
Intro
Virtual influencers are not 'real' for many consumers. Opinions or perceptions aside, however, it must be recognized that their existence has and will have a strong impact on consumption. Their photos of everyday life, albeit created from scratch, have even led brands such as BMW or Prada to choose them for their campaigns. But do they actually influence our behavior on social media?
The paper in a nutshell
Title: Artificial intelligence-generated virtual influencer: Examining the effects of emotional display on user engagement. Authors: Yu et al. Year: 2024. Journal: Journal of Retailing and Consumer Services.
Main result: emotions expressed by computer-generated imagery (CGI) influencers can actually influence user engagement on social media.
Using the computer are social actors (CASA) framework and using facial recognition, the authors tried to understand the real effects of images of virtual influencers on user behavior.
The computer-generated imagery (CGI) influencers
I was eating a plate of pasta for lunch while watching the news. The commercial interrupted the broadcast for five minutes. One of the adverts that caught my attention was one from BMW in which a young girl was driving a new electric model, inspiring modernity and a glimpse of the future. There was something about that girl, I thought I had seen her before, but I didn't understand. After a while, I connected: it was not a 'real' actress in a commercial, but Lil Miquela, one of the most successful virtual influencers.
She is just the most famous probably of a particular type of influencer: the computer-generated imagery (CGI) influencer. They are the result of increasingly elaborate and complex artificial intelligence image generation techniques that are sure to bring about significant changes in the world of virtual media1.
Unlike avatars, in fact, CGI influencers can actually express emotions in a more human way than ever before: animation and rendering techniques enable CGI influencers to replicate the subtle nuances of human expressions2.
Now, this difference, for the purposes of today's study, is crucial. When we interact with AI, although it may be the case of a very anthropomorphized AI (i.e. very similar to a human being), we somehow perceive that it is a robot, a technology not so similar to us. This feeling makes us feel uncomfortable, discouraged. This is the core (reduced to the essentials) of the so-called Uncanney Valley theory3 4.
A significant criticism of chatbots or avatars and AI concerns indeed their degree of resemblance to humans: many find these technologies disturbingly unreal, which can cause discomfort and fear in users5. According to the authors, the advancement of AI techniques that lead to the maintenance of "virtual" influencers or, as we call them today, CGI influencers, greatly reduces the discomfort we feel.
Virtual influencers like Lil Miquela, as reported by the authors, in fact:
have digital personalities;
have perceived social and physical attractiveness;
are built on the basis of marketers' intentions.
Now, if all this is true, what we have understood is that: consumers, in interacting with a more or less anthropomorphized technology, have a degree of satisfaction until they realize that, although "real-like", technology is not human (uncanny valley). Furthermore, CGI influencers, who are influencers created and managed thanks to the latest AI-driven techniques, are a particular form of influencers that can reduce these possibly negative feelings; finally, they have characteristics that make them increasingly capable of influencing user behavior on social media.
However, if you pay attention, a piece is missing: if all this is true, what drives us to create relationships with technologies? That is, what happens in us when we try to interact with a technology and end up building a relationship?
User engagement and CASA framework
In the study of human-computer interactions one of the most successful theories is the so-called CASA framework6. The principle is very simple: when we interact with a technology, we humans interact by applying the same rules of common social interactions.
Furthermore, when computers exhibit human-like attributes, consumers often expect them to adhere to various social norms7. This principle makes computers so-called "social actors". And that's why the anagram of the name of this theory is none other than Computer As Social Actors (CASA).
Now, as the researchers rightly specified, it would be almost "crazy" to see humans interacting with any type of technology in this way. What would you think of a person talking to his calculator? Here, exactly. Gambino et al. (2020) have precisely demonstrated that (reporting from today's paper): "this paradigm only applies when technological artifacts exhibit adequate social cues that imply their ability to serve as a point of reference for social interaction"8.
