Emotional Artificial Intelligence

Warning: This content has not yet been fully revised for this year.

Background

What is it?

Emotions have a massive influence on everything that we do. They can impair or aid decision making, change the way we perceive the world, impact our learning and memory, and much more. If we want artificial intelligence to be truly intelligent, it must learn how to interpret human emotions.

Emotional artificial intelligence, or affective computing, detects and interprets emotional signals. For example, it can learn the difference between a sad or a happy smile. Emotional AI is also capable of learning the optimal response to human emotions. For instance, it learns if performing an action makes the current situation worse or better than before.

How does it work?

AI machine learning models are build on:

Text sentiment analysis: detect and quantify the emotional tenor of text samples
Image and video analysis: facial expression analysis where certain physiological signals are gleaned through videos and images
Speech recognition systems: interpret a change in tone or voice inflections
Physical parameters: temperature and heart rate

Uses

Everyday emotional support

Emotech created Olly, a constantly evolving voice-activated assistant like Amazon Alexa or Google Home. Its machine learning algorithms make it learn to be more like its owner. The features that make it emotionally intelligent are its understanding of a user’s facial expressions, voice inflections, and phrases. From those inputs, it makes helpful suggestions to the owner and learns from the owner’s reactions.

Therapy

Microsoft’s Human and Understanding Empathy group are investigating the idea of emotionally intelligent therapy. Their arguments are that an AI therapist would be less costly, more personalized, and more frequent than a human therapist. The main advantage is that it is always available; it could support in a moment of crisis or take advantage of choice times when a person needs therapy the most. Microsoft’s team may still be in the research phase of this, but there are others that have already been created. For example, Woebot is an AI-powered therapy tool that offers conversations based on cognitive behavioral therapy. It is grounded in clinical research and powered by understanding and reacting to its patient’s emotions.

Rosalind Picard’s research has also contributed greatly to affective computing being used in the context of therapy. Much of her work has been on autism and emotion recognition. Her research has pioneered the idea of developing devices that could help humans recognize nuances in human emotion. In her book “Affective Computing,” she outlines framework for emotionally intelligent models, signal-based representations of emotions, moral implications of effective computing, building models of emotion for synthesizing emotions in computers, and applying the idea of affective wearable computers.

Potential Problems

What happens if an AI therapist responds poorly to a person in a highly emotional state?
Who all has access to the data the emotional AI gathered on you? What could happen if that data got leaked?
Can affective computing be used to emotionally manipulate patients?

Provocations

How can affective computing be used to improve the customer service industry?
Do you think an artificially intelligent therapist has more ability to benefit or harm its patients?
What privacy should a patient meeting with an AI therapist have? Is it ethical for the programmers monitoring and maintaining the AI therapist to have access to patient data?

Brea Koenes