Seite - 56 - in Coaching im digitalen Wandel

Strong / Terblanche56 Paradoxically, some humans find it easier to confide in an autonomous sys- tem without human characteristics – such as gathering witness statements and therapy bots. If humans already tend to assess and manage their level of open- ness with human coaches, is that very different to maintaining an awareness of a chatbot? The point is that chatbots, in any form, will need to give us the right level of visual clues to build our trust. Having a coaching conversation Let us get back to creating a coaching chatbot. The primary mode of communi- cation of a chatbot is conversation (text or voice-based). There are three dimen- sions of a conversation that must be considered (Britz, 2016): Ȥ closed vs open Ȥ short vs long Ȥ retrieval-based vs generative-based. Open conversation refers to a large array of topics and contexts with the abil- ity to switch context rapidly, e. g. from talking about the weather to switching to a discussion on climate change, whereas closed conversations are limited to a specific domain with limited complexity, e. g. ordering a pizza or asking for help on selecting a mortgage bond. Conversation duration refers to the amount of time a chatbot is able to hold a meaningful conversation. A chatbot that pro- vides very specific information probably needs a conversation duration of one minute or two, whereas a more open-ended conversation could last several min- utes or longer. Lastly, rule-based chatbots that use pre-defined scripts (retrieval- based) literally retrieve the next sentence from a finite list of options. A gen- erative-based chatbot uses machine learning to generate conversations on the fly, based on what it deems appropriate for the context. At this point, we move closer to the concept of virtual humans and sentient beings. One could, therefore, classify chatbots into two opposite camps: simple chat- bots (closed, short, retrieval-based) and complex chatbots (open, long, gener- ative-based). A quick look at some of the existing coaching chatbots such as Wysa, Youper, Woebot and Vicci reveal that they fall into the simple category of chatbots. Responses and questions are often repeated (retrieval-based) and the scope of conversation is limited (closed, short). At the complex end of the spectrum, the humanoid robot Sophia and the interviewing chatbot Intervyo are more self-aware, self-motivated and have an inner narrative (Burden & Savin- Baden, 2019). Coaching chatbots are not advanced enough as yet. Open-Access-Publikation im Sinne der CC-Lizenz BY-NC-ND 4.0