麻豆传媒

WHEN Rosalind Picard announced to the world that computers needed to understand emotion, many scoffed. But her book, Affective Computing, published 20 years ago, seeded a new field. It鈥檚 now clear that computers will serve us better if we can help them understand what matters to us 鈥� using changes in our physiology, movement, facial expression and tone of voice to discern our emotions.

Affective computing is enjoying quite a few real-world applications. What kinds of ideas are you working on at the moment?

Something I鈥檓 excited about are wearable systems that help people see their stress levels and mood changing. In the early days of affective computing, a 鈥渨earable鈥� computer might have weighed 50 pounds 鈥� totally impractical. Today, I鈥檝e got this lightweight device on my wrist, the Empatica Embrace. Its sensors record my skin conductance, motion data and skin-surface temperature 24/7 鈥� information linked to my mood, stress and activity levels. It鈥檚 connected to a smartphone app. Using our latest tech, this set-up is around 85 per cent accurate at predicting a person鈥檚 mood, stress and health tomorrow night based on their data from today.

What鈥檚 the appeal of this system?

People have huge differences in their ability to gauge their own emotions and how those feelings change over time. People with autism in particular have difficulty with this but it is by no means limited to them. I鈥檝e had computer scientists pull me aside, usually privately, and say 鈥淚 don鈥檛 understand what you mean by 鈥榝eelings鈥�.鈥� It can be a terrible handicap that affects women as well as men.

So we have an opportunity to help people learn more about their feelings. Turning on cameras to see how much someone smiles and frowns, and listening for the emotion in their speech could also be beneficial. Personally, I鈥檝e learned a lot about my stress using the wrist sensor 鈥� about what activates it, what calms me down.

How might this sort of technology help us deal with stress?

In the heat of the moment, you don鈥檛 want your phone to act like the HAL 9000 computer in the movie 2001 鈥� 鈥淗ey Dave, take a stress pill.鈥� But imagine that you鈥檙e someone who isn鈥檛 aware of stress until it suddenly becomes overwhelming 鈥� that鈥檚 true for lots of people 鈥� and that you have a wearable device that gently warns you well before the stress level gets high, giving you the chance to change what you鈥檙e doing. What I鈥檓 working on is raising a person鈥檚 awareness of subtle changes before they would naturally become aware of them, and certainly before a looming emotional storm hits that affects them and everybody around them.

鈥淚 want people to be aware of a looming emotional storm before it hits them鈥�

What will affective computing鈥檚 killer app be?

People often think of a killer app as the thing that everybody sees and names and recognises. But if we succeed in putting the emotional intelligence in our different smart interactive systems, you probably won鈥檛 notice it explicitly. You鈥檒l just feel that you had an intelligent interaction that made you feel good, and understood. That鈥檚 hard to pull off, even for another human.

A simple example would be your GPS system. You don鈥檛 want a happy, upbeat voice telling you to make a U-turn when you鈥檝e driven the wrong way and are feeling frustrated. In fact, in controlled experiments, we see that accidents happen more often when the driver is stressed and the GPS voice sounds happy. If these systems perceive our state, and respond appropriately, it makes the entire experience better 鈥� whether it鈥檚 driving or any other augmented system.

You have said that face-reading systems could help politicians better understand the feelings of their constituents. How would this work?

Many years ago at Affectiva, we started using face-reading technologies while people watched politicians speak. We hit a problem: all the facial expression analytics imposed symmetry on the tracking, but we realised people were smiling lopsidedly, smirking in scepticism at what the speakers were saying. We had to develop a smirk detector. I鈥檝e since left Affectiva, but last year they measured smiles, smirks, looks of disbelief and other expressions in people watching the US presidential debates online, and got some really nuanced feedback about what people liked and didn鈥檛 like. Giving people the power to communicate their emotion in an objective way to elected officials offers an opportunity to influence them in a powerful way.

How might that work in practice?

Imagine people watching a politician giving a talk on their smartphones, while the system watches their facial expressions and then aggregates relevant smiles, smirks and frowns across all the viewers. The team surrounding that politician might tell her, 鈥淕reat job, your message really connected!鈥�, but then facial-expression data show them that in fact 60 per cent of viewers exhibited negative expressions during that message. Such direct-to-leader data could be very powerful in helping voters communicate with their representatives effortlessly and accurately.

Social media thrives on content that angers or delights users. What would happen if these sites had unfettered access to facial reactions?

I don鈥檛 see everyone turning on their cameras all the time, giving 鈥渦nfettered access鈥�. But if social media sites did respond to your facial expressions, they could become much better recommenders for you by noting not just what you watch 鈥� which already happens 鈥� but also what you smile and laugh at, and helping you find more of that. Some people could get addicted to just watching things that make them smile.

On the other hand, you might choose settings that offer you a more diverse emotional diet. Good film-makers know that the best experiences involve a range of emotions 鈥� you have better highs when there are also lows.

You鈥檝e said that part of affective technology is actually about developing the next generation of humans. Can you elaborate on that?

Affective technologies can augment our innate ability to understand feelings 鈥� our own and those of others. People are manipulated emotionally all the time, by fake news, by friends, advertisers, teachers鈥� While some of this is bad, some of it is also good, and technology can help us measure, understand and manage how emotions influence our choices. We can create a better future by finding engaging ways to communicate with one another 鈥� and learning what people really want.

This article appeared in print under the headline 鈥淗ow am I feeling, machine?鈥