 My name is Christine Liebevecht and I received my PhD at Helberg University, Universite Nijmegen on the Department of Communication and Information Sciences and I particularly like to connect the field of Communication and Information Sciences with the field of artificial intelligence and this project that I'm going to present today is on Chattel's communication in the customer service industry. So conversational agents are artificial intelligence computer programs, oftentimes using natural language to engage the users and these agents are increasingly being deployed by organizations in customer service settings. So we see an increase in the usage of these technologies. So for example, according to the Gardner Technologies in Service Bullseye, 68 percent of these service leaders expect conversational agents will become more important in the next years. So there's a real grow in that area. So here you see such a chatbot and then you can see, well, there are certain tone of voice used in the communication and a customer can, for example, try to chase a delivery address. And that is why a topic that these kind of chatbots are really useful for these more or less frequently asked questions. So also with regard to trends watching, maybe you are familiar with the Gardner Hype Cycle for artificial intelligence. And as you can see here, chatbots are on the peak of inflated expectations, meaning that I will actually maybe the expectations on the abilities of chatbots or customer service chatbots is really a bit too high. So really complex questions, emotional questions are really hard for a customer service chatbots yet. But as I said before, frequently asked questions are kind of doable for this technology. So that is also, I think now chatbots is more in the form of disillusionment, meaning that now you can actually see for what kind of cases this technology is really useful for. And then it can actually grow in that specific area. That is also what we see in the field. So organizations train chatbots oftentimes on really specific topics, for example, with regard to delivery services, opening hours, all these easy questions that you don't want to bother your human employees with. So actually there are a few opportunities in using chatbots in this field. They are always available 24-7. They respond really fast. They can manage many clients simultaneously in contrast to human employees. And they can also answer specific questions about these specific topics. However, performance is something that can be improved oftentimes. So here you see for someone, someone is trying this IKEA chatbot, Anna. It's an old video already, but you can see that the intent recognition isn't that good yet. So that is something to improve and oftentimes also where researchers and also organizations spend many hours on in trying to improve the intent recognition, for example. But you also see here that the chatbot has a name. In this case, IKEA has chosen Anna. And you also see her facial expressions while you are communicating with her. Another example is this one. Maybe you are familiar with Microsoft Stay. Actually, the company tried to or the aim was to create a chatbot, so to say, or at least a Twitter account that automatically can respond to other people on Twitter. And the aim was to have a really personal interaction so that the tone of voice of Stay matches the tone of voice of the people on Twitter. That was the aim. And they launched Stay some years ago, 2015. But this is actually what happened on Twitter. And it was because they use machine learning approach. And you can see that they maybe did not learn the right sentences that that was Microsoft school. So you can see that machine learning approach, for example, can be, well, dangerous. If the input in the machine, of course, is not of high quality or not the type of quality that you aim. So, of course, you can imagine that Microsoft turned this one offline as soon as possible once they saw that this wasn't successful at all. But the aim to create more personalized conversations with people online, that is something that is desirable. So here, for example, you see that one of the reasons why people would stop using the chatbot is because they say the chatbot wasn't able to chat in a friendly manner. And also with regards to personalization, after feeling to being treated in a human-like way is something that is one of the barriers. So therefore, we see also some challenges for chatbot development, because chatbots could misunderstand the customer's language. Oftentimes, they cannot fix mistakes. They can activate feelings of resistance, and also they do not communicate in a human-like way yet. In our project, we particularly try to focus on the latter one, namely how to create more human-like chatbots. And today, I will tell you more about the tone of voice of chatbots. That actually connects to prior theories. So the idea is that people apply social rules to computers. That's what computer or social actors are about, and they actually treat them in a similar way as they do to other humans. And we thought, well, maybe if we give more social cues to chatbots, it can be possible to actually enhance that feeling that someone is talking to a human rather than a machine, because that is one of the disadvantages of chatbots at the current moment. So we thought, to what extent can we enhance the perceived social presence? So the feeling that you are talking to another person in the interaction, and we try to do that by adopting linguistic elements in chatbot conversations. We were not the first one who tried