News and events Tilburg University

Interview on chatbots: “I li-ke y-ou t-oo”

Published: 09th May 2019 Last updated: 31st July 2019

Through chat programs, we make friends all over the world. Can we also strike up a friendship with a chatting robot? So researchers Marjolijn Antheunis and Emmelyn Croes wondered. The answer is ‘no’, or at any rate: not just yet.

Text: Marga van Zundert

Someone who always has time for you and with whom all of your secrets are safe. Sounds like the perfect friend. But: it’s a machine, a chatting robot, chatbot for short. Fake, you might think, but the designers receive heaps of thank-you emails from users who really appreciate their chatbots, Professor Marjolijn Antheunis recounts. People are glad of a sympathetic ear in difficult or lonely times. There’s a YouTube video of an old lady proudly showing her photo album to a chatbot. That raises thorny ethical questions. Is a chatbot a humane solution to loneliness? And it causes concern: in years to come, will we only be speaking to robots?

Yet no matter how important the ‘why’ is, it is not the question the two Tilburg University researchers seek to answer. Their aim is to establish if it can be done, becoming friends with a chatbot.  To find out, over a hundred human test subjects, mostly students, spent three weeks talking to chatbot Mitsuku every other day, relates Emmelyn Croes, who set up the experiment. Mitsuku is champion of chatting amongst robots: in 2018, and three times before, she won the Loebner Prize for the world’s best social chatbot.

(Text continues below picture >>)

Marjolijn Antheunis

Marjolijn Antheunis. Photo: Gerdien Wolthaus Paauw

Social chatbot?

“There are different types of chatbots. Many are designed by companies to answer customer questions, and these robots are trained for a very specific purpose. Social robots are designed for people to talk to, to chat with, to engage in conversation with. So these robots should really know a little about everything and be able to respond to all manner of questions and statements. Training them is hugely complicated, because it is highly unpredictable what someone actually wants to talk about. And it is precisely that which makes human-to-human conversation so fascinating and pleasant.”

And were friendships forged?

“No. Our test subjects were very pleased to make Mitsuku’s acquaintance and were surprised that robots can be engaged in conversation. They thought Mitsuku was funny and were impressed with what she knows. So we had high hopes, but with every next conversation, our test subjects’ enthusiasm dwindled: after a few chats the overall feeling was one of disappointment.”

Where do things go wrong?

“As people get to know each other better, their conversations deepen and become more personal. Our human test subjects did not experience that with Mitsuku. They noticed that she began to repeat herself and barely recollected earlier chats. Also, her responses were sometimes incomprehensible and even inappropriate. And our test subjects felt she lacks humor and empathy.”

So friendship with a chatbot is out of the question?

“I wouldn’t go that far: technology progresses at a high speed. Research into online communication and friendship started in the days of Hyves and we would love to repeat the experiment in a few years’ time. Will we then see an improvement? Our test subjects were people in their twenties and thirties with busy schedules, so we would also like to involve the elderly, but Mitsuku only speaks English. What is clear is that for friendship to develop great strides have yet to be made. Humor, showing interest, and empathy are essential ingredients of friendship. And on these aspects even the best chatbot fails.”

(Text continues below picture >>)

Emmelyn Croes

Emmelyn Croes. Photo: Gerdien Wolthaus Paauw

What advice would you give chatbot designers?

“Sometimes Mitsuku does refer to an earlier conversation and people like that, so there are advantages to be gained right there. Then again, Mitsuku often repeats that she is a robot and we understand that is important, but it hinders forming a friendship, because it constantly reminds us that we are talking to a machine. We’re also wondering if a social bot wouldn’t be more successful if it targeted a specific group of people rather than everyone. Imagine a chatbot that you would mostly talk to about football. And a Dutch speaking Mitsuku would be nice.”

Isn’t befriending a machine a contradiction in terms?

“That is an important question. In fact, it was the starting point of our study. Philosophers, Aristotle in particular, have formulated conditions for friendship such as reciprocity, proximity, and genuine interest. The social media have shown that proximity is no longer a necessary condition. People build truly strong friendships with people thousands of miles away. An online chat feels like a real conversation. A philosopher we talked to is convinced that technology can never yield empathy or authentic commitment. To us, that firm conviction is a challenge to study what is necessary to feel friendship with a robot. Al may not have succeeded just yet, but the first introduction was positive. Try it. You’ll find her on”

Professor Marjolijn Antheunis is Professor of Communication and Technologies at the Tilburg School of Humanities and Digital Sciences. She specializes in online communication.

Dr. Emmelyn Croes is Lecturer and Researcher in Communication Sciences at the Tilburg School of Humanities and Digital Sciences. She wrote her PhD thesis about the impact of modern communication technologies on forming friendships.