introduction _

This year marks the 100th anniversary of the birth of Herbert A. Simon, a Nobel laureate in economics “for his groundbreaking research into the decision-making process within economic organizations.” Simon was also a visionary in the area of artificial intelligence, and his first notable work in the field, “The Logic Theory Machine,” from 1956, is celebrating its 60th anniversary in 2016. Co-created with Allen Newell, it described the first computer program designed to simulate the problem-solving skills of humans.

Building on Simon’s achievements in the field of artificial intelligence, we take a journey to explore the latest innovations in AI and, most importantly, its human element, to ultimately answer the controversial questions: What physical human characteristics and emotions must a robot have to make people react to it? And, obversely, Can AI recognize human emotions?

human emotions explained _

Decades have passed since Simon first explored the psychology of human cognition; today AI is more and more present in our lives, be it via customer service or pure entertainment. No matter what its application, the Holy Grail of any successful AI project is its ability to achieve seamless interaction with humans. And at the core is AI’s capability to recognize and react to emotions.

But first, what are the basic human emotions, and why are they so important?

Sadness? Anger? Excitement? Curiosity? Fear? Identifying the key types – and number – of human emotions was tough even for Aristotle who, in the 4th century B.C., identified the following 14: confidence, anger, friendship, fear, calm, unkindness, shame, shamelessness, pity, kindness, indignation, emulation, enmity and envy. Later, Charles Darwin’s book “The Expression of the Emotions in Man and Animals” suggested that many emotional facial expressions are universal. This subsequently led to the conventional scientific understanding that there are six key emotions: happy, sad, afraid, surprised, angry and disgusted.

More recently, however, psychologists have simplified things even further. According to new research from the Institute of Neuroscience and Psychology at the University of Glasgow, people’s facial expressions, and the emotions they signal, can be reduced to four “basics”: happy, sad, afraid/surprised and angry/disgusted. These, according to the study, are the main biologically rooted facial signals, while the distinctions between surprise and fear, and between anger and disgust, appear later as more complex socially developed expressions.

Add to this the fact that we as humans are capable of experiencing more than one emotion at a time, and the task of identifying the exact nature of emotions gets even more challenging when it comes to AI and its computational models.

humanizing artificial intelligence _

The earliest emotion-relevant work in AI dates back to the 1970s, when cognitive science came of age as a discipline, inspired in part by Allen Newell and Herbert A. Simon’s 1972 book “Human Problem Solving.” In the same year, Kenneth Colby invented one of the first devices of emotional AI: a computer system called PARRY that simulated a conversation with a human paranoiac.

In 1988, Andrew Ortony co-authored the book “The Cognitive Structure of Emotions,” introducing the OCC model, which explained various emotions and their intensities, and is still used in AI. What makes the OCC model so powerful is that it’s hinged on the basic idea of characterizing emotions in terms of different ways of feeling good or bad about things.

Today, from an emotion-theory perspective, AI developers have started to incorporate these cognitive building blocks for emotion generation and recognition. “They have also developed ways of modulating emotional intensity,” Ortony adds, “and they have started to endow their artificial agents with personality profiles that interact with emotion.”

Among them are the American David Hanson and Hiroshi Ishiguro of Japan, who have both faced the challenges of creating robots capable of recognizing the subtleties in human expressions and emotions.

Ishiguro, a professor at Osaka University’s Intelligent Robotics Laboratory, who made headlines with an android he built in his own likeness, considers that desire and intention are key prerequisites to achieving human AI. By 2020, he and his team hope to apply these emotions successfully in Erica – a collaboration between Osaka and Kyoto Universities and the Advanced Telecommunications Research Institute, she is considered the most advanced humanoid yet.

Dubbed by her creator as the “most beautiful and intelligent” android ever, Erica is 23 and is able to engage in 10-minute conversations with people. “For humans, emotions and emotional behavior are very context-dependent,” explains Ishiguro, who is trying to develop the same in Erica. To prove the importance of self-awareness (or the closest AI can get to it), Ishiguro relies on psychologists and philosophers on his team, who all work tirelessly to answer the perpetual question: What is the minimum requirement for being human?

David Hanson, on the other hand, whose humanoid Sophia was a sensation at the latest edition of SXSW, thinks that robots need to have a highly expressive face so that people can communicate with them more intuitively.

At present, Sophia, who has a skinlike rubber face and about 60 expressions, can interact and lead simple conversations. Dr. Hanson, who previously worked with Disney, believes that the ability of AI to relate convincingly to humans is something that would eventually enhance the application of artificial intelligence in our future daily lives.