However, despite the importance of this specification, you can understand the power of this paradigm today in relation to AI and today's protagonist influencers. Further research. for example, has demonstrated that service robots with gesture-based interfaces can enhance communication9 and boost user engagement10.
But if we talk about influencers, we focus above all on the behaviors that occur on social media. This is why the authors focus exactly on social media user engagement, defined as the extent of interactions, such as likes and comments, between users and social media content11.
But what generates user engagement? Anyone who has been reading this newsletter for a while will have already read several issues on the importance of emotional aspects in the use of AI. Here, even today's authors, starting from the basis of the CASA framework, have looked for evidence of a significant effect of the emotions shown by CGI influencers on user engagement.
However, there are two fundamental questions:
which emotions to consider?
how do we measure emotions?
To answer the first question, the authors focused on the six main emotions identified by Ekman (1992)12, "powerful in influencing social interactions"13. However, if this is easy to understand, it is certainly not the measurement of an emotion by a person, like Miquela who, in reality, is not "real".
Then, the authors involved the so-called use of Facial Action Coding System (FACS)14, which pinpoints facial movements that are visually observable and correspond to specific emotional expressions.
These movements linked to certain emotions, which naturally provide a standardized method for these not simple first impact analyses, are also called action units (AUs), and represent distinct actions or movements of specific facial muscles15.
In summary, the authors detected the resulting expressions from Lil Miquela's face in different posts, to verify an effective impact on user engagement. Not only did this research use such a specific methodology on virtual influencers, but it provides further evidence on the impact of the emotions of CGI influencers on users on social media (of great value for companies).
An impact to be explored
The methodology of this paper involves a sophisticated blend of data mining, image clustering, and emotion analysis. Researchers utilized facial recognition technology to decode emotional expressions from images of Lil Miquela posted on Instagram, correlating these with user engagement metrics such as likes and comments. The results are really interesting.
First of all, the authors identified 10 clusters of images, which correspond to different products in different sectors or situations: 1) Self-portrait; 2) Fashion and branded content; 3) Gastronomy and dining experience; 4) Dynamic posing and motion shot; 5) Sightseeing and entertainment; 6) Social events and gatherings; 7) Glamor photography; 8) Car model shot; 9) Urban fashion; 10) Unidentified images.
Subsequently, it should be noted that there is actually a high degree of correspondence of Lil Miquela's expressions and emotions compared to the relationship between human ones, thus denoting the great reproduction capacity of these influencers.
Focusing more specifically on the link between displayed emotions and engagement, certain emotions like happiness and surprise generally promote higher engagement, suggesting that users are more likely to interact with content that conveys these positive emotions.
Conversely, expressions of disgust and sadness, while effective in certain contexts, can sometimes lead to lower engagement. However, it should be noted that happiness does not always lead to positive results. Surprise played also a significant role, especially in its ability to generate a sense of excitement and interest.
Finally, in relation to the various clusters, the scholars also show how some specific image clusters, such as fashion and branded content and glamour photography, could not lead to always desirable signals. In fact, facial expressions in them “can create a sense of seriousness, concern or tension, which may be incongruent with the desired mood (e.g., confidence, relaxation) or image of the content (e.g., aspirational lifestyle)".
Before moving on to the research questions, I remind you that you can subscribe to Nucleus, the exclusive weekly section in which I send 4 paper summaries, links to resources and interesting readings, and interview the authors. It comes out every Wednesday.
In this (open to everyone), for example, we talked about the political opinions of ChatGPT, a new environment for human-robot interactions and interesting links such as the similarity between “the child you” and the “current you”.
Just this? No: you can also access the Society, the exclusive chat where you can discuss AI and talk about your projects, as well as find your next co-founder or collaborator. Subscribe now.
Take-aways
Lil Miquela goes further. Thanks to new AI-driven technologies, CGI influencers have great potential to go beyond the feelings of negativity felt by consumers towards anthropomorphized technologies (uncanny valley).