to do this. So Araujo already conducted a study in which two different versions of the chatbots were developed, one more human-like chatbots with a name, Anna, using a formal language, and also human-like greeting and closing compared to the machine-like chatbots. And he indeed did find some positive effects with regards to the human-like chatbot type. So a higher perceived social presence and also a higher emotional connection. However, there were also some similarities in his paper, but it was actually one of the reasons why we thought, well, maybe we can use our knowledge from communication information sciences, namely how do humans connect or chat with each other? So what tone of voice do they use when a company and a human employee is communicating with another human being, with the customer? And we tried to adopt the so-called conversational human voice from that field to the field of chatbots communication. And the conversational human voice is a personal and engaging communication style. And characteristics are cues of personalization, informal language, and also invitational rhetoric. And in business to customer context, we do see positive effects of this specific tone of voice. So for example, you see that the left is more personalized than the right one, because of the picture, but also because of the signature of Anna, for example. Here you see some elements of informal language. So for example, abbreviations, but also the usage of smileys. And here you see examples of invitational rhetoric. So for example, showing sympathy and empathy, asking questions to the customer, apologizing to create that environment that the customer is really welcome to ask questions to the company, and they are really welcome to help the customer out. So this is the example I showed you before, and that was actually one of the experimental materials that we used. And as you can see, there are some informal language cues in it. So for example, a smiley multiple, etc. And we compared such a version with a more human-like version in which internal voice was more formal. We conducted several studies, actually. So the first study we used several CHV cues, and we indeed found that the human-like chatbot enhanced the perceived social presence, which in turn enhanced the grant attitude. So positive effects of the human-like chatbot. We found similar effects when we only adopted informal language cues, so only focusing on that part. And in the last study, we also found it for informal language cues, but we also conducted a study in which we only focused on invitational rhetoric. And there we actually saw that invitational rhetoric effects actually depend on the familiarity of the customer with the grant. So if someone is familiar with the grant, so if you are a customer of IKEA, for example, then you did like the invitational rhetoric cues. But if you were not familiar with the brand at all, then it was considered as inappropriate. And we could explain that by means of the role theory, namely that it's kind of inappropriate to use such a language style if you are not familiar with each other yet. So then a more distant tone of voice, so to say, is more appropriate in that situation. These were, well, examples of tone of voice of invitational rhetoric that we use. So using humor, showing sympathy, empathy, asking questions, thanking, apologizing, et cetera. So these were the conclusions of this line of research. You were like, tone of voice generally has a positive impact on brand attitudes via social presence. But be careful when using invitational rhetoric, because then brand familiarity comes into play. And also be aware of the uncanny valley. We haven't tested it yet, but the idea of the uncanny valley is that maybe you can also be too human-like with the chatbots. It hasn't shown in the field of robots. And maybe it could also count for chatbots as well, chatbots communication. So if people have the feeling that it is too human-like, then they can also feel some kind of resistance. And that is, of course, not desirable in our context. So then I want to briefly also dive into new challenges that we aim to address in our project. So first of all, for example, how the chatbot is introduced by customers. So actually managing expectations. Sometimes you do see in practice that the chatbot says, hi, I'm your virtual assistant. You can ask me anything you want. And I'm really intelligent because I was trained really well. But oftentimes you also see another version, namely, hi, I'm just a chatbot in training. I don't know much yet, but you can ask me questions about a specific topic. And you can imagine that actually manages the expectations of the ability of the chatbots to the customer, which can be very helpful. Furthermore, we also want to address or focus on how to communicate in case of errors. So if the chatbot doesn't perform well yet on that topic, how do you communicate that in order to try to avoid negative feelings from the customer? And lastly, we're also interested in, well, if the chatbot is not able to solve the problem, how can you redirect the customer to a human employee? And especially in the customer service context, we do see oftentimes that handover. But you want to make that as smooth as possible for the user and also for the human employee. So these are challenges that we also aim to address in our study or in our project as well, besides next to other topics, of course. If you want to know more, please check out our websites. And these are the people who are involved in the project as well. So thank you very much for your attention.