Bridging Bytes and Feelings: Rana el Kaliouby's Emotional AI Crusade
Bridging Bytes and Feelings: Rana el Kaliouby's Emotional AI Crusade
This is the Three Data Point Thursday, making your business smarter with data & AI.
Want to share anything with me? Hit me up on Twitter @sbalnojan or Linkedin.
Let’s dive in!
“I am a brain, Watson. The rest of me is a mere appendix.” (Sherlock Holmes in The Adventure of Mazarin Stone)
“Even though we’re in the business of building AI, it always comes back to the humans.” (Dr. Rana el Kaliouby on the Lex Fridman Show)
Data is emotionless; algorithms are inhuman. But that’s kind of the point of it, right? Remove emotions and all the fuzzy stuff to get objective decisions and correct calculations.
So, what’s the natural reaction to data and algorithms by humans? We get scared; we have a natural aversion to both of them. In fact, the United States, home to the powerhouses of AI, is the country with the highest AI fear index following a recent survey.
"Traditional computers are emotion-blind. They have trained us to behave as if we lived in a world that was completely devoid of emotions." (Dr. Rana el Kaliouby on computers and emotions)
So, what do great data leaders have to do? They must build empathy into their business, products, and algorithms.
Easier said than done…
I remember the bestselling author Yuval Harari saying repeatedly that AI will be very good at making us feel comfortable and that it will have empathy. My question is just: when?
My answer is: Once people stop to ignore empathy as a crucial data leadership skill!
Recently, I stumbled over a book about just that. “Girl Decoded” by Dr. Rana el Kaliouby is a fascinating read and author.
So let me share the story of Dr. Rana el Kaliouby, the “Emotional AI Pioneer.” to understand how to build empathy for and into the soulless machinery of AI & data.
Why computers need to be emotional
But wait, why should we care at all about emotions? Aren’t computers just that, computing engines? Huge calculators? …
They were until the academic field of “Affective Computing” was founded in 1997 with one single most influential book.
Rana el Kaliouby grew up in Egypt and then went to Cambridge to earn her Ph.D. In that phase, she read a book that transformed how she viewed computers, machines, and their interaction with humans.
A book that changed her life and gave her a mission she’s been on for over a decade.
This book, Affective Computing by Rosalind Picard, makes a simple claim: When humans interact, we use the other side’s emotions to forecast their behavior & to make decisions. As such, emotions are critical to both our interactions as well as to our decision-making.
That means computers need two skills to be effective aids:
The skill to understand our emotions (what Yuval Harari likes to highlight)
The skill to demonstrate emotions themselves
This idea stuck with Rana, as she dedicated her life to emotional AI and founded a company with Picard. This company is called Affectiva.
To build empathy, you need to know yourself
It is striking that one of the most outspoken people on the dangers of AI is Elon Musk, a man who doesn’t have a natural feeling for empathy.
In a similar vein, Mark Zuckerberg, another tech giant not exactly famous for his empathy or his display of emotions, focuses most of his efforts on the transparency of algorithms rather than on emotions and empathy.
For the usually ultra-rational techies, both reactions are natural: fear and a tendency to overrationalize the unrationalizable.
And yet, Both Zuckerberg and Musk probably make worse data & AI leaders for their lack of empathy. So what does it take to be great at helping AI & data display and understand emotions?
It is to know yourself and have empathy for yourself.
It is to realize that it isn’t necessarily about transparency but trust.
It is about feeling safe.
It is about feeling connected.
It is not at all about engaging the frontal lobe to deduce you can trust a machine analytically, it is about feeling right, not knowing something is right.
To build empathy, you need to read faces
“I’m a very emotive person and the face to me is really key. You know, I had spent years just watching and observing people and I wanted to build an algorithm that can quantify that. So that was my PhD work.” (Rana el Kaliouby, From Cairo to MIT)
In a recent episode of the Lex Fridman show, Mark and Lex experience what makes real faces so special. For it, both of them did thorough 3d scans to perform the interview inside the metaverse.
And they notice quickly, even for people very low on the facial expression scale, the difference between zero and a few facial expressions is enormous. It immediately enables both of them to quickly read the other’s emotional state, to put words into context, and to read without listening to words.
Empathy is an exercise in diversity
“Holy shit, this thing is kind of racist.” (Lex Fridman)
“"A European car company sent us a training set of faces, but they were all blue-eyed, blond middle-aged guys," she recalls of the data set. "We told them this was not good enough. If our technology doesn't work with darker-skinned people like me, or with people who represent diversity in all of its forms, then that’s an epic fail from a business and ethical standpoint."” (Rana el Kaliouby’s Quest to make more humane technology)
Software people pride themselves on being bit-driven. That they prefer software and bits over atoms and physical goods - because, as the story goes, these scale way faster.
But humanity isn’t a homogenous blob of indifferent people - we’re a diverse set of millions of subgroups, cultures, religions, and ethnicities.
Turns out, even with bits, we’re quickly hitting boundaries, the boundaries of our own diverse culture space.
Diversity isn’t a problem; it’s a way to scale. To utilize it, you need to understand it and be able to collect diverse datasets and be able to unbias algorithms.
Rana learned this firsthand with one of her first products when her prized marketing algorithm suddenly stopped working. One of her customers had started to use the algorithm to judge facial expressions in the Chinese market, a cultural space with a very different way of using facial expressions to express emotions.
Data & Empathy & Text
Rana describes in her book how, in her childhood, her grandmother held these parties when the mangos were just picked from the trees in her backyard. She remembers every smell of the food, of the sweet fresh mangos.
Emotions are not just about faces; they are about all the senses we have.
Empathy is not just faces though! It is really about all our senses. It is sounds, smells, visuals, and touch. For computers, it can be about text, videos, and so much more.
Every form of input can help to build empathy. If your Apple watch measures your heart rate, it suddenly has a decent grasp on your emotional state.
We’re already making a lot of progress in all of these directions, albeit most research is in an early stage. A recent study, for instance showed that AI can help to enhance text-to-text conversations and make them more empathic.
So what?
Rana is an exceptional data leader - devoting her life to creating a new field of science - yet she and Rosalind Picard make a good point: We need to make data & algorithms emotionally able to make our businesses & products successful in the long run.
No one is going to sell a product or feature that scares people. To do so, we all need to enable our systems to do two things well: to read and display emotions.
Additionally, all data-heavy products are about helping people make better decisions; until now, we mostly tried to sway people into making better decisions by throwing facts and information at them. However, maybe it’s time also to help people get into the right emotional state to make decisions.
We’re going to do that by learning to become more emphatic ourselves, by helping our systems to get access to emotional signals, and we do this by teaching our systems diversity. There’s no good excuse for not doing emotional AI; we can already do it just based on text.
This is a hell of a task - especially difficult for all of us (me included) cold, technical, analytical data people - yet that’s what is needed to show great data leadership.
Here are some special goodies for my readers
👉 The Data-Heavy Product Idea Checklist - Got a new product idea for a data-heavy product? Then, use this 26-point checklist to see whether it’s good!
The Practical Alternative Data Brief - The exec. Summary on (our perspective on) alternative data - filled with exercises to play around with.
The Alt Data Inspiration List - Dozens of ideas on alternative data sources to get you inspired!