User:Ruben/Graduation/Thesis Outline - 1 Feb

Choose how you feel; you have seven options

An inquiry into algorithmic classification of expressions of emotion.

Facial emotion analysis is booming business with multimillion dollar investments, and various implementations are about to be rolled out in public places. All current tools for emotion analysis are grounded in Paul Ekmans’ theory of the 7 basic, culturally universal, expressions of emotion, even though this theory is disputed. Besides, theory shows there can be a major gap between feeling and facial expression.

Already since Plato reason and emotion have been seen as antagonists. A dual system. Classification, a child of reason, is now applied to its counterpart: the emotions. Furthermore, in the digital process of ‘emotion analysis’, reason is omnipresent. It seems therefore a paradoxical union of the dichotomy of reason and emotion. Is it a happy marriage between the two, or is it the reason subduing the emotion?

My main question then is: which classification takes place in facial emotion analysis?

Goal: Scrutinise the algorithmic classification of facial expressions in emotion-categories. (And speculate on its impact on our subjectivity.)

Scope: In this research I will look at both the technical process of classification and at its psychological foundation. And speculate on its impact on our subjectivity – which is (for now) the focus of my practical work.

Chap. 1. What is algorithmic emotion analysis? How does it (technically) classify emotions? A survey.

Based on interviews with people in the industry and existing literature on the topic, I will highlight the technical working of emotion analysis. And more in particular, I will focus in on the classification procedure.

So I will not look that much at facial recognition algorithms; rather I will examine the translation of a facial expression into parameters along the emotional dimensions. Thereby showing the ambiguity of the numbers that the algorithms produce: they represent both a certainty and an intensity.

Chap. 2. Which psychological classification for emotions is employed?

The emotion algorithms provide a categorisation that is already based on existing classification. In this chapter I will investigate the choice for the used classification. And potential problems with this approach.

It shows how these tools (quoting Femke): “connect the desirable to the possible.” (“Het koppelen van het wenselijke aan het mogelijke”)

Chap. 3. Classifying the self. A selection of applications of emotion analysis and its problems.

What sort of applications for emotion analysis exist or are proposed. And what motives for using the technology can we distinguish in these applications?

Futhermore, what (unforseen) consequences might these tools have (answer: impact on subjectivity).

Chap 4.

….

In this possible chapter I might reflect on my practical work, based on previous theory. Although I am not sure whether this fits in this thesis, or should rather be a separate document, accompanying the work.

Notes.

Although this outline seems somewhat to match up with my previous paper, it is not deliberately written like that. Especially the first chapter has a different approach than what I have currently written. The first will be a more technical consideration of the technology. Futhermore, in the third chapter I will allow myself more room for speculation than in my paper.

Key texts

1. Massumi, Brian (1995). The autonomy of affect. Cultural Critique, 83-109.

Contrary to the APA, Brian Massumi draws a distinction between feeling, emotion and affect. Feeling being conscious. Emotion is the facial expression (which can be sincere or feigned). Affect is the bodily sensation.

I also refer indirectly to this work using the explanatory text Feeling, Emotion, Affect by Shoush (2005)

2. Foucault - Technologies of the Self, in Hutton, P. H., Gutman, H., & Martin, L. H. (Eds.). (1988). Technologies of the self: A seminar with Michel Foucault. University of Massachusetts Press.

Technologies of the self are those ‘inventions’ that help us to shape our subjectivity. This can be rules (like in religions) or tools (such as with the Quantified Self movement).

Foucault shows that the ‘care of the self’ is not an objective act.

3. Ruckenstein, M (2014). Visualized and interacted life: Personal analytics and engagements with data doubles.

Ruckenstein has done an experiment with heart rate monitors showing that personal data analysis does not give the users an objective insight into their own being. Rather, it gives rise to a ‘conversation’ with their data-double.

In my argument I use this text to show that the usage of emotion analysis in a Quantified Self-like system is not an objective act. But ‘shapes’ the user’s subjectivity (self-awareness/self-image).

4. Kendal Walton - Games of make believe

Emotions a ‘real’ within their own context of expression. Walton explains these emotions based on a game a child is playing with its father. The father, acting as a monster, chases the child. The child both laughs with joy and screams with fear. According to Walton both emotions the child experiences are real within the context of that emotion. Within the game the child is chased by a monster, its fear is real; within the game that he is playing with his father, his joy is real.

After reading this text I asked myself: are emotions not all context dependent? Are our interactions not always some sort of ‘games of make-believe’?

To be honest, this text had a clear goal in my proposal. Now that my topic is shifting somewhat, I wonder how this text still relates.

5. several texts I use in my draft paper.

In my current draft I construct my argument using various articles. From eg El Kaliouby, Ekman, Hoque and a work from Baron-Cohen. None of these texts are more ‘key’ in my argument than the other. They build on each other, so I would rather mention them all.

With these texts I examine the categories existing in emotions, and the erroneous assumption that facial expressions represent a feeling.