Emotional Intelligence in Deep Learning
In the 2016 film Arrival, based on Ted Chiang’s excellent Story of Your Life, Amy Adams’ character gives a summary definition of what’s known in linguistics as the Sapir-Worf hypothesis: “it’s the theory that the language you speak determines how you think.”
This concept appears elsewhere in fiction, such as Orwell’s 1984, in which the authoritarian state creates the language Newspeak to make it impossible for people to think critically about the government, or even to contemplate their own subjugation.
And, more disturbingly, it plays a role in non-fictional contexts as well. During the Holocaust, Nazis described Jews as Untermenschen — subhuman — and rats. Hutus involved in the Rwanda genocide called Tutsis cockroaches, and slaveowners throughout history referred to slaves as animals — the same word our sitting president has used to describe immigrants.
“Dehumanizing,” as professor Brené Brown put it, “always starts with language.”
The opposite holds true, too. If the language we use with other people conveys emotional awareness, empathy, authenticity, and self-regulation, it is humanizing and considered emotionally intelligent.
In the marketing world, however, where effective communication is everything, we often forget this.
Consider the negative influence that programmatization, automation, and data-worship have had on the very way marketers speak, from the jargon of AdTech to the basic labels we apply to those people to whom we want to sell our wares: “targets” and “consumers.” In particular, as my former colleague Jonah Bloom wrote, the pervasive use of the word “consumer” both depersonalizes and reduces the other humans with whom we share the planet to “bipedal purses whose only value is as buyers of stuff”.
It is from this lack of contextual awareness, of EQ, that so much failed marketing stems.
Given the expectation, particularly among millennials, of authenticity and personalization in their interactions with brands, as well as the growing use of machine and deep learning in all areas of digital marketing, from automated media buying (Albert) to computer vision for branded object recall (Clarifai) to recurrent neural networks for language modeling, now is the time for us to be thinking structurally about the other new things machine learning could enable, and to focus on building those applications.
To this point, one burgeoning area of innovation is in “affective computing.” In order for an AI to interact with a person in a way that feels truly authentic to the human, the AI must able to convey empathy and detect emotion. The ability of a computer to do so falls into the field of affective computing.
Detecting emotion in human faces in one version of this, and not a brand new one; in 2016 Apple acquired Emotient, a startup that uses machine learning to analyze facial expressions and detect emotions, as does Affectiva, the MIT Media Lab spin-off founded in 2009.
There are contexts in which real-time, personalized communication at scale cannot be achieved with humans alone. Think of 5,000 people trying to text simultaneously with Hertz or United or, more urgently, someone with undiagnosed depression trying to talk to a mental health specialist at 2 am. These are areas in which human contact, while important, does not scale well.
The application of deep learning to solve these challenges goes beyond the analysis of images and words into generative content, powering a new kind of machine-human communication imbued with both emotional intelligence and empathy.