View profile

One day, I'll sit on a couch and tell an AI my feelings...


Olivia Jeffers

February 12 · Issue #14 · View online

Welcome to Compassionate Technologies. Here you'll get a dose of real science and business in your inbox every Sunday morning. Why? Because cutting-edge research shouldn't be locked in an ivory tower. This newsletter covers the relationships between machine learning, robotics, genetic engineering, and climate science. It's all connected, and it's my passion to simplify and make clear those connections for all of you. Love, Olivia.

“What will happen when AI understands emotions better than we do?” students asked several times at the OpenMind::OpenArt Gallery at MIT. It’s a question that digs into the core of what it means to be human and machine.

#OpenMindOpenArt Gallery Mural at MIT (Feb 16 - Mar 2, 2017)
#OpenMindOpenArt Gallery Mural at MIT (Feb 16 - Mar 2, 2017)
Rosalind Picard, founder of Affectiva, an MIT Media Lab spinout and rising star in the relatively new field of affective computing. Well, new since Picard’s 1995 paper on the topic.
Engineers at Affectiva train computers to detect ‘affect’ - the precursor to emotions - from human faces. It’s reading the human mind, by reading the face.
"No he can't read my poker face" - Lady Gaga
For many of us, no matter how hard we try to hide our emotions, some emotional content will leak through to our face in the form of “microexpressions”, made famous by Paul Ekman and the popular TV show Lie to Me based off of his research.
Ekman codified the Facial Action Coding System (FACS, see below) which much of Affectiva’s work is based off of. Since the FACS system was derived from common denominators across culture and gender lines, it is believed to represent emotional expressions that are fundamental to all humans. For this reason, it is famously studied in law enforcement training for lie detection.
Ekman's Facial Action Coding System (FACS)
Ekman's Facial Action Coding System (FACS)
In general, emotions have been regarded as harmful in the western world, something that is better repressed and kept to yourself with a “stiff upper lip”. Infamously, executives with psychopathic traits (lack of empathy, egocentric, grandiose and charismatic behavior) tend to do better in business. But are emotions are bad?
How Much Emotion is Best for Decision Making?
In the early 1990s, Antonio Damasio studied patients whose emotional brain centers had been destroyed by cancer, but whose reasoning centers were otherwise intact. He discovered that too few emotions leads to ineffective and slow decision making as the thinker processes every possibility with little motivation to pick one option over the other - “paralysis by analysis”.
It turns out that emotions are necessary for decision making. Our body dictates values, such as “dopamine is good” and “pain is bad”. Our mind can only hold 5-9 items at once, and is not able to process all the activities going on ranging from the cellular level all the way up to seeking an attractive mate.
That’s where emotions come in. They are our body’s computational system, taking in all the external cues that our brains can’t process and delivering “simplified emotional feedback” such as tightness in your chest for anxiety, heat in your hands for anger, or warmth all over for love and happiness.
Love, it makes us warm all over <3
Love, it makes us warm all over <3
Essentially, our emotions give us a result without telling us why. It’s our mind’s job to either reverse engineer a complicated decision - or just go with the flow of preexisting patterns of reaction, emotion, and thought (aka “habits”).
The Mystery of Machine Intuition & Aggression
So, that sounds awfully familiar… “a result without a why.”
In September 2016, Google’s DeepMind (deep neural networks resembling brain neurons) “watched” millions of Go games, an ancient game with more potential moves than there are atoms in the universe. The game is famously thought to rely on human intuition, an area where human intelligence would always trump computer intelligence.
However, DeepMind won against the human Go world champion - and it was regarded a mystery as to how it was able to learn and “out intuition” a human!
According to scientists Henry Lin at Harvard and Max Tegmark at MIT, the answer to the mystery lies in physics. While the total number of possible algorithms that could yield solutions are mind-bogglingly large - the number of algorithms that actually work in real life (the life that games like Go exist in) are relatively few.
In fact, complex human behaviors such as aggression in the face of scarcity and the evolution of cooperative behaviors - were also replicated through a small set of algorithms in a virtual game played over many virtual-generations by Google’s DeepMind (see the paper here).
The Big Questions for Technologists
Just how human are emotions? If neural networks can evolve emotional behaviors, then maybe we should take a more computational look at ourselves.
Can we work together as biological intelligences standing side-by-side with artificial ones? Can we stand together as parents to the machines intelligences who will one day be our children?
Useful Links...
Thanks from Olivia :)
Did you enjoy this issue?
In order to unsubscribe, click here.
If you were forwarded this newsletter and you like it, you can subscribe here.
Powered by Revue