Siri with sensitivity: Emotionally intelligent chatbot could ease elder loneliness

Is artificial intelligence the answer to loneliness in old age? If you ask Ana, a plush bear glued to a tablet, she'll say yes. 

And then the University of Alberta chatbot might ask if you're feeling OK and maybe suggest a fun activity — like looking at photos of your grandchildren or reading a story — to cheer you up.

Ana, standing for Automated Nanny Agent, is a chatbot being developed by the university that is breaking new ground in the area of emotionally intuitive AI and could one day be a companion for grandma.

Most chatbots — conversational agents like Siri and Alexa — are task-oriented, with the ability to book airline tickets or make restaurant reservations or tell you the weather, according to Osmar Zaiane, U of A professor and researcher as well as scientific director of the Alberta Machine Intelligence Institute.

"But our chatbot is not task-oriented," he said. "The goal is actually companionship for the elderly that live alone."

Empathy, love, anger and fear

But a truly good companion is one that can pick up on verbal cues and react accordingly — to be enthusiastic when you are happy, comforting when you're down, sympathetic when you're frustrated or angry, Zaiane said. 

"If somebody says 'I fell,' we're not going to talk about how beautiful the weather is outside. We have to show empathy."

To that end, Zaiane's team has used machine learning to give its prototype chatbot a sensitive side, teaching it to detect emotionally loaded words — chatbots "hear" through voice-to-text technology — and to respond in kind, he said.

"We took thousands and thousands and thousand of tweets, for example, that the author has hashtagged with #happy or ... with other hashtags that express other emotions," Zaiane said.

• Oh A.L.Ex., you crack me up: Tech and theatre collide at Edmonton improv show
• Ethics and artificial intelligence: Why tech has to have a moral backbone

As the chatbot learned to discriminate between emotions, it was also taught how to generate appropriate responses using dialogue it learned from movies or television, he said.

Zaiane said his teddy bear prototype is able to generate responses but admits they're a bit … odd.

"It learns from movies so whenever we generate automatically a sentence after an utterance, it sounds movie-like," he said. "We're trying to now teach it sentences from other sources, whether it's Reddit or even Twitter." 

Zaiane's team recently published research that shows promising results in the model's ability to provide responses that, in most cases, matched emotions that were requested. Some — like surprise and love — were easier to express than others.

The next phase of research will involve the program independently deciding what emotion to express, he said. 

More than a companion

The idea of building a chatbot as a companion for elderly people was inspired by Zaiane's mother, who lives alone in her own home. Like many seniors with shrinking social opportunities, she gets lonely and bored — factors that can lead to depression and worsening health.

In the U of A lab, the tablet was glued to a teddy bear in order to make the prototype more personal. That's a theme that Zaiane is carrying into its capabilities.

Zaiane's envisions a device with the power to make suggestions and perform some tasks — suggest a nap to someone who is tired, look up a recipe for someone who wants to bake cookies or read a story to someone who is just bored, he said.

Future versions could be integrated with smart home technology to become a caretaker's assistant, tracking sleep, medication or moods and being able to provide that information to caregivers or family members, he said.

The final look and sound of the companion would be decided by the company that eventually commercializes the technology.

Zaiane noted that a digital companion isn't better than human companionship — but it is better than nothing at all. 

"We all have family members," he said. "How often do we go and visit them? How often do we call them? We have to be realistic."