1957

Herbert A. Simon, Allen Newell and Joseph Carl Shaw develop the General Problem Solver (GPS), a program that uses cognitive psychology to simulate human behavior. It is originally intended to solve any general problem; however, GPS can only be used for well-defined problems such as chess, puzzles or logical theorems.

q&a: today’s movers and shakers _

Nadia Magnenat Thalmann,

expert in virtual humans and customer service

After studying psychology under the Swiss psychologist Jean Piaget and receiving a Ph.D. in quantum physics from the University of Geneva, Nadia Magnenat Thalmann, led by the work of Herbert A. Simon, founded the artificial intelligence research group MIRALab in 1989. This March, Thalmann introduced the world to Nadine, a lifelike humanoid robot that works as a receptionist and has the ability to recognize people and relate to them on a social and emotional level.

Please click the question to reveal the answer.

What were your priorities when designing Nadine?
The ability to recognize and acknowledge facial expressions aside, can AI systems possess the capacity for actual emotions?
Does this mean that you can program robots with certain personality traits and, as they have different interactions with different people, they can develop unique identities over time?
Any other useful applications?

David Hanson,

groundbreaking humanoid maker

Formerly a sculptor and a technical consultant at Walt Disney Imagineering, David Hanson has created hyper-realistic humanoid robots that have received international acclaim for their ability to replicate nuances in facial expression. Together with his team of scientists and developers at his eponymous company Hanson Robotics, Hanson is working on his latest project, Sophia. With an expected fall rollout, Hanson has big ambitions to change the field of artificial intelligence by creating an open-source platform for researchers everywhere.

Please click the question to reveal the answer.

Tell us about Sophia and how you achieved such a lifelike quality.
What about the AI infrastructure?
How accurate are the current emotional models?
Must robots have humanistic features for us to relate to them?

Lola Cañamero,

specialist in AI and embodied emotions

As the head of the Embodied Emotion, Cognition and (Inter-)Action Lab at the University of Hertfordshire, Lola Cañamero works on the multifaceted aspects of emotions in autonomous robots. She is currently researching how the use of emotions can help robots make better decisions and adapt to different environments and tasks, and how to build robots that socialize with the world around them and develop the way babies do. She’s also engineering a robot companion for diabetic children to help them manage their illness.

Please click the question to reveal the answer.

Why is it important that robots “feel”?
What’s the strategy and approach to endow robots with emotions? Is it about mimicry, or can you actually make robots “feel”?
Why is it so hard to create complex feeling in AIs?

what lies ahead: superhuman customer service? _

Today, as much time and effort is being put into making AI “human” as is into making it useful. In the next decade, research in AI will be largely devoted to developing social robots that can help, assist and look after people in our daily environments – homes, banks, hotels, shops – in a way that is natural for humans as well as efficient.

Breakthroughs in data processing and conversation systems are helping more and more companies to implement AI in their operations. According to some experts, well-advanced artificial intelligence could someday not only assist businesses in doing their jobs more efficiently, but also bring a more human touch back to customer service, leading consumers to prefer sophisticated and professional AI service to today's human variety.

Some human-lookalike robots have made an impact in health care. Hanson explains that, according to some theories, human emotions are so chaotic and faces display so much information that it’s almost painful to watch. “So a humanoid robot can be a safe entry point: some of my robots have been used in autism-therapy trials, and the results have been extremely positive,” he says.

His opinion is shared by Lola Cañamero, head of the Embodied Emotion, Cognition and (Inter-)Action Lab at the University of Hertfordshire, who is engineering a robot companion to help diabetic children cope with their illness. She is certain, however, that robots and AI systems will never completely replace the bulk of service providers.

“Robots can become better and better at doing specific things, learning to do things in line with our preferences and being very helpful in our homes,” she says. “The purpose of AI, however, is not to replicate human intelligence or human behavior, but to develop robots that help us do things better in a way that we can relate to and accept, while at the same time helping us understand our own intelligence and emotions better. In other words, a very efficient service provider that we really like, but is still a robot, not the replica of a human.”

Elinor C. Ostrom
Nobel Laureate, 2009

Learn more

Herbert A. Simon
Nobel Laureate, 1978

Learn more

Alvin E. Roth
Nobel Laureate, 2012

Learn more

Exploring the questions that shape our
world through Nobel Perspectives