Watching the recent Supreme Court television coverage and the wide range of emotions shown by all parties involved gave me pause to wonder whether artificial intelligence could be effectively deployed to analyze emotional response in humans.
While studying emotions is important for a number of reasons, artificial intelligence may not be able to cope:
- Emotions are the most unpredictable aspect of a person, but also a common aspect of the human condition;
- An emotion shows the way that a person perceives the world, and an accurate categorization of emotion could be a better indicator of truthfulness or deception;
- Emotions are an important aspect of human intelligence and play a significant role in human decision-making processes.
However, there is little information about what emotions really are, and the boundaries to the domain of what experts have called emotion are so blurry that it sometimes appears that everything is an emotion.
While emotions are a common aspect of the human condition, it should be evident that waiting for a consensus on how experts actually categorize emotions is unrealistic because the study of emotions is still incomplete. There is some acceptance of classifying emotions according to their physical manifestation, i.e. humans show emotion on their faces and in their body movement, and those cues could help classify and categorize emotional response. But each of us is different, and it is perhaps our pre-conditioning that dictates not only the emotional response but also the level of that response. Someone who was taught to foster joy may react differently than someone who is serially depressed when exposed to the same stimulus – hence the notion that emotions are unpredictable.
A main problem for categorizing emotions stems from language, because there are some emotion words that have different meanings in different countries, kind of like how Eskimos have hundreds of words for the condition of snow, but most folks just call it “snow”. Nevertheless, structuring emotions to be interpreted by AI means finding a common ground for the emotion words, even if the categories and classifications are different in different cultures. This may be impossible to codify, and certainly would be reductive as emotional responses and triggers evolve over time and with human experience. At one time, cars frightened people, and while today some people’s ability to not use a turn signal may be frightening, the aspect of driving a car or seeing a car does not strike the kind of fear in humans it once did.
Motoring is one of the most contemptible soul-destroying and devitalizing pursuits that the ill-fortune of misguided humanity has ever imposed upon its credulity…[they are] a pack of fiends released from the nethermost pit. – C. E. M. Joad, circa 1900
In addition, the notion of “mixed emotions” about a topic are inherent in humans as well, which may defy categorization – like the old joke about someone seeing their Mother In Law drive off a cliff in their new sport scar. One of the theories put forth is commonly known as the Strongest Emotion Model, which seeks to address this mixed emotions dilemma by assigning values for each emotion felt, and tagging the one that has the highest value as the predominant, and ultimately motivating emotional response.
While facial expressions can also be tied into mixed emotional response, they may not agree with the verbal cues offered as support of underlying emotion. If you watch the SCOTUS court proceedings, there are often facial expressions that seem to run counter to what is being said, and those facial expressions can be viewed as a more accurate representation of underlying emotion – like telling someone their disgusting dinner dish is tasty while wrinkling the nose, or when during a poker game a great hand and the exhilaration of the win is masked by a poker face. It gets deeper, as some emotions have no expression whatsoever, and that there are some others that have the same nuances across emotion, and thus, it is impossible to differentiate between them.
The goal of using artificial intelligence to analyze emotions is a noble direction in which to take machine learning. However, as humans offer such a wide variable to study, I would argue that a concrete basis on which to formulate artificial intelligence for the accurate assessment of the dynamics of human emotion is too lofty a goal for current technology.
This article was composed using the Zaphne content inspiration tool for original research. Sign up for a free demo of the Zaphne Fire platform at https://calendly.com/zaphne