The impact of emotions. The study provides detailed evidence that emotional expressions significantly impact user engagement on social media. It finds that specific emotions such as happiness, sadness, disgust, and surprise, when expressed by CGI influencers, play a crucial role in influencing user interactions like likes and comments.
Expressions and emotions. The research utilizes advanced facial recognition technology to analyze the facial expressions of CGI influencers, specifically through the Facial Action Coding System (FACS). This system helps break down emotions into individual action units (AUs), providing a granular understanding of how each facial muscle movement correlates with user engagement.
Further research directions
Explore the development of CGI influencers to enhance the generalizability of findings.
What is the synergy between post captions and images in affecting engagement rates, and how can scholars effectively uncover it?
Given that likes and comments were used as proxies for user engagement in this study, do they accurately represent the true level of engagement? How can experimental methods or biometric tools, such as eye tracking, be utilized to assess the internal states of viewers or the amount of time spent engaging with content?
Thank you for reading this issue of The Intelligent Friend and/or for subscribing. The relationships between humans and AI are a crucial topic and I am glad to be able to talk about it having you as a reader.
Has a friend of yours sent you this newsletter or are you not subscribed yet? You can subscribe here.
Surprise someone who deserves a gift or who you think would be interested in this newsletter. Share this post to your friend or colleague.
P.S. If you haven't already done so, in this questionnaire you can tell me a little about yourself and the wonderful things you do!
Drenten, J., & Brooks, G. (2020). Celebrity 2.0: Lil Miquela and the rise of a virtual star system. Feminist Media Studies, 20(8), 1319-1323.
Ahn, R. J., Cho, S. Y., & Sunny Tsai, W. (2022). Demystifying computer-generated imagery (CGI) influencers: the effect of perceived anthropomorphism and social presence on brand outcomes. Journal of interactive advertising, 22(3), 327-335.
Mori, M. (1970). Bukimi no tani [The uncanny valley]. Energy, 7, 33.
Mori, M., MacDorman, K. F., & Kageki, N. (2012). The uncanny valley [from the field]. IEEE Robotics & automation magazine, 19(2), 98-100.
Söderlund, M. (2022). Service robots with (perceived) theory of mind: an examination of humans’ reactions. Journal of Retailing and Consumer Services, 67, 102999.
Nass, C., Steuer, J., & Tauber, E. R. (1994, April). Computers are social actors. In Proceedings of the SIGCHI conference on Human factors in computing systems (pp. 72-78).
Wang, W. (2017). Smartphones as social actors? Social dispositional factors in assessing anthropomorphism. Computers in Human Behavior, 68, 334-344.
Gambino, A., Fox, J., & Ratan, R. A. (2020). Building a stronger CASA: Extending the computers are social actors paradigm. Human-Machine Communication, 1, 71-85.
Mara, M., & Appel, M. (2015). Effects of lateral head tilt on user perceptions of humanoid and android robots. Computers in Human Behavior, 44, 326-334.
Chuah, S. H. W., & Yu, J. (2021). The future of service: The power of emotion in human-robot interaction. Journal of Retailing and Consumer Services, 61, 102551.
Yu, J., & Egger, R. (2021). Color and engagement in touristic Instagram pictures: A machine learning approach. Annals of Tourism Research, 89, 103204.
Ekman, P. (1992). An argument for basic emotions. Cognition & emotion, 6(3-4), 169-200.
Landwehr, J. R., McGill, A. L., & Herrmann, A. (2011). It's got the look: The effect of friendly and aggressive “facial” expressions on product liking and sales. Journal of marketing, 75(3), 132-146.
Ekman, P., & Friesen, W. V. (1978). Facial action coding system (FACS)[Database record]. APA PsycTests.
Campos, B., Shiota, M. N., Keltner, D., Gonzaga, G. C., & Goetz, J. L. (2013). What is shared, what is different? Core relational themes and expressive displays of eight positive emotions. Cognition & emotion, 27(1), 37-52.
I have to say that the trend of AI influencers is going unnoticed by me and I don't regret it at atll.